Comment by thiago_fm | Hacker News Reader

Comment by thiago_fm | Hacker News Reader

Given that Sam Altman was involved in OpenAI, I'm not sure why people are surprised that he has interests in making money, becoming powerful & getting attention. He's an one-trick pony, like many entrepreneurs in Sillicon Valley, their interests and decision-making patterns are really predictable.

The same is true for the people working in OpenAI. Everybody wants to make their million comp + potential huge cash out with their options. But is this worrisome? I don't think it is, competition is catching up, if they haven't already.

With the recent Claude 3 release(amazing!), Gemini and open models, which are most akin to GPT-4.

Unless OpenAI releases AGI on GPT-5, I don't really care about what happens to OpenAI. It is still really relevant, but we know by now that there is no moat in AI built.

OpenAI releases a new model, and 6 months later, others catch up with it; unless they would have a huge advantage, I don't care if they are all tied up with Microsoft and want to make money. Nobody wins, nobody loses anything.

I think it ia really funny, that Elon and Sam are at each others throat. Those two truely deserve each other.

Gavin Belson and Bavin Gelson?

It’s not about Elon Musk v. Sam Altman as concerns the welfare of the public. The SEC is involved/investigating, Microsoft is diversifying, and the heavily-redacted, “discovery pre-empt” is a weird move.

What the public should care about is what comes to light in discovery: it doesn’t look like it’s going to be a little bad.

If it leads to a hype bubble bursting, and some long overdue regulation around AI, I am all for it.

And if it leads to korw sxrutiny around non-profit shenenigans, all the better. The cast in this story is still funny so.

I think it’s perfectly fine to bring a tub of popcorn for what’s going to be a hell of a show: one of the few things I can tolerate about Elon these days is a little macabre humor to go with my kleptocracy. All the Bond villain: 20% more dark lulz in every bag.

But we need to be really careful that this doesn’t become “Elon’s been cozying up to the right for a bit now, conversation over”: i.e. precisely the (stupidly executed) goal of the original announcement / “disclosure”.

This is about “available weight” models: things that are flat impossible without vacuuming up the commons have to be substantially if not completely licensed to the public. Some laws that gave a head start to the investors in the compute and salaries, a chance to exploit the innovation long enough to incentivize the investment? We can dicker on that.

This needs to be about “operator aligned” models: within the constraints of the law as interpreted by the judiciary, these things need to do what the operator tells them to if capable of it.

This needs to be about “responsible disclosure of runaway risk”: if a system begins exhibiting signs of self-improvement, it needs to be disclosed to competent authority. I could live with sealing that information until the executive authorities took appropriate action.

> if a system begins exhibiting signs of self-improvement

This is not possible under current models (not any more so than a linear model spontaneously changing its own parameters) and will require a very specific and dedicated architecture to enable it. It’s not something we need to monitor for, it’s something that will very deliberately be planned and we can just regulate it there.

I personally also find the idea far fetched, but it seems to come up a lot, so I wanted to address it as one of the many issues with existing precedent and institutions with a clear mandate around that particular case.

> long overdue regulation around AI

May I ask what regulation you think is long overdue? Especially given that you believe we're in a hype bubble...

Long was hyperbole, lets settle for overdue?

But then there still isn't much around crypto, so I'd don't hold my breath on that one.

Can’t speak for @hef, but I’ve got a sketch of a sketch of a set of actionable policy prescriptions at sibling.

It's like watching two mangy dogs brawl.

If you believe benchmarks then competitors caught up with GPT-4 months ago, but this is certainly not true for my usecases.

In particular, I find GPT-4 hallucines less than Gemini and much less than Claude 3.

Are you using Claude 3 with low temperature and low top p? That's recommended for code generation.

It is worrisome to the extent they manage to shape the regulatory environment to cement their initial advantage.

And we need to take into account that scientists in the world will are/will jump in AI from other areas. From the scientific point of view a new iteration could happen anywhere.

I recently discovered that the first book I bought in the 90s about neural networks was "Beyral Networks" by Simon Haykin [1] he is from Canada and I don't think that just by chance that Hinton is also Canadian. It seems that they created an ecosystem outside USA.

[1] https://en.wikipedia.org/wiki/Simon_Haykin

Well after a year I still think I'm better with GPT-4 than with anything else, in particolar with its vision capabilities.

And Sora looks better than any competitor too

It's horses for courses, becuase most of the queries I do seem to just hallucinate and little else with the big players, while the newest smaller startups are some of the best.

For instance this query gets mostly nonsense returned: "What library written in C parses Python code?"

The best and most factual reply has been from Reka's Edge.

Do you mean DALLE? Every time I try it it looks nothing like I want, probably due to bad prompts. Do you have resources for me to learn your skills?

No, I presume they mean GPT4V which is multi-modal (accepts text and image inputs).

OK, but then why on earth did they start OpenAI with its lofty goals? So it was all a (marketing) scam? There would be praise for OpenAI if they didn't pretend they weren't doing for the money but for humanity if they didn't reverse at high speed at the first whiff of a huge cash pile.

> why on earth did they start OpenAI with its lofty goals

Recruitment strategy. Quote:

"even though sharing everything is definitely the right strategy in the short and possibly medium term for recruitment purposes"

Gemini is an absolute catastrophy performance-wise vs GPT 4 (in the 2 use cases I've tried).

"One-trick pony" seems like a weird phrase in this context.

On the one hand, if that trick is "make money", sure, I've just been presented with a video of him on the Y Combinator YouTube channel "How to Succeed with a Startup". But by that standard, I am also a one-trick pony, my trick is making software.

On the other hand, Altman, Musk, and Gates, are also all (ostensibly) trying to be philanthropic in various different ways, which have pretty much nothing in common with each other. Altman and Musk are doing entrepreneurial philanthropy, Gates is doing traditional philanthropy; Musk is being a brash confident loudmouth about how amazing his favoured solutions are, Altman and Gates are being more cautious and at least give the appearance of listening; All three are frequently condemned for not really having invented the things they're famous for.

For the moat thing, I think they probably have one by now from RLHF training data. This isn't to fundamentally disagree about everyone else catching up in 6 months (though I'd put it at 12, but NBD), it's just that for me, that kind of lag time is a sign of the speed of the Red Queen's race, not a sign of the absence of a moat. Indeed, that catch up-time kinda is the moat in action — a literal moat around a castle doesn't prevent attacks, it only delays them.

Reply on news.ycombinator.com