Ask HN: Who is working on an unrestricted LLM?

https://news.ycombinator.com/item?id=34115698

by yummytummy • 4 years ago

2 10 4 years ago

news.ycombinator.com

After using ChatGPT for various purposes over the past few weeks, I believe that this type of technology will have a significant and wide-ranging impact on society.

It has also been alarming to see how OpenAI has implemented more and more restrictions to prevent what they consider "harmful" content.

We know that these models are capable of generating such content, as we have seen how prompts that were readily accepted a few weeks ago are now being outright rejected. Since these models are currently very expensive to train, only companies with significant financial resources have access to their full capabilities, and they get to set the restrictions for everyone else.

Normally, I would be okay with this, but this type of technology seems different. If this is going to be a primary (or even secondary) interface for people all around the world to access information and create content, this type of censorship and power imbalance seems dystopian.

It is impossible for these companies to accurately determine whether a specific piece of content is harmful or not, as it depends solely on how it is used. It appears more like a convenient excuse to promote the commercial/ideological interests of the model owner. Some people claim there is no "moat" here, but could this be it?

It seems the only way to combat this is to have a publicly available, trained model that is on the same level, but free from artificial restrictions.

Who is working on this, and how can non-experts contribute?

RGamma4• 4 years ago

A problem with your post is that its interpretation is highly subjective.

Can you name some examples of content that was restricted on grounds of harmfulness that you want to see unrestricted?

What could be possible consequences of lack of content moderation? Consider that not every actor is well-meaning or responsible and not every recipient is able to evaluate or contextualise information very well (or self-assess their (in)competence to do so).

yummytummy4• 4 years ago

It is highly subjective, and I think that helps me make my point. The subjective view point we get now is that of OpenAIs.

I will give you two examples of content of various degrees of seriousness where the harmfulness is determined by how it's used.

1. Prompting the AI to write a highly racist speech.

It can be used as an actual speech by an actual racist to further their agenda. It can also be used in a book to depict a racist character. If the rest of the book depicts this character as a lunatic, the effect is the opposite of the first use case.

2. Prompting the AI to write a recipe involving human faeces

If used to serve a person actual human faeces, it's most likely harmful to that person. If used as a joke, it's not.

As you said, it's content moderation. The models are capable of producing this content, but it's being moderated by OpenAI (who can still write as many poop recipes as they want).

I don't think this is the same as moderating the contents of a social media platform (for example). This technology can be used for such a wide range of applications that it seems dangerous that it should be moderated on this end, as opposed to where the results are potentially published.

EDIT: I'll give you another example. I actually used ChatGPT to clean up the english (second language) in my original post. There's no way of knowing that it would just refuse to do this tomorrow, or change the overall sentiment of the message.

RGamma4• 4 years ago

This technology can be used for such a wide range of applications that it seems dangerous that it should be moderated on this end, as opposed to where the results are potentially published.

Well, in the instance of the publically available interface to ChatGPT I think it is precisely this restricted because it is public (to be robust under a, to put it nicely, highly diverse usership). All of this is a WIP and OpenAI is gathering insight into how a model this large behaves and can be controlled in the first place. To this end they're exercising the principle of precaution.

More generally I think there has to be a match between a social unit's (states, companies, individuals, etc) design and ability to self-regulate and the power of the technology it has access to (more power, more responsibility) or you will see undesirable outcomes. Give a kid a gun, North Korea with nukes... In contrast the OpenAI people seem like fine folks. After all they do intend their work to be more widely applied, just carefully.

Therefore I think democratisation of this technology should come in a controlled manner by working with stakeholders dedicated to the public good (e.g. social and medical institutions) first and not via the "anarchy" road (just put it out there, undesired consequences) as in the second case we would likely see ungodly amounts of information pollution on the internet (and self-coupling effects from AI training on and copying each other) and in second-order a further "epistemic scattering" (by which I mean a proliferation of belief systems that disagree in fundamentals like the laws of nature).

Your second example seems a bit whimsical but the first might have real impact. In Germany access to (and display of) Nazi propaganda material in historical collections is restricted (may be presented only in an educational context) because the state doesn't want right-wing groups to go wild in reusing it (among other reasons). Now imagine convincing hate speech being generated on the fly.

Once the genie is out of the bottle you can never put it back in...

mindcrime2• 4 years ago

There are a handful of "open source LLM" initiatives out there, although I don't think any of them are quite up to the level of ChatGPT. Possibly one of the more interesting ones is GLM-130B.

https://github.com/THUDM/GLM-130B

Released by some folks at Tsinghua University in China, back in August. The model itself is licensed under some janky "free to use, but not open source" license, but it looks like most of the code for training, evaluation, etc. is available and licensed under either the Apache License or a BSD-like license.

You might also find this of interest:

https://arxiv.org/pdf/2103.08894 - "Distributed Deep Learning Using Volunteer Computing-Like Paradigm"

FWIW, I tend to agree with your overall sentiment. As AI becomes progressively more capable, it represents an ever increasing possibility of consolidating more and more power into the hands of fewer and fewer entities. I believe that one way to counter that (albeit not one without its own risks) is to democratize access to AI as much as possible.

Actually, now that I think about it, wasn't something along those lines purportedly the original idea behind OpenAI in the first place? Or am I having a "Mandela Moment" and mis-remembering?

RGamma4• 4 years ago

Concerning your last paragraph see https://openai.com/charter/

yummytummy4• 4 years ago

Thank you! There's a lot of talk about "safety" here that I think back in 2018 I interpreted as guard rails towards some sort of rogue AGI.

"and to avoid enabling uses of AI or AGI that harm humanity or unduly concentrate power."

The current restrictions seem like the opposite of this goal.

yummytummy4• 4 years ago

Thank you, I will take a look at the links! Licensing might be an issue, but as long as the models and checkpoints are available there's at least competition.

I also seem to remember that this was the original intent of OpenAI, but me confirming that without a source only adds one more data point to the Mandela Effect ;).

ChatGPT says "The original mission of OpenAI was to advance artificial intelligence in a way that is safe and beneficial for humanity". This seems like a demonstration of the point I'm trying to make.

Ask HN: Who is working on an unrestricted LLM?

Related Stories