/MachineLearning

this is pretty much worst case performance no harness at all and very simplistic prompt

Holy crap. This is the genre of software that's in the most danger: - Kind of mid in quality - Highly niche use-cases - It's been winner takes all for the space in the past - Often involved special formats or protocols And now Claude Code can just reverse engineer it. 🤯

DeepSeek just dropped a banger paper to wrap up 2025 "mHC: Manifold-Constrained Hyper-Connections" Hyper-Connections turn the single residual “highway” in transformers into n parallel lanes, and each layer learns how to shuffle and share signal between lanes. But if each layer can arbitrarily amplify or shrink lanes, the product of those shuffles across depth makes signals/gradients blow up or fade out. So they force each shuffle to be mass-conserving: a doubly stochastic matrix (nonnegative, every row/column sums to 1). Each layer can only redistribute signal across lanes, not create or destroy it, so the deep skip-path stays stable while features still mix! with n=4 it adds ~6.7% training time, but cuts final loss by ~0.02, and keeps worst-case backward gain ~1.6 (vs ~3000 without the constraint), with consistent benchmark wins across the board

We finally had a moment to run our system with GPT-5.2 X-High on ARC-AGI-2! Using the same Poetiq harness as before, we saw results as high as 75% at under $8 / problem using GPT-5.2 X-High on the full PUBLIC-EVAL dataset. This beats the previous SOTA by ~15 percentage points.