
Steven Groves
10.3K posts

Steven Groves
@stevengroves
ProRelevant Marketing VP, Online Strategist, Economic Gardener - personal account. Social Strategy topics at @SocialMktConvo; ProRelevant stuff at @ProRelevant



Yesterday I was talking with an AI professor, who taught many of today’s AI leaders, who I won’t name (for a reason I will get to in a second). He laid out that LLM’s won’t get us to AGI. A new architecture is needed, he told me. One built round human cognitive abilities. He says most of his former AI students are math geniuses, but don’t understand human beings and how they learn, which led them down the LLM path of “get more data and more NVIDIA cards.” LLMs are struggling to learn reasoning and will always hallucinate, he says. They also require huge datacenter expenditures ands huge pile of data to build them. He says a new approach from the ground up could get us to AGI with far less compute, more privacy, and be highly personalized. Think of a robot that could learn on its own to do new things. The problem is that most he talks with think his ideas are too weird. Which is why I am keeping his name out of this and it isn’t @GaryMarcus or @ylecun. Here is to the weird ones who see the world differently than anyone else. I have learned to listen to them, because every once in a while they turn out to be right. But until that happens they are just crackpots telling everyone they are wrong. Ahhh, humans are so fun, aren’t they? Anyway his point is that it would take two years to build this new approach and get us to AGI. Crackpot thinking. :-)










Gigantic Announcement for Language Models That Run on your CPU!💥📣 We are releasing: - GPT4All-Snoozy: the strongest local LLM that runs on your private CPU hardware! - The first local OS native LLM app verified by Apple ! Try it at: gpt4all.io











