Use ChatGPT less and support open source more! New York University professor Nature writes: For the future of science

The free ChatGPT is very fun to use, but the biggest disadvantage of this closed-source language model is that it is not open source. The outside world cannot understand the training data behind it and whether it will leak user privacy. This has also caused issues such as Subsequently, the industry and academia jointly open sourced a series of alpaca models such as LLaMA.

Recently, the Nature Worldview column published an article. Arthur Spirling, professor of politics and data science at New York University, called on everyone to use more open source models. The experimental results can be reproduced, and Comply with academic ethics.

The point is, if one day OpenAI becomes unhappy and closes the language model interface, or if it relies on a closed monopoly to increase prices, then the user can only say one helpless sentence, "After all, academics lost to capital".

##The author of the article, Arthur Spirling, will join Princeton University in July this year to teach political science. His main research direction is political methodology and Legislative behavior, specifically the application of text-as-data, natural language processing, Bayesian statistics, machine learning, item response theory, and generalized linear models in political science.

It seems that a new large language model (LLM) is launched every day, and its creators and relevant people in the academic community will comment on the new model every time. The ability to communicate fluently with humans is generous, for example, you can help users change code, write letters of recommendation, write summaries of articles, etc.

As a political and data scientist who is using and teaching how to use these models, I think academics should be wary because the most popular language models currently remain private and Closed, that is, run by a company, they will not disclose specific information about the basic model, and will only independently check or verify the capabilities of the model, so researchers and the public do not know what files were used to train the model.

The rush to incorporate language models into one's own research process may cause problems and may threaten hard-won progress in "research ethics" and "result reproducibility."

Not only cannot they rely on commercial models, researchers must also work together to develop open source large-scale language models that are transparent and not dependent on the interests of a specific company.

Although commercial models are very convenient and can be used out of the box, investing in open source language models is a historical trend. We must not only find ways to promote development, but also apply the models to future research. middle.

I optimistically estimate that the future of language model tools must be open source, similar to the development history of open source statistical software. Commercial statistical software was very popular at the beginning, but currently almost all communities All are using open source platforms such as R or Python.

For example, the open source language model BLOOM was released in July last year. Its development team Hugging Face is an artificial intelligence company headquartered in New York, working together with more than a thousand volunteers and researchers It is jointly built by people, and part of the research and development funds are provided by the French government; other teams are also working hard to open source large language models.

I think open source projects like this are great, but we also need more cooperation and the pooling of international resources and expertise.

Teams that open source large language models are usually not as well-funded as large companies, and the development team also needs to continue operations to track the latest progress in the field: the AI ​​field is developing too fast Even most language models become obsolete weeks or months after they are introduced.

So the more scholars involved in open source, the better the final open source model will be.

Using open source LLM is crucial for "reproducible research" because closed source commercial language model owners can change their products or their training data at any time, which may cause problems. Change the model's generated results.

For example, a research group might publish a paper testing whether wording suggested by a commercial language model can help clinicians communicate more effectively with patients; if another group attempts to replicate the study, who Do you know whether the basic training data of the model is the same as at that time? Even whether the model is still operational is unknown.

GPT-3, the auxiliary tool commonly used by researchers in the past, has been replaced by GPT-4. All research based on the GPT-3 interface is likely to be unable to be reproduced in the future. For companies, , keeping the old model running is not a high priority.

In contrast, using open source LLM, researchers can view the model’s internal architecture, weights, understand how the model operates, customize the code and point out errors. These details include the model’s Adjustable parameters and data to train the model, community involvement and oversight all help keep this model robust in the long term.

The use of commercial language models in scientific research also has negative implications for research ethics because the text used to train these models is unknown and may include users on social media platforms. direct messages or content written by children.

While the person producing the public text may have agreed to the platform’s terms of service, this may not be the standard of informed consent that researchers want to see.

In my opinion, scientists should stay away from using these models in their work as much as possible. We should move to open language models and promote them to others.

Also, I don’t think academics, especially those with large social media followings, should push others to use commercially available models because if prices spike, or the company goes out of business, the researchers may I will regret promoting the technology to my colleagues.

Researchers can currently turn to open language models produced by private organizations, such as LLaMA, which is open sourced by Facebook parent company Meta. It was initially issued based on user application and review, but The full version of the model was subsequently leaked online; Meta's open language model OPT-175 B

is also available. The downside in the long run is that the release of these models relies too much on the benevolence of the company. , this is an unstable situation.

In addition to this, there should be a code of conduct for academics working with language models, as well as corresponding regulatory measures, but these all take time. According to my experience as a political scientist, I expect that these regulations will be very imperfect at first and will be slow to take effect.

At the same time, support is urgently needed for large-scale collaborative projects to train open source language models for research, such as CERN, the International Organization for Particle Physics, The government should increase funding through grants.

The field is evolving at lightning speed and coordination of domestic and international support needs to begin now.

The scientific community needs to be able to assess the risks of resulting models, and releases to the public need to be cautious, but it is clear that an open environment is the right thing to do.

The above is the detailed content of Use ChatGPT less and support open source more! New York University professor Nature writes: For the future of science. For more information, please follow other related articles on the PHP Chinese website!