Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

OpenAI's comment to the NTIA on open model weights

openai.com

Post reply on HN
Print view View original

41–50 of 74 posts

Re: OpenAI's comment to the NTIA on open model weights

#41

Post by visarga »

benreesman wrote: post #28
rbren wrote: post #26

Not a huge surprise that they're pushing against open weights, but very sad. I posted my comments on the RFC as well: https://rbren.substack.com/p/banning-open-weight-models-woul...

This is phenomenal piece of writing on this topic. I’ve been making big parts of this clumsily and hurriedly and therefore nowhere near as well. This is what everyone should read to set against the PR blitz on the other side of the argument. Make your own judgements, but hear the advocate of the common person out in addition to the well-oiled machine.

This is indeed a good writeup. Just one small quip:

> We’ll need to clarify copyright law when it comes to disseminating derivative AI-generated works.

Generated content can be either derivative or transformative, and this distinction is important. It's not automatically derivative because

- a model can receive new knowledge and skill demonstrations from the user at test time, that effectively take it out of its initial training distribution (contextual learning)

- the model can draw from multiple sources performing cross-input analysis, such as finding inconsistencies or ranking quality (comparison and cross referencing)

- a model can learn from experimental feedback, such as running code or a complex simulation to see the outcomes, and iterating over the search space. For example AlphaTensor discovered an improved matmul algo (models can discover new knowledge from the environment, they are not restricted to learning from human text)

So models can get new information from users, textual analysis or from experiment based learning. In all these cases it does more than derivative work.

Re: OpenAI's comment to the NTIA on open model weights

#42

Post by visarga »

SirensOfTitan wrote: post #20

Closed weight LLMs are unethical forms of theft that will privatize profits on a work that includes virtually all of humanity’s digital written work and serve to ultimately heighten wealth inequality as they get more sophisticated and start eliminating jobs. The only path forward is open model weights, Sam Altman is on the wrong side of history here, and I hope he fails to convince regulators.

While I agree with your message, it needs more nuance. Second order effects exist: most open models have been boosted with data generated by closed SOTA models. GPT-4 has been the teacher of a whole generation of AI models, closed as it is, and even if OpenAI officially opposed this practice.

Re: OpenAI's comment to the NTIA on open model weights

#44

Post by visarga »

internetter wrote: post #16
jrflowers wrote: post #12

TLDR: OpenAI says it is a moral and safety imperative to pay OpenAI for all eternity

For the copyrighted content they stole

They did make the API public and it was often used for skill distillation by input-output pairs. So they grudgingly contributed to the advancement of open models.

Re: OpenAI's comment to the NTIA on open model weights

#45

Post by visarga »

benreesman wrote: post #15
simonw wrote: post #13

Earlier quoted context omitted.

I don't understand what you're saying about GPT detectors. Are you angry that people are promoting detectors that don't work, or are you angry that OpenAI used to offer one and no longer do?

The latter, in the context of the justification. It was pulled because while it caught on the order of 25 of pure AI output, it missed the rest. But I’m cool with that number, in a world where so much AI in human writing? That’s a total win on low-effort propaganda. Anyone should be cool with that number. Unfortunately they had to pull it because something like 9% of human authored text got hit with the AI flag. Agai…

> Again, some people are starting to write like it. It’s gonna happen.

Interesting point you touched here. Let's do the math for OpenAI: 100M users, with 10k tokens/user/month, that means 1 trillion tokens/month are read by people. That has got to influence speech and circulate information faster between all fields.

Re: OpenAI's comment to the NTIA on open model weights

#46

Post by dang »

rbren wrote: post #26

Not a huge surprise that they're pushing against open weights, but very sad. I posted my comments on the RFC as well: https://rbren.substack.com/p/banning-open-weight-models-woul...

Related ongoing thread:

Banning open weight models would be a disaster - https://news.ycombinator.com/item?id=39901978 - April 2024 (32 comments)

Re: OpenAI's comment to the NTIA on open model weights

#48

Post by Llamamoe »

CharlesW wrote: post #6

My attempt at a TLDR for the piece: • The audiences are policymakers and government agencies like NTIA, the broader AI research community, and existing and potential partners/customers. • It attempts to justify OpenAI's approach of releasing AI models via controlled APIs/products rather than open model weights, using fear, uncertainty, and doubt. • It portrays OpenAI as a thoughtful steward of AI, and is designed to…

And so it serves as yet another reminder that any corporation trying to "do good" is just the usual sociopathic anti-human bullshit doing unusually good PR.

Re: OpenAI's comment to the NTIA on open model weights

#49

Post by maeil »

Llamamoe wrote: post #48
CharlesW wrote: post #6

My attempt at a TLDR for the piece: • The audiences are policymakers and government agencies like NTIA, the broader AI research community, and existing and potential partners/customers. • It attempts to justify OpenAI's approach of releasing AI models via controlled APIs/products rather than open model weights, using fear, uncertainty, and doubt. • It portrays OpenAI as a thoughtful steward of AI, and is designed to…

And so it serves as yet another reminder that any corporation trying to "do good" is just the usual sociopathic anti-human bullshit doing unusually good PR.

Unfortunately that kind of generalization being inaccurate is exactly what makes this problem so difficult. Every case needs to be judged independently, but that takes a lot of time and effort that no one person individually has.

A corporation is purely a legal structure. There exist people who do actually spend their time "doing good", and they too use such legal structures when that's helpful.

It's unfortunate that so many were deceived by Sam Altman, and that the majority of OpenAI employees voted money over honesty when they had a direct, hugely impactful vote. On the other hand, it's not like Altman's history was a closely guarded secret, it was quite easy to look up. So ironically this in itself is a great example of a decemption that could've been prevented quite easily by some cursory research and solidarity.

Post reply on HN