Anthropic's Breakthrough Is RSI, Not True AGI
Critical context on the new Anthropic blog: 1, AGI is *harder* than RSI (as used below). AGI: machine can do anything human can do, autonomously [not achieved] RSI (as used below): AI is a useful coding tool that humans can leverage [achieved]; it’s great at (some) code optimizations The results in the blog are about RSI, not AGI. Getting to AGI will require new ideas, not just new code optimizations. So we don’t need to panic yet. 2. Technical note: Mythos and Claude Code are neurosymbolic systems; this isn’t a victory for pure scaling per se, it’s victory for harnesses and symbolic tools. Deep learning did hit a wall; neurosymbolic AI rescued it.* The new results show progress on AI; they don’t show that that progress is general, and they don’t show that massive data centers will be critical. —— *My own claim was never that AI would hit a wall, or that deep learning should be abandoned, but that we would need to supplement it with neurosymbolic AI, in order to address the limitation of pure scaling. And that’s exactly what happened. Go back and reread my 2022 paper if you don’t believe me.
Neurosymbolic AI Will Outperform LLMs, Render Big Bets Premature
Three Predictions: 1. Some form of AI, probably neurosymbolic in nature, will come that is far more economical and data- and energy-efficient than LLMs, and it will make an absolute fortune. 2. LLMs, on the other hand, will never be all...
Hyperscalers Adopt Pay‑Per‑Use Pre‑IPO Due to Cash Crunch
⚠️Why didn’t the hyperscalers wait until after the IPO in switching to pay-by-usage-charging? My guess is that *they literally could not afford to* —because it would bankrupt them. My reasoning: in “all you can eat” agent era they were...
Neurosymbolic AI Complements Deep Learning, Proving 2022 Prediction
People who say this kind of thing are completely lost about what I actually said about deep learning, and I would strongly encourage them to read “Deep learning is hitting a wall” (2022). What I said there (and in 2018...
Pretrained Models Help, but AI Needs Deeper Reasoning Foundation
This was right five years ago, and still is: “Large scale pretrained models are certainly likely to figure prominently in artificial intelligence for the near future, and play an important role in commercial AI for some time to come. The results...
Real-World Messy Data Cripples AI Demos, Costs Billions
“The mechanism is always the same in every story I've been covering. The demo works in a controlled environment with clean inputs. The deployment fails because real kitchens, real intersections, and real warehouses produce messy inputs the demo never tested....
Scaling Obsession Delayed AGI; Neurosymbolic Focus Needed
If we had done everything I suggested in my 2020 arXiv article “The Next Decade in AI”, we might actually have reached AGI by now. In the last three years, after a detour driven by the false promise of pure scaling,...
Neurosymbolic Model Outperforms OpenAI’s Hurried Erdos Win
neurosymbolic by @swarat et al for the Erdos win, with much more careful, quantitative work than openai’s in hindsight i wonder whether OpenAI rushed theirs out, knowing this was coming?
Current AI Falls Short of True AGI Criteria
AGI is certainly not here by the definitions I have repeatedly laid out. or by criteria that @hendrycks @Yoshua_Bengio and I and others recently laid about at https://t.co/ogf90ZCVKs i don’t think any current AI can even reliably do any of the...
Experts Tackle Hard Questions Beyond LLMs
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...
Claude Code Proves Neurosymbolic AI Beats Pure LLMs
🤩🤯🤩 Claude Code (still not AGI but biggest advance since GPT-4) is the most neurosymbolic thing I have ever seen in my life. 53 symbolic tools, 500,000 lines of symbolic code, combined with a state-of-the-art LLM. It is categorically *not* a...
LLM Progress Now Incremental, Not Game‑changing
When is the last time a general purpose LLM (putting aside hybrid systems like Claude Code with special purpose symbolic harnesses) last completely blew away all competing prior models? GPT-4 relative to GPT 3.5? That’s what incremental change with no real moat...
Generative AI’s $1.6 T Odds Beat Roulette’s Zero
Which has better odds? Generative AI earning $1.6 trillion/year or the roulette wheel landing on zero? https://t.co/52cTL3iSHM
AI Must Reach $1.6 Trillion—Four Times Google’s Peak
One estimate of how much annual revenue AI needs to “make sense”: 1.6 trillion. That’s four times what Google made in its best year. (total revenue so far is perhaps on order of 100 billion.)
Denies “AI Just Regurgitates” Claim, Cites Hallucination Warnings
Dear @geoffreyhinton, I literally never said that AI systems “JUST regurgitate”; that’s plainly false. I don’t believe it, and I didn’t say it. (They do *sometimes* regurgitate, and the evidence for that is overwhelming.) I further discuss the rest of your...