10^7-Dimensional LLM Memory, but Only If it Stays Sparse
A BDH seminar summary circulating in recent technical discussion frames LLM memory as a tradeoff between the familiar transformer KV…
-
-
LLM Failure Modes Start in the Stack, Not the Chat
LLM failure modes are easiest to understand if you stop treating them as personality flaws, “the model lied,” “the chatbot…
-
66 Tokens Make a Diffusion Language Model Look Easy
A diffusion language model generates text by starting from masked or otherwise corrupted tokens and iteratively restoring them. In this…
-
Language-Agnostic Representations Show a Shared Semantic Workspace
The standard story is that LLMs work in words. They predict the next token, so surely their internal reasoning is…
-
The Abstraction Fallacy Makes Conscious AI Harder to Prove
Alexander Lerchner’s paper on conscious AI does something unusual: it does not start by asking whether today’s models seem conscious….
-
Claude Code Lost Its Thinking Budget: and Broke Workflows
A few weeks ago, one of the most useful facts about Claude Code stopped being visible. Not the code it…
-
Open Models Are Winning code arena rankings by Fitting the Loop
A strange thing happened to code arena rankings. They stopped being just a nerdy scoreboard and started acting like a…
-
Supercomputer Breach: Why One Hack Matters More Than It Seems
A supercomputer breach is not just a bigger data breach. If CNN’s reporting on the alleged compromise of China’s National…
-
Meta’s Muse Spark Changes AI Competition Because a reasoning model Is Now a Distribution Layer
Meta has launched Muse Spark, a new reasoning model that it says is competitive on multimodal, health, and agentic tasks…
