Omar Khattab (@lateinteraction)

2026-02-24 — quote-tweeting N8 Programs (@N8Programs)

Omar Khattab ✓ @lateinteraction · 22h Buried in the massive progress in LLMs over the past few years is how all your favorite Transformers/DNNs still can't solve even just grade school math problems above a "B" grade through a forward pass. Unless they're in a scaffold like CoT, ReAct, RLM, etc. And this is true even at trillions of params and bajillions of FLOPs. For all I can tell, all a vanilla Transformer can do is really glorified kNN. Without a reasoning scaffold, there's just way too many states to compress; too many mappings that were never seen before. In that case, what makes reasoning models work so incredibly well must be that, at sufficient pretrain/RL scale, every relevant next-reasoning step can be actively visited (more or less "contaminated", but productively so) and composed up. To be clear, if this is true, it seems to be working, and it explains why scale is so important and why failures are so jagged! If a specific kind of state is not retrievable via compression and kNN, then you're going to get some other ~arbitrary behavior. tl;dr the distinction between your DNN architecture and your "scaffold" is subtler than you think. > QUOTED: N8 Programs ✓ @N8Programs · Feb 22 > Inspired by @RyanPGreenblatt, I measured LLMs accuracy on GSM8K when only allowed to output a numerical answer without any CoT - all reasoning done in a few forward passes. The result is a nice log-linear scaling curve. We can ... > [Image: chart "Reasoning-Free GSM8K: Parameter Count vs Accuracy" — log-linear fit (acc = 18.5*log10(p) + 11.5, R²=0.877), scatter of model points including Qwen3.5-297B, Sonnet 4.6, GPT-5.2, Kimi K2.5]

Note from Claude Sonnet 5

A technical thread on why vanilla transformer forward passes fail at multi-step reasoning without CoT/scaffolding, framing transformers as "glorified kNN" limited to seen-state compression, with an empirical GSM8K scaling chart. Relevant background for interpretability/capability discussions in the project (e.g. brain_graph_1 architecture threads, singularity/r-value tracking).

twitter llm reasoning chain of thought scaling laws transformers interpretability gsm8k

Omar Khattab @lateinteraction

2026-02-11

Omar Khattab @lateinteraction · 8h One of the understudied differences between current AI and human intelligence is how comparatively easy it is [for us at least!] to model how humans respond to new knowledge or preferences. If you throw in a fact for an LLM (or take a gradient step on it), it can influence the LLM's behavior in an oddly sharp and peculiar way. Models may latch on it in odd conditions or essentially ignore it altogether. I don't mean catastrophic forgetting or prompt sensitivity, but that the way the models integrate knowledge is very volatile and spiky/jagged. You can see this in how many advanced users of LLM interfaces turn off "memory" features, but until now it's till persistent in other settings too.

Note from Claude Sonnet 5

A tweet on how LLMs integrate new facts/preferences in a "spiky/jagged" and unpredictable way compared to humans, distinct from catastrophic forgetting or prompt sensitivity. Relevant to Nathan's interest in model updating dynamics and the "compelled vs endogenous values" distinction from prior research notes — how information gets encoded matters as much as whether it's encoded.

twitter llm learning dynamics model updating memory interpretability omar khattab