OpenAI co-founder on past approach to openly sharing research: ‘We were wrong’

Openness is probably the reason chatGPT was possible in the first place. Machine learning research has been able to progress so rapidly precisely because information has been all open and publicly available.

I think trying to turn research back into company silos is going to be a death sentence for innovation in the sector. What happened with stable diffusion was amazing for generative art. But I guess openAI has learnt from what they probably see as a wasted opportunity and a mistake.

Now, let's talk about datasets and information as well; these LLMs all rely on the wealth of public information we have created as societies, so it seems to me pretty scummy to simply exploit that wealth without really giving much in return. I don't think it's going to be possible to put the genie back in the bottle however, I think people in the tech sector are avid for publicly available LLMs after tasting what happened with stable diffusion. We'll see, in any case, I think this is a good business decision by "open"AI; though they might want to rebrand.

Also, let me just say this: safety is a much more encompassing issue that needs to be discussed in the social and political sphere. These decisions need to be made much more democratically. The amount of people that I've seen disappointed of the safeguards that chatGPT offers, and why? Who decided those were the right safeguards to put?

Anyway, boo for openAI but chatGPT is pretty cool.

Such nonsense of unbelievable proportions. Remind me whose comments are in the model (almost all of their software is parameters, the code is a tiny bit). It's a shame people are no longer using GPL and other copyleft licenses to enforce openness. Thankfully researchers will be pissed off enough that they will stop enabling their behavior and start building competing systems

The decades old poison of Micro$oft: Embrace, extend, and extinguish.

On the hardware level, it feels like computing just has gone back to the era of mainframes. Hopefully the era of PC soon comes and everybody can start running their own "PC" for fun.

The precedent has been set by ClosedAI. They don't even talk about number of parameters. That's the most disturbing aspect of this action. Imagine Google starts enforcing its patents.

PatheticAI, this is like IBM Watson all over, just that if you try exposing the raw API to your customers you really can't guard against jailbreaks/guardrails.