Difference between revisions of "Timeline of transformers"
From Timelines
| Line 41: | Line 41: | ||
|- | |- | ||
| 2020 || June 11 || || OpenAI releases Generative Pre-trained Transformer 3 ({{w|GPT-3}}) in beta. | | 2020 || June 11 || || OpenAI releases Generative Pre-trained Transformer 3 ({{w|GPT-3}}) in beta. | ||
| + | |- | ||
| + | | 2023 || February 15 || || A paper presents a pilot study that evaluates the cognitive abilities of two recently released generative transformer models, ChatGPT and DALL-E 2, in decision-making and spatial reasoning. The study constructs input prompts following neutral a priori guidelines and finds that DALL-E 2 can generate at least one correct image for each spatial reasoning prompt, but most images generated are incorrect. Similarly, ChatGPT demonstrates some level of rational decision-making, but many of its decisions violate at least one of the axioms under the classical Von Neumann-Morgenstern utility theorem. ChatGPT's outputs tend to be unpredictable and can make irrational decisions for simpler problems while drawing correct conclusions for more complex bet structures. The paper discusses the challenges of scaling up such cognitive evaluations for generative models and conducting them with a closed set of answer keys.<ref>{{cite journal |last1=Tang |first1=Zhisheng |last2=Kejriwal |first2=Mayank |title=A Pilot Evaluation of ChatGPT and DALL-E 2 on Decision Making and Spatial Reasoning |journal=arXiv:2302.09068 [cs] |date=15 February 2023 |doi=10.48550/arXiv.2302.09068 |url=https://arxiv.org/abs/2302.09068 |access-date=7 March 2023}}</ref> | ||
|- | |- | ||
| 2023 || February 18 || || A paper evaluates the performance of Generative Pre-trained Transformer (GPT) models for machine translation, covering various aspects such as the quality of different GPT models, the effect of prompting strategies, robustness towards domain shifts and document-level translation. The experiment includes eighteen different translation directions involving high and low resource languages, as well as non English-centric translations. The results show that GPT models achieve competitive translation quality for high resource languages, while having limited capabilities for low resource languages. Hybrid approaches, which combine GPT models with other translation systems, can further enhance the translation quality. The paper provides valuable insights for researchers and practitioners in the field to understand the potential and limitations of GPT models for translation.<ref>{{cite journal |last1=Hendy |first1=Amr |last2=Abdelrehim |first2=Mohamed |last3=Sharaf |first3=Amr |last4=Raunak |first4=Vikas |last5=Gabr |first5=Mohamed |last6=Matsushita |first6=Hitokazu |last7=Kim |first7=Young Jin |last8=Afify |first8=Mohamed |last9=Awadalla |first9=Hany Hassan |title=How Good Are GPT Models at Machine Translation? A Comprehensive Evaluation |journal=arXiv:2302.09210 [cs] |date=17 February 2023 |doi=10.48550/arXiv.2302.09210 |url=https://arxiv.org/abs/2302.09210}}</ref> | | 2023 || February 18 || || A paper evaluates the performance of Generative Pre-trained Transformer (GPT) models for machine translation, covering various aspects such as the quality of different GPT models, the effect of prompting strategies, robustness towards domain shifts and document-level translation. The experiment includes eighteen different translation directions involving high and low resource languages, as well as non English-centric translations. The results show that GPT models achieve competitive translation quality for high resource languages, while having limited capabilities for low resource languages. Hybrid approaches, which combine GPT models with other translation systems, can further enhance the translation quality. The paper provides valuable insights for researchers and practitioners in the field to understand the potential and limitations of GPT models for translation.<ref>{{cite journal |last1=Hendy |first1=Amr |last2=Abdelrehim |first2=Mohamed |last3=Sharaf |first3=Amr |last4=Raunak |first4=Vikas |last5=Gabr |first5=Mohamed |last6=Matsushita |first6=Hitokazu |last7=Kim |first7=Young Jin |last8=Afify |first8=Mohamed |last9=Awadalla |first9=Hany Hassan |title=How Good Are GPT Models at Machine Translation? A Comprehensive Evaluation |journal=arXiv:2302.09210 [cs] |date=17 February 2023 |doi=10.48550/arXiv.2302.09210 |url=https://arxiv.org/abs/2302.09210}}</ref> | ||
Revision as of 13:02, 7 March 2023
This is a timeline of FIXME.
Contents
Sample questions
The following are some interesting questions that can be answered by reading this timeline:
Big picture
| Time period | Development summary | More details |
|---|
Full timeline
| Year | Month and date | Event type | Details |
|---|---|---|---|
| 2017 | June | Google researchers first describe the transformer algorithm that would turbocharge the power of chatbots. | |
| 2018 | June 11 | OpenAI releases a paper entitled Improving Language Understanding by Generative Pre-Training, in which they introduces the Generative Pre-trained Transformer (GPT).[1] | |
| 2019 | February 14 | OpenAI releases Generative Pre-trained Transformer 2 (GPT-2). | |
| 2020 | June 11 | OpenAI releases Generative Pre-trained Transformer 3 (GPT-3) in beta. | |
| 2023 | February 15 | A paper presents a pilot study that evaluates the cognitive abilities of two recently released generative transformer models, ChatGPT and DALL-E 2, in decision-making and spatial reasoning. The study constructs input prompts following neutral a priori guidelines and finds that DALL-E 2 can generate at least one correct image for each spatial reasoning prompt, but most images generated are incorrect. Similarly, ChatGPT demonstrates some level of rational decision-making, but many of its decisions violate at least one of the axioms under the classical Von Neumann-Morgenstern utility theorem. ChatGPT's outputs tend to be unpredictable and can make irrational decisions for simpler problems while drawing correct conclusions for more complex bet structures. The paper discusses the challenges of scaling up such cognitive evaluations for generative models and conducting them with a closed set of answer keys.[2] | |
| 2023 | February 18 | A paper evaluates the performance of Generative Pre-trained Transformer (GPT) models for machine translation, covering various aspects such as the quality of different GPT models, the effect of prompting strategies, robustness towards domain shifts and document-level translation. The experiment includes eighteen different translation directions involving high and low resource languages, as well as non English-centric translations. The results show that GPT models achieve competitive translation quality for high resource languages, while having limited capabilities for low resource languages. Hybrid approaches, which combine GPT models with other translation systems, can further enhance the translation quality. The paper provides valuable insights for researchers and practitioners in the field to understand the potential and limitations of GPT models for translation.[3] |
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
- ↑ Radford, Alec; Narasimhan, Karthik; Salimans, Tim; Sutskever, Ilya (11 June 2018). "Improving Language Understanding by Generative Pre-Training" (PDF). OpenAI. p. 12. Archived from the original (PDF) on 26 January 2021. Retrieved 23 January 2021.
- ↑ Tang, Zhisheng; Kejriwal, Mayank (15 February 2023). "A Pilot Evaluation of ChatGPT and DALL-E 2 on Decision Making and Spatial Reasoning". arXiv:2302.09068 [cs]. doi:10.48550/arXiv.2302.09068. Retrieved 7 March 2023.
- ↑ Hendy, Amr; Abdelrehim, Mohamed; Sharaf, Amr; Raunak, Vikas; Gabr, Mohamed; Matsushita, Hitokazu; Kim, Young Jin; Afify, Mohamed; Awadalla, Hany Hassan (17 February 2023). "How Good Are GPT Models at Machine Translation? A Comprehensive Evaluation". arXiv:2302.09210 [cs]. doi:10.48550/arXiv.2302.09210.
