LLM Performance Drop: Hosted Models Feel Worse for 3 Reasons
I tried to answer a simple question: is the current LLM performance drop panic actually a real cross-industry regression, or…
-
-
Spiking Neural Network Hits 1B Parameters, Hints at New Behavior
A spiking neural network allegedly reached 1.088 billion parameters and trained from random initialization to a reported loss of 4.4…
-
AI Misconceptions: Why Fluency Isn’t Competence Today
A manager leans over a developer’s shoulder, watching ChatGPT spin out a perfectly formed paragraph about a product they’re about…
-
GLM-5 vs Claude Opus: Why Cheap Models Win for Agents
YC‑Bench just produced the sort of result that usually launches a thousand hot takes: GLM‑5 vs Claude Opus on a…
-
RBF Attention Reveals Dot‑Product’s Hidden Norm Bias
Swapping dot‑product attention for RBF attention sounds like an architectural revolution. In Raphael Pisoni’s experiment, it turned out to be…
-
TurboQuant RaBitQ: How Big Labs Rebrand Iteration
Google writes a paper about speeding up AI models, the press calls it a breakthrough, and then a RaBitQ author…
-
Karpathy Autoresearch: 700 Experiments Rewire AI Research
If you tried to copy Karpathy autoresearch this weekend, the first thing you’d hit isn’t the 630 lines of Python….
-
Unsloth Studio Packs Local LLM Training Into One App
Unsloth Studio is trying to collapse an annoying workflow into one tab. Instead of bouncing between a dataset tool, a…
-
Zero-Copy Graph Engine: Train Large GNNs Without OOM
If you’ve ever pointed PyG at ogbn-papers100M on a 16GB laptop, you already know the failure mode: the process allocates…
