He's begging for regulatory capture. "I can destroy the world but I won't. My competitor's will, so regulate them." A shrewd plan considering he's not offering something beyond what another company with a large nvidia cluster could offer.
That's the thing with technology, people get used to it and then trying to ban/control it makes you look ridiculous. It's like how now that Tesla has made it normal to have driving assistance (and calling it FSD) there is little appetite outside of contrarian circles for serious regulation. If, however, regulation was proposed before Tesla shipped then it might have passed.
These sorts of "public/private" control patterns and the destruction of competition while maintaining private monopoly control in the pursuit of "safety"/"equity" is America's most reliable religion.
HN is always full of people pushing for it in one way or the other, so it's not entirely foreign to us either. Nor something we can merely blame on politicians or some intentional conspiracy by a subset of the population.
It's been 4 months, no one has released anything nearly as good as the initial release of ChatGPT. Meanwhile OpenAI has released GPT-4 and is trialing plugins and 32k context.
Either their competition is incompetent or OpenAI is doing something right.
In 2019 they Planned to spend the then 1 bn funding within the next 5 years or so.
So I suppose they’re not far away from net zero at least.
[1]: https://www.reuters.com/business/chatgpt-owner-openai-projec...
Anyway, there have been several open source versions in the gpt3-3.5 range. Those are your base to build off of. Their's just little incentive to try to go participate in a moneysink competition with Google and Microsoft.
Yeah - I'd say a bit of both. Google certainly seem to have dropped the ball, and now seem to be desperately trying to play catch up.
OpenAI certainly seem to be doing a lot right though, in addition to being all-in and laser focused on this tech. It seems there's a lot more to improving these models than just scaling them up. Altman said "there's a lot of understanding went into building GPT-4", and they've invested a lot of effort into the "alignment" aspect - how to control these models to modify behavior - which is what makes the difference between a neat tech demo and something actually useful.
Google home was the perfect chance to showcase their engineering/AI ability, and yet some days it still struggles to interpret a voice command its done 100 times prior, even when you verify that it heard the exact same command you used yesterday.
As with most things, it probably depends on how it's done.
You're limited by two prisoners and separate, sometimes antagonistic, countries. For example, we have little agreement on nuclear weapons, at best we've gotten concessions on testing, a few types of missiles, and so on. Same with current climate legislation. So getting global agreement is hard outside of the bare minimum. Most of the inner county approaches seem to be either panic, political distraction, or like Sam, regulatory capture, as they are ignoring that it means nothing if another country pursues it.
So I'd focus on what simple agreements we could get worldwide.
So why does Bard seem inferior to GPT-4?