Hacker News Reader

Comment by swagmoney1606 | Hacker News Reader

One thing that really bothers me is how much of a contradiction "OpenAI" is. There's virtually nothing open about it.

You can ask ChatGPT if you can run it locally, and it will happily say it is open source, then give 2 github.com links that 404 and then say it doesn't know where the code is if you press it for answers.

ChatGPT wasn't trained on text that talks about its own build process, source code, Github repos, etc. so it doesn't know any more than you and any such answers would be hallucinations.

I can't wait for the inevitable lawsuits that will result when this technology is white-labeled and misrepresents the business to customers or investors.

> I can't wait for the inevitable lawsuits

But there's no ill intent. As long as it's not labeled as human. I guess it applies to everything, even sentient AIs, that it's fair as long as they don't promise services of a human.

It doesn't seem like that would magically protect a business. Otherwise, you could just have an AI write all your terms of service and marketing copy. That way, if you get sued, you can just say "oh, an AI wrote that marketing copy. We never claimed that a human made those promises."

AI grant is another one. It's not actually a grant at all, they take ownership of your company.

There's something about AI tech where it seems important to the owners that it comes across as open and philanthropic, even when it isn't.

Because nobody adopts your product if you advertise yourself as "the Oracle of {new thing}".

The industry has a potent weaponization aspect/risk to it, that the tech overlords recognized a very long time ago. They also know how parts of the tech landscape (eg SV workers) responds to such things as AI being used for military purposes (or similar). So naturally the people in charge of big tech (including the financiers) have to try to lure that labor in to creating what they want, via misleading fakery. Thus it is you get the non-open OpenAI.

That sounds paranoid, but from what I see it is true.

It's a good reminder that names don't mean anything. Words don't mean anything ("don't be evil" comes to mind).

Only actions matter.

well clearly it was open for business.

I remember the first time I read "OpenAI" I thought it was a new project of the Khronos Group consortium

IIRC Musk wanted the auto industry to standardize on an "Open AI" for autopilot, to waive liability and build consensus around ethical decisions.

Maybe he should start with standardizing his charger

They are doing that: https://www.tesla.com/blog/opening-north-american-charging-s...

Does this mean I can buy an adapter and use a supercharger?

I feel like they must've had the relevant XKCD* framed somewhere when they came up with that plan.

* https://xkcd.com/927/

Industry is too early IMO. There’s a huge need to innovate on rapid recharge.

I’d worry standardizing would negatively impact this.

Seeing as right now there are two connectors, Tesla's and literally everybody else, I'm not sure where innovation is necessary.

That we still don’t have rapid recharge close to gas?

I dont know... high voltage and gas fumes dont sound like they mix.

Well done sir.

* literally everybody else _per continent_ please, also CCS is an abomination of a charging port.

If you would let the CCS consortium design a megawatt charger like Tesla‘s V4, mere mortals would probably not even able to lift it.

Disagreed. I'd say there's almost nothing open about it. GPT-1, GPT-2, Whisper, Point e, various papers etc. were all "open" contributions.

Also nothing intelligent.

In what way? The code is public

It's "open" relative to all of the research being IP of Google and FB.

They pulled researchers into their own org and out of FAANG which I think was part of the motivation. They pitched it as open to the public, but I'm not sure that's what it was.

I think opinions also changed because of goal alignment and AGI safety ideas becoming better understood.

FAANG engineers usually publish their research in top NLP conferences ran by the ACL, which forces all published papers to have a permissive license and highly incentives people to contribute code along with the paper for reproducibility reasons.

You don't know what you're talking about. OpenAI has done a lot of self publishing/arxiv to avoid getting slammed in peer review by the fact that none of what they have is open source.

> what they have is open source

How is https://github.com/openai/whisper not open source?

That's the only thing they open sourced. Also not NLP. That's speech to text.

How is speech to text (speech recognition) not NLP?

I used to work in NLP and the team next to me worked in speech to text. Though this was a decade ago and maybe terms have drifted? But it's listed on https://en.wikipedia.org/wiki/Natural_language_processing#Co...

Reply on news.ycombinator.com