Hacker News Reader

Comment by appplication | Hacker News Reader

Apologies, but I’m seeing this everywhere so I have to quip on this: OpenAI is no closer to developing AGI than you or I am. Transformer models and the GPT series are huge breakthroughs, but they are not AGI and there’s no indication the path to AGI is via transformer models. It seems increasingly like transformers/LLMs are just a local optima on the path to AGI.

View on HN

Right, right, nothing surprising about computers suddenly creating original works of art, poetry, legal arguments, bedtime stories and screenplays. AGI? Pshaw. Wake me up when they’ve proved the Riemann Hypothesis using nothing but a collection of vintage cookbooks.

But OpenAI believe that they will develop AGI, maybe they will, and may they won’t, and also highly unlikely if they don’t close the next round, but that misses the point.

They believe that they will and therefore we need to ask is their governance structure fit for purpose if they achieve their goals?

AI researchers have been making breakthroughs and proclaiming that AGI is "just around the corner" since 1959.

There's absolutely no reason to believe OpenAI is right about this when thousands of experts have been wrong. AGI is deceptively difficult because it's almost impossible to see how far away you are from the goal. You can see how far you've come and extrapolate, but that's always a mistake.

They operate inside a legal framework. If they do achieve AGI, various parliaments will quickly pass laws. Congress is likely to take action.

It has been done for weapons, it has been done for cryptography in the 80s, it has been discussed for vaccines during the pandemic, it has been done for finance and surely it will be done for AGI.

Reply on news.ycombinator.com