you need both for optimal function; that’s the claim i have been making for 30 years. neuro+symbolic. as a point of fact though, most of waymo’s LLM stuff at least as of last summer was still experimental. i am not actually sure how much lift they are getting from LLMs in
@garymarcus
-
LLMs and production code debugging limitations
By
–
pure llms aren’t what’s debugging production code
-
The shift from pure LLMs to neurosymbolic AI architectures
By
–
The pure LLM debate – which I had for many years, here and elsewhere – is indeed no longer relevant. Why? Because I won; nobody uses pure LLMs anymore. Nowadays all deployed objects are neurosymbolic, which was exactly the point of my infamous 2022 paper, Deep Learning is
-
Technical Discussion on LLM Reasoning and Architectural Iteration
By
–
literally i didn’t say that. adding “reasoning” already borrows tools like iteration and evaluation from classical AI and isn’t a pure LLM. and the reasoning has all kinds of problem. and i didn’t say “just”; i was careful to say “basically”, suggesting an approximation.
-
Debating LLM scaling versus symbolic integration for AI progress
By
–
this is a such a muddle. (at least relative to my views) LLMs are more or less just autcomplete, but (as I have always said) they have their uses. And the real progress now is coming from adding new (symbolic) techniques to the mix, not from pure scaling.
-
Critique of pure LLM architecture and the role of symbolic integration
By
–
I love AI, it’s pure LLMs I hate. Pure LLMs *are* basically just autocomplete. Recent progress (e.g. Claude Code) doesn’t show otherwise Rather, lot of the progress in the last two years has come from *introducing* other things – mainly classic symbolic techniques and tools,
-
Critical Perspective on GPT Development and AI Capabilities
By
–
What I am about to describe ain’t AGI; it’s a sign of a trillion dollar trainwreck. If I had told you in 2022 that the 2026 version of GPT (which by the way would only be GPT 5.5 and not GPT-6 or 7 like many people fantasized about) would still have strange quirks like inserting
-
Study finds memory in LLM agents remains unreliable
By
–
Breaking new study: memory in LLM agents still can’t really be trusted, even after over trillion dollars has gone into the development of the field.
-
Moving beyond LLMs toward world models in AI research
By
–
Truly an all-star cast, on one of the most important questions in AI. Thrilled to see some many people finally willing to confront the hard questions of how we can move beyond LLMs, and into what world models are really about.
-
LLM limitations versus AGI capabilities
By
–
Real AGI would not do this. Even after a trillion dollars in LLMs still do.