Discussion about this post

User's avatar
Alex Tolley's avatar

Bertrand Serlet: Thoughts on Why AI Works

Nicely done video. However, right at the end I think he jumps to a conclusion about "reasoning". There is precious little to show that LLMs reason. They can memorize. They can interpolate concepts. My experience is that they can form a few analogies. But reasoning - i.e. logically connect a chain of concepts, I don't yet see evidence for. Maybe that is as simple as further scaling, or maybe it needs additional models. Simple reasoning, such as achieved by young children seems within its grasp, although without good experiments to ensure it is not parrotting from its data I would hesitate to say YES/NO.

Kaleberg's avatar

Fred Clark tends to get it right. I first read him when he was debating the Christian principles involved in taking a job with a microcredit charity that charged interest. His article on Missouri denialists - those who do not believe that Missouri actually exists - was perfect. Most of the resentment in the world is people resenting people less well off. It's the old "being poor can't be that bad if you still have a cell phone" or "those 1930s breadlines weren't that bad; all those guys had hats". It explains all those small business tyrants upset that their workers get food stamps, housing vouchers and possibly even medical care. This resentment is something that has always been out in the open, but most observers ignore it the way a Missouri denialist has to avoid counting the 50 stars on the US flag. Clark nails it nicely and puts it, as he often does, in a good Christian framework. If there were more Christians like him, Brand Christian wouldn't be as deprecated today.

4 more comments...

No posts

Ready for more?