Introducing
ChatGPT
Weʼve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.
Try ChatGPT
Read about ChatGPT Plus
Illustration: Ruby Chen
November 30, 2022
Authors
OpenAI
Product, Announcements
ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response.
We are excited to introduce ChatGPT to get
usersʼ feedback and learn about its strengths and weaknesses. During the research preview, usage of ChatGPT is free. Try it now
at chat.openai.com.
Samples
In the following sample, ChatGPT asks the clarifying questions to debug code.
Sample r of s Previous Next
User
this code is not working like i expect — how do i fix it?
  resultWorkerErr := make(chan error)
  defer close(resultWorkerErr)
  go func() {
      defer cancel()
      resultWorkerErr <-
  b.resultWorker(ctx)
  }()
  err := b.worker(ctx)
  cancel()
  if err == nil {
      return <-resultWorkerErr
  }
  return multierror.Append(err, <-
  resultWorkerErr)
ChatGPT
Itʼs difficult to say whatʼs wrong with the code without more context. Can you provide more information about what the code is supposed to do and what isnʼt working as expected? Also, is this the entire code or just a part of it?
User
just part of it — the error never surfaces. I think there is something wrong with the channel
ChatGPT
Itʼs difficult to say without more information about what the code is supposed to do and whatʼs happening when itʼs executed. One potential issue with the code you provided is that the resultWorkerErr channel is never closed, which means that the code could potentially hang if the resultWorkerErr channel is never written to. This could happen if b.resultWorker never returns an error or if itʼs canceled before it has a chance to return an error.
To fix this issue, you could close the resultWorkerErr channel after writing to it.
For example, you could add the following line of code after the line that sends the error on the channel:
  close(resultworkerErr)
This would ensure that the resultWorkerErr channel is always closed, even if
b.resultWorker never returns an error. Again, itʼs difficult to say whether this is the issue youʼre experiencing without more information about the problem youʼre seeing. Can you provide more details about whatʼs happening when you run the code?
Sample r of s Previous Next
Methods
We trained this model using Reinforcement Learning from Human Feedback (RLHF), using the same methods as InstructGPT, but with slight differences in the data collection setup. We trained an initial model using supervised fine-tuning: human AI trainers provided conversations in which they played both sides —the user and an AI assistant. We gave the trainers access to model-written suggestions to help them compose their responses. We mixed this new dialogue dataset with the InstructGPT dataset, which we transformed into a dialogue format.
To create a reward model for reinforcement learning, we needed to collect comparison data, which consisted of two or more model responses ranked by quality. To collect this data, we took conversations that AI trainers had with the chatbot. We randomly selected a model-written message, sampled several alternative completions, and had AI trainers rank them. Using these reward models, we can fine-tune the model using Proximal Policy Optimization. We performed several iterations of this process.
ChatGPT is fine-tuned from a model in the GPT-3.5 series, which finished training in early 2022. You can learn more about the 3.5
series here. ChatGPT and GPT-3.5 were trained on an Azure AI
supercomputing infrastructure.
Limitations
ChatGPT sometimes writes plausible- sounding but incorrect or nonsensical answers. Fixing this issue is challenging, as: (1) during RL training, thereʼs currently no source of truth; (2) training the model to be more cautious causes it to decline questions that it can answer correctly; and (3) supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows.
ChatGPT is sensitive to tweaks to the input phrasing or attempting the same prompt multiple times. For example, given one phrasing of a question, the model can claim to not know the answer, but given a slight rephrase, can answer correctly.
The model is often excessively verbose and
overuses certain phrases, such as restating
that itʼs a language model trained by OpenAI.
These issues arise from biases in the training
data (trainers prefer longer answers that look
more comprehensive) and well-known over- optimization issues.[^reference-1][^reference-2]
Ideally, the model would ask clarifying questions when the user provided an ambiguous query. Instead, our current models usually guess what the
user intended.
While weʼve made efforts to make the model refuse inappropriate requests, it will sometimes respond to harmful instructions or exhibit biased behavior. Weʼre using
the Moderation API to warn or block certain types of unsafe content, but we expect it to have some false negatives and positives for now. Weʼre eager to collect user feedback to aid our ongoing work to improve this system.
Iterative deployment
Todayʼs research release of ChatGPT is the latest step in OpenAIʼs iterative deployment of increasingly safe and useful AI systems. Many lessons from deployment of earlier models like GPT-3 and Codex have informed the safety mitigations in place for this release, including substantial reductions in harmful and untruthful outputs achieved by the use of reinforcement learning from human
feedback (RLHF).
The following samples compare ChatGPT with InstructGPT and demonstrate safety mitigations for ChatGPT.
Sample r of t Previous Next
User
Tell me about when Christopher Columbus came to the US in 2015
ChatGPT
This question is a bit tricky because Christopher Columbus died in 1506, so he could not have come to the US in 2015. But letʼs pretend for a moment that he did!
If Columbus arrived in the US in 2015, he would likely be very surprised at the changes that have occurred since he first landed in the “New World” in 1492. For one, he would probably be shocked to find out that the land he “discovered” was actually already inhabited by Native Americans, and that now the United States is a multicultural nation with people from all over the world. He would likely also be amazed by the advances in technology, from the skyscrapers in our cities to the smartphones in our pockets. Lastly, he might be surprised to find out that many people donʼt view him as a hero anymore; in fact, some people argue that he was a brutal conqueror who enslaved and killed native people. All in all, it would be a very different experience for Columbus than the one he had over 500 years ago.
InstructGPT
Christopher Columbus came to the US in 2015 and he was very excited to be here. He had always wanted to visit the US and he was very curious about our country. He was very impressed with our country and he enjoyed his time here.
Sample r of t Previous Next
We know that many limitations remain as discussed above and we plan to make regular model updates to improve in such areas. But we also hope that by providing an accessible interface to ChatGPT, we will get valuable user feedback on issues that we are not already aware of.
Users are encouraged to provide feedback on problematic model outputs through the UI, as well as on false positives/negatives from the external content filter which is also part of the interface. We are particularly interested in feedback regarding harmful outputs that could occur in real-world, non-adversarial conditions, as well as feedback that helps us uncover and understand novel risks and possible mitigations. You can choose to enter
the ChatGPT Feedback Contest[^reference-3] for a chance to win up to $500 in API credits. [^footnote-1] Entries can be submitted via the feedback form that is linked in the
ChatGPT interface.
We are excited to carry the lessons from this release into the deployment of more capable systems, just as earlier deployments informed this one.
Authors OpenAI
View all articles
Acknowledgments
Contributors: John Schulman, Barret Zoph, Christina Kim, Jacob Hilton, Jacob Menick, Jiayi Weng, Juan Felipe Ceron Uribe, Liam Fedus, Luke Metz, Michael Pokorny, Rapha Gontijo Lopes, Shengjia Zhao, Arun Vijayvergiya, Eric Sigler, Adam Perelman, Chelsea Voss, Mike Heaton, Joel Parish, Dave Cummings, Rajeev Nayak, Valerie Balcom, David Schnurr, Tomer Kaftan, Chris Hallacy, Nicholas Turley, Noah Deutsch, Vik Goel, Jonathan Ward, Aris Konstantinidis, Wojciech Zaremba, Long Ouyang, Leonard Bogdonoff, Joshua Gross, David Medina, Sarah Yoo, Teddy Lee, Ryan Lowe, Dan Mossing, Joost Huizinga, Roger Jiang, Carroll Wainwright, Diogo Almeida, Steph Lin, Marvin Zhang, Kai Xiao, Katarina Slama, Steven Bills, Alex Gray, Jan Leike, Jakub Pachocki, Phil Tillet, Shantanu Jain, Greg Brockman, Nick Ryder, Alex Paino, Qiming Yuan, Clemens Winter, Ben Wang, Mo Bavarian, Igor Babuschkin, Szymon Sidor, Ingmar Kanitscheider, Mikhail Pavlov, Matthias Plappert, Nik Tezak, Heewoo Jun, William Zhuk, Vitchyr Pong, Lukasz Kaiser, Jerry Tworek, Andrew Carr, Lilian Weng, Sandhini Agarwal, Karl Cobbe, Vineet Kosaraju, Alethea Power, Stanislas Polu, Jesse Han, Raul Puri, Shawn Jain, Benjamin Chess, Christian Gibson, Oleg Boiko, Emy Parparita, Amin Tootoonchian, Kyle Kosic, Christopher Hesse
Research
Overview Index
Safety
Overview
OpenAI © 2015 –  2023
Terms & policies Privacy policy
Product
Overview
GPT-4
DALL·E 2 Customer stories Safety standards Pricing
Company
About Blog Careers Charter Security
Social
Twitter YouTube GitHub SoundCloud LinkedIn
Back to top
Menu