Post #1558768
2026-04-23 03:08 UTC
🔥 take for this audience. LLMs are a dead end for AGI, a dead end for many applications, and that still makes them useful.
They're just not the genies that certain financially incentivized people claim they are.
The big model makers are incredibly wasteful and incredibly not transparent. Why would they be? Their ability to make money hinges on this stuff seeming to be magic.
It's not magic. It's math. We've taught old rocks new tricks.
1/
Replies (1)
-
@raiderrobert@mastodon.social 2026-04-23 03:10
But I will double down on this being even more of a dead end. Every month that goes by without the quantum leap in "intelligence" being proffered by the foundation models while gemma4 and qwen and other open weights models that also walk through their training approach is yet another nail in the coffin for the argument for closed source and closed training sets. These things are not inherently dangerous. Not unless you made them with dangerous things. 2/