@katiecholden@adocomplete Computer use is an awfully inefficient temporary solution. Businesses will be forced to implement agent friendly interface sooner or later
@jon3k LLM isn’t conscious for sure. However, we have no idea how consciousness works. You can argue that human brain also just fires electrochemical signals betwern neurons too.
@pmitu Strong survival bias. And not all people want to devote their 24/7 to build a successful business when they can make very good living working 9-5
@Glennmon In real companies your impact is rarely coming from your work alone. There’s always a team behind you who helped you get that result. Someone who says “I” did everything in my job is a narcissist.
@_littleplato@AestherML and then see whether it can systematically solve novel compositions far outside the examples it was given. That would be much stronger evidence of reasoning than benchmark performance alone.
@_littleplato@AestherML That’s a great question. It’s hard to know when LLMs are trained on essentially all publicly available human text. I think a stronger test would be to train a model that has never seen a particular formal system or class of problems, teach it the rules only at test time