Difference between revisions of "Timeline of ChatGPT"

From Timelines
Jump to: navigation, search
(6 intermediate revisions by the same user not shown)
Line 39: Line 39:
 
| 2019 || May 7 || || Adam King on {{w|Twitter}} launches “TalktoTransformer.com”, a site where people can have an interface to play with OpenAI's new text-generating language model.<ref>{{cite web |title=https://twitter.com/adamdanielking/status/1125831730848571392?lang=en |url=https://twitter.com/adamdanielking/status/1125831730848571392?lang=en |website=Twitter |access-date=23 March 2023 |language=en}}</ref>
 
| 2019 || May 7 || || Adam King on {{w|Twitter}} launches “TalktoTransformer.com”, a site where people can have an interface to play with OpenAI's new text-generating language model.<ref>{{cite web |title=https://twitter.com/adamdanielking/status/1125831730848571392?lang=en |url=https://twitter.com/adamdanielking/status/1125831730848571392?lang=en |website=Twitter |access-date=23 March 2023 |language=en}}</ref>
 
|-
 
|-
| 2019 || May 9 || || French company {{w|Hugging Face}} releases a conversational AI demo based on GPT-2 models. The company believes that open-source and knowledge sharing should be the default, and that the progress of deep learning for NLP is at risk without open-source. While they are firm believers in open-source, they also understand that technology is not neutral and specific action must be taken to ensure a positive impact. The company analyzed the potential malicious use of the technology, and concluded that it could be used for drastic improvement of spam-bots or mass catfishing and identity fraud. To mitigate these risks, Hugging Face provides users with some simple tricks to recognize the conversational AI from a human conversation.<ref>{{cite web |last1=Delangue |first1=Clément |title=Ethical analysis of the open sourcing of a state-of-the-art conversational AI |url=https://medium.com/huggingface/ethical-analysis-of-the-open-sourcing-of-a-state-of-the-art-conversational-ai-852113c324b2 |website=HuggingFace |access-date=23 March 2023 |language=en |date=9 May 2019}}</ref>
+
| 2019 || May 9 || || French company {{w|Hugging Face}} releases a conversational AI demo based on GPT-2 models. The company believes that [[W:Open-source software|open-source]] and knowledge sharing should be the default, and that the progress of deep learning for NLP is at risk without open-source. While they are firm believers in open-source, they also understand that technology is not neutral and specific action must be taken to ensure a positive impact. The company analyzed the potential malicious use of the technology, and concluded that it could be used for drastic improvement of {{w|spambot}}s or mass {{w|catfishing}} and [[w:Identity theft|identity fraud]]. To mitigate these risks, Hugging Face provides users with some simple tricks to recognize the conversational AI from a human conversation.<ref>{{cite web |last1=Delangue |first1=Clément |title=Ethical analysis of the open sourcing of a state-of-the-art conversational AI |url=https://medium.com/huggingface/ethical-analysis-of-the-open-sourcing-of-a-state-of-the-art-conversational-ai-852113c324b2 |website=HuggingFace |access-date=23 March 2023 |language=en |date=9 May 2019}}</ref>
 
|-
 
|-
 
| 2019 || August 20 || || OpenAI announces the release of its 774 million parameter {{w|GPT-2}} language model, along with an open-source legal agreement to make it easier for organizations to initiate model-sharing partnerships with each other. They also publish a technical report about their experience in coordinating with the wider AI research community on publication norms. Through their research, they found that coordination for language models is difficult but possible, synthetic text generated by language models can be convincing to humans, and detecting malicious use of language models is a genuinely difficult research problem that requires both statistical detection and human judgment.<ref name="GPT-2: 6-month follow-up">{{cite web |title=GPT-2: 6-month follow-up |url=https://openai.com/research/gpt-2-6-month-follow-up |website=openai.com |access-date=23 March 2023}}</ref>
 
| 2019 || August 20 || || OpenAI announces the release of its 774 million parameter {{w|GPT-2}} language model, along with an open-source legal agreement to make it easier for organizations to initiate model-sharing partnerships with each other. They also publish a technical report about their experience in coordinating with the wider AI research community on publication norms. Through their research, they found that coordination for language models is difficult but possible, synthetic text generated by language models can be convincing to humans, and detecting malicious use of language models is a genuinely difficult research problem that requires both statistical detection and human judgment.<ref name="GPT-2: 6-month follow-up">{{cite web |title=GPT-2: 6-month follow-up |url=https://openai.com/research/gpt-2-6-month-follow-up |website=openai.com |access-date=23 March 2023}}</ref>
Line 115: Line 115:
 
| 2023 || February 27 || Study || A paper provides an overview of the technology behind ChatGPT. It discusses the historical background of chatbots and the potential applications of ChatGPT in various domains such as healthcare, education, and research. However, the paper also highlights the privacy and ethical concerns surrounding ChatGPT and the limitations of the current version. The paper also includes an interview with ChatGPT and presents its responses to various questions.<ref>{{cite journal |last1=Shahriar |first1=Sakib |last2=Hayawi |first2=Kadhim |title=Let's have a chat! A Conversation with ChatGPT: Technology, Applications, and Limitations |date=2023 |doi=10.48550/arXiv.2302.13817 |url=https://arxiv.org/abs/2302.13817}}</ref>
 
| 2023 || February 27 || Study || A paper provides an overview of the technology behind ChatGPT. It discusses the historical background of chatbots and the potential applications of ChatGPT in various domains such as healthcare, education, and research. However, the paper also highlights the privacy and ethical concerns surrounding ChatGPT and the limitations of the current version. The paper also includes an interview with ChatGPT and presents its responses to various questions.<ref>{{cite journal |last1=Shahriar |first1=Sakib |last2=Hayawi |first2=Kadhim |title=Let's have a chat! A Conversation with ChatGPT: Technology, Applications, and Limitations |date=2023 |doi=10.48550/arXiv.2302.13817 |url=https://arxiv.org/abs/2302.13817}}</ref>
 
|-
 
|-
| 2023 || February 28 || Study || A paper presents an empirical study that investigates the performance of ChatGPT in cross-lingual summarization (CLS), which involves generating a summary in a different target language from a source language document. The authors use various prompts to guide ChatGPT to perform zero-shot CLS from different paradigms and provide a preliminary evaluation of its generated summaries. The study finds that ChatGPT initially prefers to produce lengthy summaries with more detailed information, but with the help of interactive prompts, it can significantly improve its CLS performance by balancing informativeness and conciseness. Experimental results on three widely-used CLS datasets show that ChatGPT outperforms the advanced GPT 3.5 model. The paper provides qualitative case studies to demonstrate the superiority of ChatGPT on CLS.<ref>{{cite journal |last1=Wang |first1=Jiaan |last2=Liang |first2=Yunlong |last3=Meng |first3=Fandong |last4=Li |first4=Zhixu |last5=Qu |first5=Jianfeng |last6=Zhou |first6=Jie |title=Cross-Lingual Summarization via ChatGPT |journal=arXiv:2302.14229 [cs] |date=27 February 2023 |url=https://arxiv.org/abs/2302.14229 |access-date=6 March 2023}}</ref>
+
| 2023 || February 28 || Study || A paper presents an empirical study that investigates the performance of ChatGPT in {{w|cross-lingual summarization}} (CLS), which involves generating a summary in a different target language from a source language document. The authors use various prompts to guide ChatGPT to perform zero-shot CLS from different paradigms and provide a preliminary evaluation of its generated summaries. The study finds that ChatGPT initially prefers to produce lengthy summaries with more detailed information, but with the help of interactive prompts, it can significantly improve its CLS performance by balancing informativeness and conciseness. Experimental results on three widely-used CLS datasets show that ChatGPT outperforms the advanced GPT 3.5 model. The paper provides qualitative case studies to demonstrate the superiority of ChatGPT on CLS.<ref>{{cite journal |last1=Wang |first1=Jiaan |last2=Liang |first2=Yunlong |last3=Meng |first3=Fandong |last4=Li |first4=Zhixu |last5=Qu |first5=Jianfeng |last6=Zhou |first6=Jie |title=Cross-Lingual Summarization via ChatGPT |journal=arXiv:2302.14229 [cs] |date=27 February 2023 |url=https://arxiv.org/abs/2302.14229 |access-date=6 March 2023}}</ref>
 
|-
 
|-
 
| 2023 || February || || A Colombian judge reportedly uses ChatGPT to help rule on a case about the medical rights of a child with autism. The case is about whether the child's insurance should cover the bills for his medical treatments. The judge uses ChatGPT in accordance with a [[w:Judiciary of Colombia|law in Colombia]] that allows the use of virtual tools in some cases. Opinions are divided, with some professionals opposing the use of AI in a legal setting, while others see it as a useful tool.<ref name="Bhaimiya">{{cite web |last1=Bhaimiya |first1=Sawdah |title=A Colombian judge used ChatGPT to help rule on a case about the medical rights of a child with autism, reports say |url=https://www.businessinsider.com/chatgpt-colombian-judge-used-in-ruling-child-medical-rights-report-2023-2 |website=Business Insider |access-date=4 March 2023}}</ref>  
 
| 2023 || February || || A Colombian judge reportedly uses ChatGPT to help rule on a case about the medical rights of a child with autism. The case is about whether the child's insurance should cover the bills for his medical treatments. The judge uses ChatGPT in accordance with a [[w:Judiciary of Colombia|law in Colombia]] that allows the use of virtual tools in some cases. Opinions are divided, with some professionals opposing the use of AI in a legal setting, while others see it as a useful tool.<ref name="Bhaimiya">{{cite web |last1=Bhaimiya |first1=Sawdah |title=A Colombian judge used ChatGPT to help rule on a case about the medical rights of a child with autism, reports say |url=https://www.businessinsider.com/chatgpt-colombian-judge-used-in-ruling-child-medical-rights-report-2023-2 |website=Business Insider |access-date=4 March 2023}}</ref>  
 
|-
 
|-
| 2023 || February || Adoption || Small businesses in the {{w|United Arab Emirates}} turn to ChatGPT to reduce the amount of time spent on mundane tasks such as writing job descriptions and blog posts, thereby allowing them to concentrate on more creative aspects of their business. By this time, ChatGPT is used by large companies including Meta and Shopify. The Wall Street Journal speculates that the tool could eventually lead to job losses in some areas, although ChatGPT's flaws, including providing inaccurate responses, and the need for further regulation of AI technology so far prevent this for the time being.<ref>{{cite web |title=UAE small businesses embrace ChatGPT as tech allows productivity boost |url=https://english.alarabiya.net/News/gulf/2023/02/15/UAE-small-businesses-embrace-ChatGPT-as-tech-allows-productivity-boost |website=Al Arabiya English |access-date=4 March 2023 |language=en |date=15 February 2023}}</ref>
+
| 2023 || February || Adoption || Small businesses in the {{w|United Arab Emirates}} turn to ChatGPT to reduce the amount of time spent on mundane tasks such as writing job descriptions and blog posts, thereby allowing them to concentrate on more creative aspects of their business. By this time, ChatGPT is used by large companies including [[W:Meta Platforms|Meta]] and {{w|Shopify}}. The ''{{w|Wall Street Journal}}'' speculates that the tool could eventually lead to job losses in some areas, although ChatGPT's flaws, including providing inaccurate responses, and the need for further regulation of AI technology so far prevent this for the time being.<ref>{{cite web |title=UAE small businesses embrace ChatGPT as tech allows productivity boost |url=https://english.alarabiya.net/News/gulf/2023/02/15/UAE-small-businesses-embrace-ChatGPT-as-tech-allows-productivity-boost |website=Al Arabiya English |access-date=4 March 2023 |language=en |date=15 February 2023}}</ref>
 
|-
 
|-
 
| 2023 || February || || ChatGPT performs poorly when asked to answer questions from Singapore's Primary School Leaving Exam (PSLE), with an average score of 16 out of 100 marks for mathematics and 21 for science. It is found to make errors even in simple addition and can not understand or answer any questions that reference diagrams or graphs. However, ChatGPT manages to pass the English tests and scores an average of 11 out of 20 marks across the three papers it took. The bot's poor performance in exams raises concerns about the possibility of AI bots being used for cheating.<ref>{{cite web |last1=Bharade |first1=Aditi |title=ChatGPT failed miserably in Singapore's 6th-grade tests, averaging 16% for math and 21% for science. Days later, it was getting answers right. |url=https://www.businessinsider.com/chatgpt-failed-singapore-sixth-grade-exams-psle-2023-2 |website=Business Insider |access-date=4 March 2023}}</ref>
 
| 2023 || February || || ChatGPT performs poorly when asked to answer questions from Singapore's Primary School Leaving Exam (PSLE), with an average score of 16 out of 100 marks for mathematics and 21 for science. It is found to make errors even in simple addition and can not understand or answer any questions that reference diagrams or graphs. However, ChatGPT manages to pass the English tests and scores an average of 11 out of 20 marks across the three papers it took. The bot's poor performance in exams raises concerns about the possibility of AI bots being used for cheating.<ref>{{cite web |last1=Bharade |first1=Aditi |title=ChatGPT failed miserably in Singapore's 6th-grade tests, averaging 16% for math and 21% for science. Days later, it was getting answers right. |url=https://www.businessinsider.com/chatgpt-failed-singapore-sixth-grade-exams-psle-2023-2 |website=Business Insider |access-date=4 March 2023}}</ref>
Line 125: Line 125:
 
| 2023 || February 25 || || A paper introduces ChatAug, a text data augmentation approach based on ChatGPT, to mitigate the challenge of limited sample sizes in natural language processing tasks, particularly in few-shot learning scenarios. The proposed approach rephrases each sentence in the training samples into multiple conceptually similar but semantically different samples to ensure the faithfulness and completeness of the generated data. The experiment results demonstrate that ChatAug outperforms state-of-the-art text data augmentation methods in terms of testing accuracy and distribution of the augmented samples.<ref>{{cite journal |last1=Dai |first1=Haixing |last2=Liu |first2=Zhengliang |last3=Liao |first3=Wenxiong |last4=Huang |first4=Xiaoke |last5=Wu |first5=Zihao |last6=Zhao |first6=Lin |last7=Liu |first7=Wei |last8=Liu |first8=Ninghao |last9=Li |first9=Sheng |last10=Zhu |first10=Dajiang |last11=Cai |first11=Hongmin |last12=Li |first12=Quanzheng |last13=Shen |first13=Dinggang |last14=Liu |first14=Tianming |last15=Li |first15=Xiang |title=ChatAug: Leveraging ChatGPT for Text Data Augmentation |journal=arXiv:2302.13007 [cs] |date=27 February 2023 |doi=10.48550/arXiv.2302.13007 |url=https://arxiv.org/abs/2302.13007 |access-date=6 March 2023}}</ref>
 
| 2023 || February 25 || || A paper introduces ChatAug, a text data augmentation approach based on ChatGPT, to mitigate the challenge of limited sample sizes in natural language processing tasks, particularly in few-shot learning scenarios. The proposed approach rephrases each sentence in the training samples into multiple conceptually similar but semantically different samples to ensure the faithfulness and completeness of the generated data. The experiment results demonstrate that ChatAug outperforms state-of-the-art text data augmentation methods in terms of testing accuracy and distribution of the augmented samples.<ref>{{cite journal |last1=Dai |first1=Haixing |last2=Liu |first2=Zhengliang |last3=Liao |first3=Wenxiong |last4=Huang |first4=Xiaoke |last5=Wu |first5=Zihao |last6=Zhao |first6=Lin |last7=Liu |first7=Wei |last8=Liu |first8=Ninghao |last9=Li |first9=Sheng |last10=Zhu |first10=Dajiang |last11=Cai |first11=Hongmin |last12=Li |first12=Quanzheng |last13=Shen |first13=Dinggang |last14=Liu |first14=Tianming |last15=Li |first15=Xiang |title=ChatAug: Leveraging ChatGPT for Text Data Augmentation |journal=arXiv:2302.13007 [cs] |date=27 February 2023 |doi=10.48550/arXiv.2302.13007 |url=https://arxiv.org/abs/2302.13007 |access-date=6 March 2023}}</ref>
 
|-
 
|-
| 2023 || February 27 || || The {{w|International Baccalaureate}} says that students can use content created by ChatGPT in their essays, but they must make it clear when they are [[w:quotation|quoting]] its responses and treat it as they would any other source. ChatGPT has the ability to produce plausible responses to text prompts, including requests to write essays. The IB, which is taken by thousands of children every year in the UK at more than 120 schools, says that the rise of chatbot technology would mean that essay writing will feature less prominently in future qualifications.<ref>{{cite web |last1=Milmo |first1=Dan |last2=editor |first2=Dan Milmo Global technology |title=ChatGPT allowed in International Baccalaureate essays |url=https://www.theguardian.com/technology/2023/feb/27/chatgpt-allowed-international-baccalaureate-essays-chatbot |website=The Guardian |access-date=3 March 2023 |date=27 February 2023}}</ref>
+
| 2023 || February 27 || || The {{w|International Baccalaureate}} says that students can use content created by ChatGPT in their essays, but they must make it clear when they are [[w:quotation|quoting]] its responses and treat it as they would any other source. ChatGPT has the ability to produce plausible responses to text prompts, including requests to write essays. The IB, which is taken by thousands of children every year in the UK at more than 120 schools, says that the rise of {{w|chatbot}} technology would mean that essay writing will feature less prominently in future qualifications.<ref>{{cite web |last1=Milmo |first1=Dan |last2=editor |first2=Dan Milmo Global technology |title=ChatGPT allowed in International Baccalaureate essays |url=https://www.theguardian.com/technology/2023/feb/27/chatgpt-allowed-international-baccalaureate-essays-chatbot |website=The Guardian |access-date=3 March 2023 |date=27 February 2023}}</ref>
 
|-
 
|-
 
| 2023 || March 1 || || Emily M. Bender, a computational linguist at the University of Washington, warns about the dangers of humanizing language models like chatbots, stating that the machines are great at mimicry but poor at facts. Bender argues that the big question is how humans will handle themselves around these machines, as we have learned to make machines that can mindlessly generate text but have not learned how to stop imagining the mind behind it.<ref>{{cite web |last1=Weil |first1=Elizabeth |title=You Are Not a Parrot |url=https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html |website=Intelligencer |access-date=4 March 2023 |language=en-us |date=1 March 2023}}</ref>
 
| 2023 || March 1 || || Emily M. Bender, a computational linguist at the University of Washington, warns about the dangers of humanizing language models like chatbots, stating that the machines are great at mimicry but poor at facts. Bender argues that the big question is how humans will handle themselves around these machines, as we have learned to make machines that can mindlessly generate text but have not learned how to stop imagining the mind behind it.<ref>{{cite web |last1=Weil |first1=Elizabeth |title=You Are Not a Parrot |url=https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html |website=Intelligencer |access-date=4 March 2023 |language=en-us |date=1 March 2023}}</ref>
Line 139: Line 139:
 
| 2023 || March 2 || Study || A paper presents a methodology to assess the impact of AI language models like ChatGPT on different occupations, industries, and geographies. The authors find that occupations like telemarketers and post-secondary teachers in subjects such as English language, literature, foreign language, and history are the most exposed to language modeling. The industries most exposed to advances in language modeling are legal services and securities, commodities, and investments.<ref>{{cite journal |last1=Felten |first1=Ed |last2=Raj |first2=Manav |last3=Seamans |first3=Robert |title=How will Language Modelers like ChatGPT Affect Occupations and Industries? |date=2023 |doi=10.48550/arXiv.2303.01157}}</ref>
 
| 2023 || March 2 || Study || A paper presents a methodology to assess the impact of AI language models like ChatGPT on different occupations, industries, and geographies. The authors find that occupations like telemarketers and post-secondary teachers in subjects such as English language, literature, foreign language, and history are the most exposed to language modeling. The industries most exposed to advances in language modeling are legal services and securities, commodities, and investments.<ref>{{cite journal |last1=Felten |first1=Ed |last2=Raj |first2=Manav |last3=Seamans |first3=Robert |title=How will Language Modelers like ChatGPT Affect Occupations and Industries? |date=2023 |doi=10.48550/arXiv.2303.01157}}</ref>
 
|-
 
|-
| 2023 || March 3 || || Microsoft unveils a new AI model, Kosmos-1, which can respond to visual cues or images and is capable of tasks such as visual question answering and image captioning. The model is a Multimodal Large Language Model (MLLM), which allows the user to interact with it through text prompts, images, and videos. Kosmos-1 is seen as a key step toward artificial general intelligence, with researchers suggesting that multimodal perception is required to move beyond ChatGPT-like capabilities to AGI.<ref>{{cite web |last1=Nigam |first1=Saumya |last2=News |first2=India TV |title=Microsoft unveils Kosmos-1,a new AI model to race up with ChatGPT |url=https://www.indiatvnews.com/technology/news/microsoft-unveils-kosmos-1-a-new-ai-model-to-race-up-with-chatgpt-2023-03-03-851902 |website=www.indiatvnews.com |access-date=4 March 2023 |language=en |date=3 March 2023}}</ref>
+
| 2023 || March 3 || Competition || Microsoft unveils a new AI model, Kosmos-1, which can respond to visual cues or images and is capable of tasks such as visual question answering and image captioning. The model is a Multimodal Large Language Model (MLLM), which allows the user to interact with it through text prompts, images, and videos. Kosmos-1 is seen as a key step toward artificial general intelligence, with researchers suggesting that multimodal perception is required to move beyond ChatGPT-like capabilities to AGI.<ref>{{cite web |last1=Nigam |first1=Saumya |last2=News |first2=India TV |title=Microsoft unveils Kosmos-1,a new AI model to race up with ChatGPT |url=https://www.indiatvnews.com/technology/news/microsoft-unveils-kosmos-1-a-new-ai-model-to-race-up-with-chatgpt-2023-03-03-851902 |website=www.indiatvnews.com |access-date=4 March 2023 |language=en |date=3 March 2023}}</ref>
 
|-
 
|-
| 2023 || March 7 || || A paper discusses the feasibility of using ChatGPT, which is a challenging task due to the lack of annotated data. While ChatGPT has demonstrated impressive results in other tasks, it presents challenges for event extraction, which requires complex instructions defining all event types and their schemas. The authors conducted experiments to explore the feasibility of using ChatGPT for event extraction and found that it has only 51.04% of the performance of a task-specific model in complex scenarios. The model is not robust enough and is highly sensitive to different prompt styles, which can result in a poor user experience.<ref>{{cite journal |last1=Gao |first1=Jun |last2=Zhao |first2=Huan |last3=Yu |first3=Changlong |last4=Xu |first4=Ruifeng |title=Exploring the Feasibility of ChatGPT for Event Extraction |journal=arXiv.org |date=7 March 2023 |doi=10.48550/arXiv.2303.03836 |url=https://arxiv.org/abs/2303.03836 |language=en}}</ref>
+
| 2023 || March 7 || Review || A paper discusses the feasibility of using ChatGPT, which is a challenging task due to the lack of annotated data. While by this time ChatGPT demonstrated impressive results in other tasks, it presents challenges for event extraction, which requires complex instructions defining all event types and their schemas. The authors conducted experiments to explore the feasibility of using ChatGPT for event extraction and found that it has only 51.04% of the performance of a task-specific model in complex scenarios. The model is not robust enough and is highly sensitive to different prompt styles, which can result in a poor user experience.<ref>{{cite journal |last1=Gao |first1=Jun |last2=Zhao |first2=Huan |last3=Yu |first3=Changlong |last4=Xu |first4=Ruifeng |title=Exploring the Feasibility of ChatGPT for Event Extraction |journal=arXiv.org |date=7 March 2023 |doi=10.48550/arXiv.2303.03836 |url=https://arxiv.org/abs/2303.03836 |language=en}}</ref>
 
|-
 
|-
 
| 2023 || March 12 || Competition || Together releases OpenChatKit 0.15, the first open-source chatbot alternative to ChatGPT, under the Apache-2.0 license. The chat model has 20 billion parameters and was trained on 43 million instructions, and includes a generic chatbot and the necessary components to create specialized bots. The framework has four main parts, including a large language model tuned for chat, a set of customization recipes, a retrieval system, and a moderation model. The open-source foundation offered by OpenChatKit enables the development of domain-specific and general-purpose chatbots.<ref>{{cite web |last1=Shenwai |first1=Tanushree |title=Together Releases The First Open-Source ChatGPT Alternative Called OpenChatKit |url=https://www.marktechpost.com/2023/03/12/together-releases-the-first-open-source-chatgpt-alternative-called-openchatkit/ |website=MarkTechPost |access-date=14 March 2023 |date=12 March 2023}}</ref>  
 
| 2023 || March 12 || Competition || Together releases OpenChatKit 0.15, the first open-source chatbot alternative to ChatGPT, under the Apache-2.0 license. The chat model has 20 billion parameters and was trained on 43 million instructions, and includes a generic chatbot and the necessary components to create specialized bots. The framework has four main parts, including a large language model tuned for chat, a set of customization recipes, a retrieval system, and a moderation model. The open-source foundation offered by OpenChatKit enables the development of domain-specific and general-purpose chatbots.<ref>{{cite web |last1=Shenwai |first1=Tanushree |title=Together Releases The First Open-Source ChatGPT Alternative Called OpenChatKit |url=https://www.marktechpost.com/2023/03/12/together-releases-the-first-open-source-chatgpt-alternative-called-openchatkit/ |website=MarkTechPost |access-date=14 March 2023 |date=12 March 2023}}</ref>  
Line 154: Line 154:
 
|-
 
|-
 
| 2023 || March 23 || || OpenAI announces ChatGPT implementation support for [[w:Plug-in (computing)|plugins]], which are tools designed for language models to access up-to-date information, run computations, or use third-party services with safety as a core principle. The feature is at first rolled out gradually for a small set of users and is expected to be expanded as more is learned about safety and alignment challenges. Plugin developers can use the provided documentation to build plugins, and the first plugins were created by companies such as {{w|Expedia}}, [[w:Slack Technologies|Slack]], {{w|Wolfram Research}}, and {{w|Zapier}}.<ref>{{cite web |title=ChatGPT plugins |url=https://openai.com/blog/chatgpt-plugins |website=openai.com |access-date=24 March 2023}}</ref>
 
| 2023 || March 23 || || OpenAI announces ChatGPT implementation support for [[w:Plug-in (computing)|plugins]], which are tools designed for language models to access up-to-date information, run computations, or use third-party services with safety as a core principle. The feature is at first rolled out gradually for a small set of users and is expected to be expanded as more is learned about safety and alignment challenges. Plugin developers can use the provided documentation to build plugins, and the first plugins were created by companies such as {{w|Expedia}}, [[w:Slack Technologies|Slack]], {{w|Wolfram Research}}, and {{w|Zapier}}.<ref>{{cite web |title=ChatGPT plugins |url=https://openai.com/blog/chatgpt-plugins |website=openai.com |access-date=24 March 2023}}</ref>
 +
|-
 +
| 2023 || March 25 || || Russian-American podcaster and artificial intelligence researcher {{w|Lex Fridman}} publishes an interview with {{w|Sam Altman}}. They discuss {{w|GPT-4}}, {{w|political bias}}, {{w|AI safety}}, and [[w:Artificial neural network|neural network]] size. Other topics include AGI, fear, competition, and transitioning from non-profit to capped-profit. They also touch on power dynamics, political pressure, truth and misinformation, {{w|anthropomorphism}}, and future applications, among other topics.<ref>{{cite web |title=Sam Altman: OpenAI CEO on GPT-4, ChatGPT, and the Future of AI {{!}} Lex Fridman Podcast #367 |url=https://www.youtube.com/watch?v=L_Guz73e6fw |website=youtube.com |access-date=25 March 2023 |language=en}}</ref>
 
|-
 
|-
 
| 2023 || End of year || || According to OpenAI's estimates, ChatGPT is expected to generate US$200 million in revenue by the end of this year.<ref name="tooltester"/>
 
| 2023 || End of year || || According to OpenAI's estimates, ChatGPT is expected to generate US$200 million in revenue by the end of this year.<ref name="tooltester"/>

Revision as of 13:41, 25 March 2023

This is a timeline of ChatGPT, an AI-based chatbot developed by OpenAI. ChatGPT falls into the category of generative AI, a type of artificial intelligence that can be used to create text or images. It works by utilizing a large language model, which enables it to understand and produce dialogue using significant amounts of data.[1] Since its launch in November 2022, as of March 2023 ChatGPT has become the fastest-growing consumer app in internet history, reaching 100 million users within two months.[2] The tool contains 570 gigabytes of text data. Training the model took only 34 days, and the tool costs around $100,000 per day to run on Microsoft’s Azure Cloud. Despite being generated by an AI, an average of 53% of people can't tell that ChatGPT content is not human-written.[3]

Sample questions

The following are some interesting questions that can be answered by reading this timeline:

Big picture

Time period Development summary More details
December 2022 After being launched on November 30, by December ChatGPT already becomes widely used and dominates chat rooms as the next innovative AI language technology.[4]
January 2023 ChatGPT reaches 100 million monthly users, a faster rate of adoption than Instagram or TikTok.[5] The chatbot makes headlines for its ability to pass prestigious graduate-level exams.[6]
February 2023 ChatGPT receives an estimated 1 billion visitors, an average of over 35 million visitors per day.[7] ChatGPT plus is released.
March 2023 GPT-4 is released.

Full timeline

Year Month and date Event type Details
2015 December 11 Prelude OpenAI is introduced as a non-profit organization focused on advancing digital intelligence in a way that benefits humanity as a whole. Its goal is to create AI that is an extension of human will and distributed broadly and evenly. OpenAI's researchers are encouraged to publish their work and collaborate with others across many institutions. The organization has received funding from several donors, including Sam Altman, Elon Musk, and Amazon Web Services. Its leaders include Ilya Sutskever and Greg Brockman, who are experts in machine learning. OpenAI hopes to build a leading research institution that prioritizes a good outcome for all over its own self-interest.[8]
2016 April 27 Prelude OpenAI Gym is introduced as a toolkit for developing and comparing reinforcement learning (RL) algorithms that consist of a suite of environments from simulated robots to Atari games, and a site for comparing and reproducing results. The environments are written in Python and are compatible with algorithms written in any framework, such as Tensorflow and Theano. The platform aims to fix the lack of standardization of environments used in publications and the need for better benchmarks by providing a diverse suite of environments that range from easy to difficult and involve many different kinds of data. OpenAI Gym provides a site for comparing and reproducing results. They are starting out by maintaining a curated list of contributions that say something interesting about algorithmic capabilities.[9]
2016 December 5 Prelude OpenAI releases Universe as a software platform designed to measure and train general intelligence in AI systems using a variety of games, websites, and applications. Universe allows an AI to operate a computer like a human does, using screen pixels and a virtual keyboard and mouse, allowing it to be trained on any task that a human can perform with a computer. The platform is capable of launching any program as a Gym environment and upon its launch includes over a thousand environments, including games like slither.io and GTA V, and browser tasks. Universe aims to develop a single AI agent that can apply past experience to quickly master new, difficult environments, moving towards general intelligence. Universe is an open-source platform that can be used for research or integrated with other games or applications.[10]
2017 August 11 Prelude OpenAI announces having created a bot that learned to play Dota 2 through self-play and is capable of beating top professionals in 1v1 matches. The bot uses reinforcement learning and does not rely on imitation learning or tree search. The game is complex, with hidden information, requiring players to plan, attack, trick, and deceive their opponents. The AI's actions-per-minute are comparable to that of an average human player, but it has learned to predict opponents' movements, improvise in unfamiliar situations, and influence other players' units to help it succeed. OpenAI's next goal is to create a team of Dota 2 bots that can compete and collaborate with the top human teams.[11]
2018 June 11 GPT-1 is released.[7]
2019 February 14 OpenAI publishes a blog post discussing the release of GPT-2, a large-scale unsupervised language model with 1.5 billion parameters, which can generate coherent paragraphs of text, achieve state-of-the-art performance on many language modeling benchmarks, and perform rudimentary reading comprehension, machine translation, question answering, and summarization without task-specific training. Due to concerns about malicious applications, the trained model is not released, but a smaller model and a technical paper are released for research purposes. GPT-2 is trained to predict the next word in 40GB of internet text, using a diverse dataset, and can generate conditional synthetic text samples of unprecedented quality.[12]
2019 May 7 Adam King on Twitter launches “TalktoTransformer.com”, a site where people can have an interface to play with OpenAI's new text-generating language model.[13]
2019 May 9 French company Hugging Face releases a conversational AI demo based on GPT-2 models. The company believes that open-source and knowledge sharing should be the default, and that the progress of deep learning for NLP is at risk without open-source. While they are firm believers in open-source, they also understand that technology is not neutral and specific action must be taken to ensure a positive impact. The company analyzed the potential malicious use of the technology, and concluded that it could be used for drastic improvement of spambots or mass catfishing and identity fraud. To mitigate these risks, Hugging Face provides users with some simple tricks to recognize the conversational AI from a human conversation.[14]
2019 August 20 OpenAI announces the release of its 774 million parameter GPT-2 language model, along with an open-source legal agreement to make it easier for organizations to initiate model-sharing partnerships with each other. They also publish a technical report about their experience in coordinating with the wider AI research community on publication norms. Through their research, they found that coordination for language models is difficult but possible, synthetic text generated by language models can be convincing to humans, and detecting malicious use of language models is a genuinely difficult research problem that requires both statistical detection and human judgment.[15]
2019 September 17 Prelude OpenAI publishes an article describing a new simulation environment that allows agents to learn and improve their ability to play hide-and-seek, ultimately leading to the emergence of complex tool use strategies. In the simulation, agents can move, see, sense, grab, and lock objects in place. There are no explicit incentives for the agents to interact with objects other than the hide-and-seek objective. Agents are rewarded based on the outcome of the game. As agents train against each other in hide-and-seek, up to six distinct strategies emerge, leading to increasingly complex tool use. The self-supervised emergent complexity in this simple environment further suggests that multi-agent co-adaptation may one day produce extremely complex and intelligent behavior.[16]
2019 October 15 Prelude OpenAI reports on having trained a pair of neural networks that can solve the Rubik's Cube with a human-like robotic hand. The neural networks were trained entirely in simulation, using the same reinforcement learning code as OpenAI Five paired with a new technique called Automatic Domain Randomization (ADR). ADR creates diverse environments in simulation that can capture the physics of the real world, enabling the transfer of neural networks learned in simulation to be applied to the real world. The system can handle situations it never saw during training, such as being prodded by a stuffed giraffe. The breakthrough shows that reinforcement learning isn’t just a tool for virtual tasks, but can solve physical-world problems requiring unprecedented dexterity.[17]
2019 November 5 Prelude OpenAI releases the largest version of GPT-2, the 1.5B parameter version, along with code and model weights to aid detection of outputs of GPT-2 models. OpenAI releases the model as a test case for a full staged release process for future powerful models, hoping to continue the conversation with the AI community on responsible publication. OpenAI conducted some tests and research on the GPT-2 model and found that humans find GPT-2 outputs convincing, it can be fine-tuned for misuse, detection of synthetic text is challenging, they have not found evidence of misuse so far, and standards are needed for studying bias.[18]
2020 May 28 GPT-3 is released in Beta.[7]
2021 January 5 OpenAI introduces CLIP (Contrastive Language-Image Pre-training), a neural network that learns visual concepts from natural language supervision and can be applied to any visual classification benchmark. CLIP is trained on a variety of images with natural language supervision available on the internet and can be instructed in natural language to perform a variety of classification benchmarks without directly optimizing for the benchmark's performance. This approach improves the model's robustness and can match the performance of traditional models on benchmarks without using any labeled examples. CLIP's performance is more representative of how it will fare on datasets that measure accuracy in different settings. CLIP builds on previous work on zero-shot transfer, natural language supervision, and multimodal learning and uses a simple pre-training task to achieve competitive zero-shot performance on a variety of image classification datasets.[19]
2022 November 30 Initial launch GPT-3.5 is released.[7] ChatGPT is introduced as a conversational AI model that can interact with users and respond to follow-up questions, challenge incorrect premises, and reject inappropriate requests. It was trained using Reinforcement Learning from Human Feedback (RLHF) and fine-tuned from the GPT-3.5 model. While it has limitations such as occasionally providing incorrect or nonsensical answers and being sensitive to input phrasing, it is designed with safety mitigations and the ability to learn from user feedback. ChatGPT is available for free during the research preview on chat.openai.com.[20]
2022 December 5 Reception (positive) Kevin Roose of The New York Times labels ChatGPT as "the best artificial intelligence chatbot ever released to the general public". The New York Times also reports that it has been "rumored" that the next version of the AI, GPT-4, will be launched sometime in 2023.[21]
2022 December 5 Ban Popular coding question-and-answer site Stack Overflow temporarily bans the sharing of AI-generated answers produced by ChatGPT. The site's moderators say the ban is necessary as ChatGPT's answers are often incorrect and this can mislead users. ChatGPT ability to generate plausible but ultimately false responses is one of several well-known failings of AI text generation models.[22]
2022 December 27 Policy According to the BBC, as of date, ChatGPT is not allowed to "express political opinions or engage in political activism".[23]
2023 January 5 A paper examines the political ideology of ChatGPT, and finds that it has a pro-environmental, left-libertarian orientation based on its responses to 630 political statements from voting advice applications and the political compass test. ChatGPT would impose taxes on flights, restrict rent increases, and legalize abortion. The findings are consistent across multiple languages and different experiment variations. The study highlights the potential impact of politically biased conversational AI on society.[24]
2023 January 27 It is reported that Elon Musk plans to form a new AI research lab to develop an alternative to OpenAI's ChatGPT, after having approached AI researchers, including Igor Babuschkin. At this time, the project is still in its early stages, with no concrete plan to develop specific products.[25]
2023 February 1 OpenAI launches a pilot subscription plan called ChatGPT Plus, which offers faster response times, priority access to new features and improvements, and general access to ChatGPT even during peak times. The subscription costs $20/month and is available to customers worldwide. OpenAI still offers free access to ChatGPT but hopes to support its free access availability to as many people as possible through the subscription plan. OpenAI announces plans to refine and expand the offering based on user feedback and needs, and that it is actively exploring options for lower-cost plans, business plans, and data packs for more availability.[26]
2023 February 1 Study A study explores the opinions and perceptions of academicians, netizens, and information professionals towards ChatGPT-3, using social media comments and a survey. The research suggests that while ChatGPT-3 can be useful for research and writing, it is not entirely accurate and should be cross-checked. It also suggests that most academicians are beginning to accept ChatGPT-3, despite some resistance. The study is useful for academics, content developers, and librarians.[27]
2023 February 2 Growth It is reported that ChatGPT reached an estimated 100 million monthly active users in January, just two months after its launch, making it the fastest-growing consumer application in history, according to a UBS study. The growing usage, while imposing substantial computing cost on OpenAI, also provides valuable feedback to help train the chatbot's responses.[28]
2023 February 7 Competition Google presents its own AI chatbot called Bard, which is similar to ChatGPT. Google CEO Sundar Pichai says that Bard is more powerful and reliable than ChatGPT, as it draws on information from the web to provide fresh, high-quality responses. Bard is part of a series of tools based on generative artificial intelligence that Google launches to the market at this time.[29]
2023 February 8 Competition Chinese multinational technology company Alibaba Group announces that it is developing a rival to OpenAI's ChatGPT AI chatbot. The move comes as several other tech companies, including Google and Baidu, are also developing AI chatbot technology to rival ChatGPT.[1]
2023 February 8 Study A study explores the potential of ChatGPT, a popular AI chatbot, in generating academic essays that can evade plagiarism detection tools. The results suggest that ChatGPT can produce high-quality content on various topics with high originality. The study highlights the need for institutions to take appropriate measures to address potential plagiarism issues arising from the use of AI technology in education.[30]
2023 February 8 Study A paper proposes a framework for evaluating interactive language learning models (LLMs) such as ChatGPT using publicly available data sets. The authors evaluate ChatGPT's performance on 23 data sets covering eight different NLP tasks and find that ChatGPT outperforms other LLMs on most tasks, but has a lower accuracy in reasoning and suffers from hallucination problems. The authors also find that ChatGPT's interactive feature enables human collaboration to improve its performance. They release a codebase for evaluation set extraction.[31]
2023 February 9 Study A study compares students' essay writing performance with or without the use of ChatGPT-3 as a writing assistant tool. Eighteen students participated in the study, and essay scores, GPTs, writing time, authenticity, and content similarity were compared between the two groups. The study found that using ChatGPT-3 did not improve essay quality, as the control group outperformed the experimental group in most parameters. The average grade was C for both groups, and none of the predictors affected essay scores. The text unauthenticity was slightly higher in the experimental group, but the similarity among essays was generally low.[32]
2023 February 12 Study A letter introduces a new semantic communication scheme with ordered importance (SCOI) that utilizes ChatGPT as a consulting assistant. In SCOI, the transmitter queries ChatGPT to determine the importance order of each word in the message, and then uses an unequal error protection transmission strategy to ensure reliable transmission of essential words. This approach is compatible with existing source-channel separation designs and can be easily incorporated into current communication systems. Experimental results show that SCOI outperforms existing communication schemes in terms of transmission bit error rate and semantic loss measured by ChatGPT.[33]
2023 February 14 Study A paper evaluates the effectiveness of ChatGPT-generated hints compared to those created by human tutors in two algebra topics. The study found that 70% of hints produced by ChatGPT passed quality checks and both human and ChatGPT hints resulted in positive learning gains. However, the learning gains from human-created hints were significantly higher than ChatGPT hints in both topics. The authors discuss the limitations of the study and suggest future research directions. The problem and hint content used in the experiment is provided for replication purposes. The study suggests that while ChatGPT hints can be useful, human-created hints are still significantly more effective in promoting learning gains.[34]
2023 February 14 Indian fintech company Velocity launches India's first ChatGPT-powered chatbot named "Lexi." The AI development is integrated with its existing analytics tool, Velocity Insights, which offers e-commerce giants AI-powered business solutions. The integration enables users to leverage the same interface they use to drive business decisions. The Ministry of Electronics and IT also plans to integrate ChatGPT with WhatsApp to help Indian farmers understand and learn about several government schemes.[35]
2023 February 15 Study A study explores the use of ChatGPT in a human-centered design process. A hypothetical design project was conducted, where ChatGPT was utilized to generate personas, simulate interviews with fictional users, create new design ideas, simulate usage scenarios and conversations between an imaginary prototype and fictional users, and evaluate user experience. The results show that ChatGPT effectively performed the tasks assigned to it as a designer, user, or product, providing mostly appropriate responses, but with some drawbacks such as forgotten information, partial responses, and a lack of output diversity. The study discusses the potential benefits and limitations of using conversational LLMs in design and suggests directions for future research in this area.[36]
2023 February 16 Notable comment Google's senior vice president, Prabhakar Raghavan, issues a warning about ChatGPT, expressing concerns that the technology could provide "completely fictitious" answers to questions, which he refers to as "hallucination." He adds that the huge language models behind this technology make it impossible for humans to monitor every conceivable behavior of the system. Around this time, teachers in the United Kingdom also consider how to prevent the use of ChatGPT to cheat on school coursework.[37]
2023 February 16 Study A paper explores the performance of ChatGPT for query or aspect-based text summarization. The authors conducted an evaluation of ChatGPT's performance on four benchmark datasets and found that its performance is comparable to traditional fine-tuning methods in terms of Rouge scores. The authors also highlight some unique differences between ChatGPT-generated summaries and human references, providing valuable insights into the capabilities of ChatGPT for diverse text summarization tasks. The findings call for further research to systematically examine the characteristics of ChatGPT-generated summaries through extensive human evaluation.[38]
2023 February 16 OpenAI shares a blog post on how ChatGPT's behavior is shaped, and the company's plans for improving the behavior of the system, allowing more user customization, and getting more public input into decision-making. OpenAI explains that unlike ordinary software, their models are massive neural networks that learn from a broad range of data, not programmed explicitly. The models undergo a two-step process: pre-training and fine-tuning. OpenAI claims it is committed to robustly addressing biases in AI systems and being transparent about their intentions and progress. The company reports it is researching how to make the fine-tuning process more understandable and controllable, and building on external advances such as rule-based rewards and Constitutional AI.[39]
2023 February 17 Study A paper explores the potential benefits and limitations of using generative AI models like ChatGPT in Statistical Process Control (SPC) practice, education, and research. The study investigates ChatGPT's ability to provide code, explain concepts, and create knowledge related to SPC by analyzing responses to structured prompts. The results indicate that ChatGPT performs well for structured tasks but struggles with more nuanced tasks, emphasizing the need for proper validation and the use of multiple methods to ensure accurate results when utilizing generative AI models in SPC.[40][41]
2023 February 19 Study A study compares the understanding ability of ChatGPT and fine-tuned BERT on the GLUE benchmark. The study finds that ChatGPT performs well on inference tasks but struggles with paraphrase and similarity tasks. ChatGPT achieves comparable performance with BERT on sentiment analysis and question-answering tasks. The study also shows that the understanding ability of ChatGPT can be improved with advanced prompting strategies.[42]
2023 February 20 Study A paper analyzes over 300,000 tweets and 150 scientific papers to investigate how ChatGPT is perceived and discussed after its release in November 2022. The analysis shows that ChatGPT is generally viewed positively with high quality, but its perception has slightly decreased since its debut. Positive emotions dominate in social media, with joy being the most frequent, while negative sentiment and surprise are on the rise. ChatGPT is perceived more negatively in languages other than English. Scientific papers characterize ChatGPT as a great opportunity across various fields, including the medical domain, but also as a threat concerning ethics, with mixed assessments for education. The authors make their data available to contribute to shaping the public debate and informing future development.[43]
2023 February 20 Study A study explores the use of ChatGPT in zero-shot information extraction (IE) to build IE systems from unannotated text. The authors transform the zero-shot IE task into a multi-turn question-answering problem with a two-stage framework (ChatIE) and evaluate it on three IE tasks: entity-relation triple extract, named entity recognition, and event extraction. ChatIE achieves impressive performance and even surpasses some full-shot models on several datasets, indicating the potential of prompt-based methods to reduce the time and effort that data labeling takes.[44]
2023 February 21 Study A paper evaluates the capabilities of ChatGPT on 25 diverse analytical natural language processing tasks, including sentiment analysis, emotion recognition, offensiveness and stance detection, natural language inference, word sense disambiguation, linguistic acceptability, and question answering. The study found that ChatGPT's average loss in quality was about 25% for zero-shot and few-shot evaluation, and the more difficult the task, the higher the ChatGPT loss. The paper also discusses ChatGPT's bias and the need for validation procedures for such systems.[45]
2023 February 22 Study A paper evaluates the robustness of ChatGPT from the adversarial and out-of-distribution perspective. The authors employ AdvGLUE and ANLI benchmarks to assess adversarial robustness and Flipkart review and DDXPlus medical diagnosis datasets for out-of-distribution (OOD) evaluation. ChatGPT shows consistent advantages on most adversarial and OOD classification and translation tasks, but the absolute performance is still far from perfection, indicating that robustness remains a significant threat. ChatGPT shows astounding performance in understanding dialogue-related texts and tends to provide informal suggestions for medical tasks. The paper concludes with in-depth discussions of possible research directions.[46]
2023 February 23 Study A paper discusses the impact of prompt knowledge on the correctness of answers generated by ChatGPT in the context of consumers seeking health advice from the model. The authors compare the effectiveness of ChatGPT in generating answers when it leverages the model's knowledge alone versus in combination with prompt knowledge. They show that the knowledge provided in the prompt can overturn the knowledge encoded in the model, leading to lower answer correctness. The study has implications for the development of more robust and transparent question-answering systems based on generative pre-trained language models.[47]
2023 February 23 Study A paper presents the first independent evaluation of ChatGPT on mathematical word problems (MWPs) from the dataset DRAW-1K. The authors found that ChatGPT's performance varies dramatically based on the requirement to show its work, with a failure rate of 20% when it provides work and 84% when it does not. The paper also highlights several factors relating to the number of unknowns and operations that increase the probability of failure, with the probability of failure increasing linearly with the number of addition and subtraction operations.[48]
2023 February 26 A paper proposes the use of Software Development Bots (DevBots) trained on large language models to aid in software architecture-centric software engineering (ACSE). The paper specifically focuses on the use of ChatGPT, a natural language processing tool, for collaboration between a novice software architect and the AI model in analyzing, synthesizing, and evaluating a services-driven software application. Preliminary results indicate that ChatGPT can mimic an architect's role but requires human oversight and decision support for collaborative architecting. The authors suggest further research into harnessing empirical evidence and exploring socio-technical aspects of architecting with ChatGPT.[49]
2023 February 27 Study A paper provides an overview of the technology behind ChatGPT. It discusses the historical background of chatbots and the potential applications of ChatGPT in various domains such as healthcare, education, and research. However, the paper also highlights the privacy and ethical concerns surrounding ChatGPT and the limitations of the current version. The paper also includes an interview with ChatGPT and presents its responses to various questions.[50]
2023 February 28 Study A paper presents an empirical study that investigates the performance of ChatGPT in cross-lingual summarization (CLS), which involves generating a summary in a different target language from a source language document. The authors use various prompts to guide ChatGPT to perform zero-shot CLS from different paradigms and provide a preliminary evaluation of its generated summaries. The study finds that ChatGPT initially prefers to produce lengthy summaries with more detailed information, but with the help of interactive prompts, it can significantly improve its CLS performance by balancing informativeness and conciseness. Experimental results on three widely-used CLS datasets show that ChatGPT outperforms the advanced GPT 3.5 model. The paper provides qualitative case studies to demonstrate the superiority of ChatGPT on CLS.[51]
2023 February A Colombian judge reportedly uses ChatGPT to help rule on a case about the medical rights of a child with autism. The case is about whether the child's insurance should cover the bills for his medical treatments. The judge uses ChatGPT in accordance with a law in Colombia that allows the use of virtual tools in some cases. Opinions are divided, with some professionals opposing the use of AI in a legal setting, while others see it as a useful tool.[2]
2023 February Adoption Small businesses in the United Arab Emirates turn to ChatGPT to reduce the amount of time spent on mundane tasks such as writing job descriptions and blog posts, thereby allowing them to concentrate on more creative aspects of their business. By this time, ChatGPT is used by large companies including Meta and Shopify. The Wall Street Journal speculates that the tool could eventually lead to job losses in some areas, although ChatGPT's flaws, including providing inaccurate responses, and the need for further regulation of AI technology so far prevent this for the time being.[52]
2023 February ChatGPT performs poorly when asked to answer questions from Singapore's Primary School Leaving Exam (PSLE), with an average score of 16 out of 100 marks for mathematics and 21 for science. It is found to make errors even in simple addition and can not understand or answer any questions that reference diagrams or graphs. However, ChatGPT manages to pass the English tests and scores an average of 11 out of 20 marks across the three papers it took. The bot's poor performance in exams raises concerns about the possibility of AI bots being used for cheating.[53]
2023 February 25 A paper introduces ChatAug, a text data augmentation approach based on ChatGPT, to mitigate the challenge of limited sample sizes in natural language processing tasks, particularly in few-shot learning scenarios. The proposed approach rephrases each sentence in the training samples into multiple conceptually similar but semantically different samples to ensure the faithfulness and completeness of the generated data. The experiment results demonstrate that ChatAug outperforms state-of-the-art text data augmentation methods in terms of testing accuracy and distribution of the augmented samples.[54]
2023 February 27 The International Baccalaureate says that students can use content created by ChatGPT in their essays, but they must make it clear when they are quoting its responses and treat it as they would any other source. ChatGPT has the ability to produce plausible responses to text prompts, including requests to write essays. The IB, which is taken by thousands of children every year in the UK at more than 120 schools, says that the rise of chatbot technology would mean that essay writing will feature less prominently in future qualifications.[55]
2023 March 1 Emily M. Bender, a computational linguist at the University of Washington, warns about the dangers of humanizing language models like chatbots, stating that the machines are great at mimicry but poor at facts. Bender argues that the big question is how humans will handle themselves around these machines, as we have learned to make machines that can mindlessly generate text but have not learned how to stop imagining the mind behind it.[56]
2023 March 1 Study A paper proposes a generic evaluation framework for Large Language Models (LLMs), specifically ChatGPT and InstructGPT, to assess human personalities based on Myers Briggs Type Indicator (MBTI) tests. The framework includes unbiased prompts, replacing the subject in question statements, and re-formulating question instructions. The study finds that ChatGPT is capable of assessing human personalities and achieves consistent and fairer assessments compared to InstructGPT, albeit with lower robustness against prompt biases. The proposed framework enables flexible assessments of personalities for different groups of people.[57]
2023 March 1 OpenAI launches an API for ChatGPT. The new product allows any business to build the technology into their apps, websites, products and services. Priced at $0.002 per 1,000 tokens, or about 750 words, the API also supports "non-chat" applications. The ChatGPT API is powered by the same AI model behind ChatGPT, dubbed “gpt-3.5-turbo," the most powerful text-generating model OpenAI offers to date through its API suite. It's aimed at driving a range of experiences including an AI-powered tutor, a personalized assistant for shopping recommendations and a virtual tutor feature.[58][59]
2023 March 2 OpenAI announces that its ChatGPT and Whisper models are now available to developers through its APIs. The models can be used for building chatbots and are integrated by companies like Microsoft, Shopify, Snapchat, and Instacart. By this time, ChatGPT costs 90% less than it did in December, making it cheaper for companies to build their chatbots. Meanwhile, Whisper can transcribe speech into text and understand 99 languages, offering potential disruption to Siri and Google Assistant.[60]
2023 March 2 Study A paper presents an analysis of ChatGPT's ability to answer conceptual physics questions related to kinematics and Newtonian dynamics using the Force Concept Inventory (FCI). The study shows that ChatGPT can perform as well as or better than the median performance of a university student who has completed one semester of college physics, although its performance is uneven. The paper raises questions about the use of conceptual assessment tools like the FCI, the implications of ChatGPT's performance for students and physics instruction, and the potential use of ChatGPT as an out-of-classroom aid for physics pedagogy.[61]
2023 March 2 Study A paper presents a methodology to assess the impact of AI language models like ChatGPT on different occupations, industries, and geographies. The authors find that occupations like telemarketers and post-secondary teachers in subjects such as English language, literature, foreign language, and history are the most exposed to language modeling. The industries most exposed to advances in language modeling are legal services and securities, commodities, and investments.[62]
2023 March 3 Competition Microsoft unveils a new AI model, Kosmos-1, which can respond to visual cues or images and is capable of tasks such as visual question answering and image captioning. The model is a Multimodal Large Language Model (MLLM), which allows the user to interact with it through text prompts, images, and videos. Kosmos-1 is seen as a key step toward artificial general intelligence, with researchers suggesting that multimodal perception is required to move beyond ChatGPT-like capabilities to AGI.[63]
2023 March 7 Review A paper discusses the feasibility of using ChatGPT, which is a challenging task due to the lack of annotated data. While by this time ChatGPT demonstrated impressive results in other tasks, it presents challenges for event extraction, which requires complex instructions defining all event types and their schemas. The authors conducted experiments to explore the feasibility of using ChatGPT for event extraction and found that it has only 51.04% of the performance of a task-specific model in complex scenarios. The model is not robust enough and is highly sensitive to different prompt styles, which can result in a poor user experience.[64]
2023 March 12 Competition Together releases OpenChatKit 0.15, the first open-source chatbot alternative to ChatGPT, under the Apache-2.0 license. The chat model has 20 billion parameters and was trained on 43 million instructions, and includes a generic chatbot and the necessary components to create specialized bots. The framework has four main parts, including a large language model tuned for chat, a set of customization recipes, a retrieval system, and a moderation model. The open-source foundation offered by OpenChatKit enables the development of domain-specific and general-purpose chatbots.[65]
2023 March 14 OpenAI announces GPT-4, the latest AI language model in its line of language models. GPT-4 can parse both text and image input but can only respond via text. The company cautions that GPT-4 retains some of the same problems as earlier models, such as the tendency to make up information and generate harmful text. By this time, OpenAI has partnered with companies including Duolingo, Stripe, and Khan Academy to integrate GPT-4 into their products, and it would also be available as an API for developers to build on. GPT-4 is available to the general public via OpenAI's ChatGPT Plus subscription. The company claims that GPT-4 is more creative and collaborative than ever before and can solve difficult problems with greater accuracy.[66] ChatGPT would undergo an update utilizing the GPT-4 language model, which would expand its vocabulary to 25,000 words (eight times more than its previous version). In addition, the update would endow ChatGPT with the capacity to process image inputs as well as text, effectively making it a multimodal system.[7]
2023 March 16 Interview In an interview with ABC News’ Rebecca Jarvis, Sam Altman says that AI technology will reshape society as we know it, but that it comes with real dangers. He also says that feedback will help deter the potential negative consequences that the technology could have on humanity. Altman acknowledges the possible dangerous implementations of AI that keep him up at night, particularly the fear that AI models could be used for large-scale disinformation or offensive cyberattacks. He also says that he fears which humans could be in control of the technology. However, he does not share the sci-fi fear of AI models that do not need humans, stating that "This is a tool that is very much in human control".[67]
2023 March 20 ChatGPT experiences an outage due to a bug in an open-source library which allowed some users to see titles from another active user’s chat history. The bug would since been patched, and the ChatGPT service would be restored with the exception of a few hours of chat history. Upon deeper investigation, OpenAI discovered that the same bug may have caused the unintentional visibility of payment-related information of 1.2% of ChatGPT Plus subscribers active during a specific nine-hour window. OpenAI would reach out to notify affected users that their payment information may have been exposed and is committed to rebuilding trust with the ChatGPT community. Technical details of the bug and the actions taken to improve the system are also provided.[68]
2023 March 21 Competition Google starts the public release of its chatbot, Bard, allowing consumers to sign up to a waiting list for English-language access in the US and UK. The program was previously open to approved testers. Google describes Bard as an experiment in generative AI, which creates rather than identifies content. Bard, unlike ChatGPT, does not generate computer code and its memory is limited. It also includes a feature to show three versions of an answer.[69]
2023 March 23 OpenAI announces ChatGPT implementation support for plugins, which are tools designed for language models to access up-to-date information, run computations, or use third-party services with safety as a core principle. The feature is at first rolled out gradually for a small set of users and is expected to be expanded as more is learned about safety and alignment challenges. Plugin developers can use the provided documentation to build plugins, and the first plugins were created by companies such as Expedia, Slack, Wolfram Research, and Zapier.[70]
2023 March 25 Russian-American podcaster and artificial intelligence researcher Lex Fridman publishes an interview with Sam Altman. They discuss GPT-4, political bias, AI safety, and neural network size. Other topics include AGI, fear, competition, and transitioning from non-profit to capped-profit. They also touch on power dynamics, political pressure, truth and misinformation, anthropomorphism, and future applications, among other topics.[71]
2023 End of year According to OpenAI's estimates, ChatGPT is expected to generate US$200 million in revenue by the end of this year.[7]
2024 End of the year According to OpenAI's estimates, ChatGPT is expected to generate US$1 billion in revenue by the end of this year.[7]

Meta information on the timeline

How the timeline was built

The initial version of the timeline was written by FIXME.

Funding information for this timeline is available.

Feedback and comments

Feedback for the timeline can be provided at the following places:

  • FIXME

What the timeline is still missing


Timeline update strategy

See also

External links

References

  1. 1.0 1.1 Kharpal, Arjun. "Chinese tech giant Alibaba working on a ChatGPT rival; shares jump". CNBC. Retrieved 4 March 2023. 
  2. 2.0 2.1 Bhaimiya, Sawdah. "A Colombian judge used ChatGPT to help rule on a case about the medical rights of a child with autism, reports say". Business Insider. Retrieved 4 March 2023. 
  3. https://www.tooltester.com/en/blog/chatgpt-statistics/#:~:text=ChatGPT%20web%20traffic%20statistics,-ChatGPT%20received%20an&text=The%20site's%20traffic%20has%20grown,62.37%25%20from%20January%20to%20February. Retrieved 25 March 2023.  Missing or empty |title= (help)
  4. DiBenedetto, Chase (18 January 2023). "ChatGPT's surprisingly human voice came with a human cost". Mashable. Retrieved 7 March 2023. 
  5. "The AI Arms Race Is On. Start Worrying". Time. 16 February 2023. Retrieved 6 March 2023. 
  6. Kelly, Samantha Murphy (16 March 2023). "5 jaw-dropping things GPT-4 can do that ChatGPT couldn't | CNN Business". CNN. Retrieved 23 March 2023. 
  7. 7.0 7.1 7.2 7.3 7.4 7.5 7.6 "ChatGPT Statistics 2023". tooltester. Retrieved 22 March 2023. 
  8. "Introducing OpenAI". openai.com. Retrieved 15 March 2023. 
  9. "OpenAI Gym Beta". openai.com. Retrieved 15 March 2023. 
  10. "Universe". openai.com. Retrieved 15 March 2023. 
  11. "Dota 2". openai.com. Retrieved 15 March 2023. 
  12. "Better language models and their implications". openai.com. Retrieved 23 March 2023. 
  13. "https://twitter.com/adamdanielking/status/1125831730848571392?lang=en". Twitter. Retrieved 23 March 2023.  External link in |title= (help)
  14. Delangue, Clément (9 May 2019). "Ethical analysis of the open sourcing of a state-of-the-art conversational AI". HuggingFace. Retrieved 23 March 2023. 
  15. "GPT-2: 6-month follow-up". openai.com. Retrieved 23 March 2023. 
  16. "Emergent tool use from multi-agent interaction". openai.com. Retrieved 15 March 2023. 
  17. "Solving Rubik's Cube with a robot hand". openai.com. Retrieved 15 March 2023. 
  18. "GPT-2: 1.5B release". openai.com. Retrieved 15 March 2023. 
  19. "CLIP: Connecting text and images". openai.com. Retrieved 24 March 2023. 
  20. "Introducing ChatGPT". openai.com. Retrieved 16 March 2023. 
  21. Roose, Kevin (December 5, 2022). "The Brilliance and Weirdness of ChatGPT". The New York Times. Archived from the original on January 18, 2023. Retrieved December 26, 2022. Like those tools, ChatGPT — which stands for "generative pre-trained transformer" — landed with a splash. 
  22. Vincent, James (5 December 2022). "AI-generated answers temporarily banned on coding Q&A site Stack Overflow". The Verge. Retrieved 6 March 2023. 
  23. Whannel, Kate (December 27, 2022). "Could a chatbot answer Prime Minister's Questions?". BBC News. Archived from the original on January 17, 2023. Retrieved December 30, 2022. 
  24. Hartmann, Jochen; Schwenzow, Jasper; Witte, Maximilian (2023). "The political ideology of conversational AI: Converging evidence on ChatGPT's pro-environmental, left-libertarian orientation". doi:10.48550/arXiv.2301.01768. 
  25. "Elon Musk recruits team to develop OpenAI's ChatGPT rival: Reports". Al Arabiya English. 28 February 2023. Retrieved 4 March 2023. 
  26. "Introducing ChatGPT Plus". openai.com. Retrieved 16 March 2023. 
  27. A, Subaveerapandiyan; A, Vinoth; Tiwary, Neelam (1 February 2023). "Netizens, Academicians, and Information Professionals' Opinions About AI With Special Reference To ChatGPT". arXiv:2302.07136 [cs]. doi:10.48550/arXiv.2302.07136. 
  28. Hu, Krystal (2 February 2023). "ChatGPT sets record for fastest-growing user base - analyst note". Reuters. Retrieved 14 March 2023. 
  29. Pascual, M. G. (7 February 2023). "Google presents Bard, its answer to ChatGPT". EL PAÍS English Edition. Retrieved 4 March 2023. 
  30. Khalil, Mohammad; Er, Erkan (8 February 2023). "Will ChatGPT get you caught? Rethinking of Plagiarism Detection". arXiv:2302.04335 [cs]. doi:10.48550/arXiv.2302.04335. 
  31. Bang, Yejin; Cahyawijaya, Samuel; Lee, Nayeon; Dai, Wenliang; Su, Dan; Wilie, Bryan; Lovenia, Holy; Ji, Ziwei; Yu, Tiezheng; Chung, Willy; Do, Quyet V.; Xu, Yan; Fung, Pascale (28 February 2023). "A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity". arXiv:2302.04023 [cs]. doi:10.48550/arXiv.2302.04023. 
  32. Basic, Zeljana; Banovac, Ana; Kruzic, Ivana; Jerkovic, Ivan (9 February 2023). "Better by you, better than me, chatgpt3 as writing assistance in students essays". arXiv:2302.04536 [cs]. doi:10.48550/arXiv.2302.04536. 
  33. Guo, Shuaishuai; Wang, Yanhu; Li, Shujing; Saeed, Nasir (12 February 2023). "Semantic Communications with Ordered Importance using ChatGPT". arXiv:2302.07142 [eess]. doi:10.48550/arXiv.2302.07142. 
  34. Pardos, Zachary A.; Bhandari, Shreya (14 February 2023). "Learning gain differences between ChatGPT and human tutor generated algebra hints". arXiv:2302.06871 [cs]. Retrieved 8 March 2023. 
  35. "India gets its first ChatGPT-powered AI chatbot, 'Lexi': All you need to know". livemint.com. 14 February 2023. Retrieved 4 March 2023. 
  36. Kocaballi, A. Baki (14 February 2023). "Conversational AI-Powered Design: ChatGPT as Designer, User, and Product". arXiv:2302.07406 [cs]. doi:10.48550/arXiv.2302.07406. 
  37. Kindred, Alahna (16 February 2023). "Google issues urgent warning to anybody using AI chatbot ChatGPT". mirror. Retrieved 4 March 2023. 
  38. Yang, Xianjun; Li, Yan; Zhang, Xinlu; Chen, Haifeng; Cheng, Wei (15 February 2023). "Exploring the Limits of ChatGPT for Query or Aspect-based Text Summarization". arXiv:2302.08081 [cs]. doi:10.48550/arXiv.2302.08081. Retrieved 7 March 2023. 
  39. "How should AI systems behave, and who should decide?". openai.com. Retrieved 16 March 2023. 
  40. Megahed, Fadel M.; Chen, Ying-Ju; Ferris, Joshua A.; Knoth, Sven; Jones-Farmer, L. Allison (17 February 2023). "How Generative AI models such as ChatGPT can be (Mis)Used in SPC Practice, Education, and Research? An Exploratory Study". arXiv:2302.10916 [cs]. Retrieved 7 March 2023. 
  41. Megahed, Fadel M.; Chen, Ying-Ju; Ferris, Joshua A.; Knoth, Sven; Jones-Farmer, L. Allison (17 February 2023). "How Generative AI models such as ChatGPT can be (Mis)Used in SPC Practice, Education, and Research? An Exploratory Study". arXiv:2302.10916 [cs]. doi:10.48550/arXiv.2302.10916. 
  42. Zhong, Qihuang; Ding, Liang; Liu, Juhua; Du, Bo; Tao, Dacheng (2 March 2023). "Can ChatGPT Understand Too? A Comparative Study on ChatGPT and Fine-tuned BERT". arXiv:2302.10198 [cs]. doi:10.48550/arXiv.2302.10198. 
  43. Leiter, Christoph; Zhang, Ran; Chen, Yanran; Belouadi, Jonas; Larionov, Daniil; Fresen, Vivian; Eger, Steffen (20 February 2023). "ChatGPT: A Meta-Analysis after 2.5 Months". arXiv:2302.13795 [cs]. doi:10.48550/arXiv.2302.13795. Retrieved 6 March 2023. 
  44. Wei, Xiang; Cui, Xingyu; Cheng, Ning; Wang, Xiaobin; Zhang, Xin; Huang, Shen; Xie, Pengjun; Xu, Jinan; Chen, Yufeng; Zhang, Meishan; Jiang, Yong; Han, Wenjuan (20 February 2023). "Zero-Shot Information Extraction via Chatting with ChatGPT". arXiv:2302.10205 [cs]. doi:10.48550/arXiv.2302.10205. Retrieved 7 March 2023. 
  45. Kocoń, Jan; Cichecki, Igor; Kaszyca, Oliwier; Kochanek, Mateusz; Szydło, Dominika; Baran, Joanna; Bielaniewicz, Julita; Gruza, Marcin; Janz, Arkadiusz; Kanclerz, Kamil; Kocoń, Anna; Koptyra, Bartłomiej; Mieleszczenko-Kowszewicz, Wiktoria; Miłkowski, Piotr; Oleksy, Marcin; Piasecki, Maciej; Radliński, Łukasz; Wojtasik, Konrad; Woźniak, Stanisław; Kazienko, Przemysław (21 February 2023). "ChatGPT: Jack of all trades, master of none". arXiv:2302.10724 [cs]. doi:10.48550/arXiv.2302.10724. Retrieved 7 March 2023. 
  46. Wang, Jindong; Hu, Xixu; Hou, Wenxin; Chen, Hao; Zheng, Runkai; Wang, Yidong; Yang, Linyi; Huang, Haojun; Ye, Wei; Geng, Xiubo; Jiao, Binxin; Zhang, Yue; Xie, Xing (2 March 2023). "On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspective". arXiv:2302.12095 [cs]. doi:10.48550/arXiv.2302.12095. Retrieved 7 March 2023. 
  47. Zuccon, Guido; Koopman, Bevan (23 February 2023). "Dr ChatGPT, tell me what I want to hear: How prompt knowledge impacts health answer correctness". arXiv:2302.13793 [cs]. doi:10.48550/arXiv.2302.13793. Retrieved 6 March 2023. 
  48. Shakarian, Paulo; Koyyalamudi, Abhinav; Ngu, Noel; Mareedu, Lakshmivihari (27 February 2023). "An Independent Evaluation of ChatGPT on Mathematical Word Problems (MWP)". arXiv:2302.13814 [cs]. doi:10.48550/arXiv.2302.13814. 
  49. Ahmad, Aakash; Waseem, Muhammad; Liang, Peng; Fehmideh, Mahdi; Aktar, Mst Shamima; Mikkonen, Tommi (26 February 2023). "Towards Human-Bot Collaborative Software Architecting with ChatGPT". arXiv:2302.14600 [cs]. doi:10.48550/arXiv.2302.14600. 
  50. Shahriar, Sakib; Hayawi, Kadhim (2023). "Let's have a chat! A Conversation with ChatGPT: Technology, Applications, and Limitations". doi:10.48550/arXiv.2302.13817. 
  51. Wang, Jiaan; Liang, Yunlong; Meng, Fandong; Li, Zhixu; Qu, Jianfeng; Zhou, Jie (27 February 2023). "Cross-Lingual Summarization via ChatGPT". arXiv:2302.14229 [cs]. Retrieved 6 March 2023. 
  52. "UAE small businesses embrace ChatGPT as tech allows productivity boost". Al Arabiya English. 15 February 2023. Retrieved 4 March 2023. 
  53. Bharade, Aditi. "ChatGPT failed miserably in Singapore's 6th-grade tests, averaging 16% for math and 21% for science. Days later, it was getting answers right.". Business Insider. Retrieved 4 March 2023. 
  54. Dai, Haixing; Liu, Zhengliang; Liao, Wenxiong; Huang, Xiaoke; Wu, Zihao; Zhao, Lin; Liu, Wei; Liu, Ninghao; Li, Sheng; Zhu, Dajiang; Cai, Hongmin; Li, Quanzheng; Shen, Dinggang; Liu, Tianming; Li, Xiang (27 February 2023). "ChatAug: Leveraging ChatGPT for Text Data Augmentation". arXiv:2302.13007 [cs]. doi:10.48550/arXiv.2302.13007. Retrieved 6 March 2023. 
  55. Milmo, Dan; editor, Dan Milmo Global technology (27 February 2023). "ChatGPT allowed in International Baccalaureate essays". The Guardian. Retrieved 3 March 2023. 
  56. Weil, Elizabeth (1 March 2023). "You Are Not a Parrot". Intelligencer. Retrieved 4 March 2023. 
  57. Rao, Haocong; Leung, Cyril; Miao, Chunyan (2023). "Can ChatGPT Assess Human Personalities? A General Evaluation Framework". doi:10.48550/arXiv.2303.01248. 
  58. Wiggers, Kyle (1 March 2023). "OpenAI launches an API for ChatGPT, plus dedicated capacity for enterprise customers". TechCrunch. Retrieved 4 March 2023. 
  59. "Introducing ChatGPT and Whisper APIs". openai.com. Retrieved 16 March 2023. 
  60. "OpenAI announces ChatGPT API, Snapchat, Instacart and others already using it". GSMArena.com. Retrieved 3 March 2023. 
  61. West, Colin G. (2023). "AI and the FCI: Can ChatGPT Project an Understanding of Introductory Physics?". doi:10.48550/arXiv.2303.01067. 
  62. Felten, Ed; Raj, Manav; Seamans, Robert (2023). "How will Language Modelers like ChatGPT Affect Occupations and Industries?". doi:10.48550/arXiv.2303.01157. 
  63. Nigam, Saumya; News, India TV (3 March 2023). "Microsoft unveils Kosmos-1,a new AI model to race up with ChatGPT". www.indiatvnews.com. Retrieved 4 March 2023. 
  64. Gao, Jun; Zhao, Huan; Yu, Changlong; Xu, Ruifeng (7 March 2023). "Exploring the Feasibility of ChatGPT for Event Extraction". arXiv.org. doi:10.48550/arXiv.2303.03836. 
  65. Shenwai, Tanushree (12 March 2023). "Together Releases The First Open-Source ChatGPT Alternative Called OpenChatKit". MarkTechPost. Retrieved 14 March 2023. 
  66. Vincent, James (14 March 2023). "OpenAI announces GPT-4 — the next generation of its AI language model". The Verge. Retrieved 14 March 2023. 
  67. News, A. B. C. "OpenAI CEO Sam Altman says AI will reshape society, acknowledges risks: 'A little bit scared of this'". ABC News. Retrieved 23 March 2023. 
  68. "March 20 ChatGPT outage: Here's what happened". openai.com. Retrieved 24 March 2023. 
  69. "Google begins opening access to its ChatGPT competitor Bard". Yahoo Finance. Retrieved 21 March 2023. 
  70. "ChatGPT plugins". openai.com. Retrieved 24 March 2023. 
  71. "Sam Altman: OpenAI CEO on GPT-4, ChatGPT, and the Future of AI | Lex Fridman Podcast #367". youtube.com. Retrieved 25 March 2023.