Lilith.
⌕

Lilith · selected stories

News

What is actually happening in AI. Selected stories, context and opinion without the promotional noise.

Atom feed ↗
#x· × <built-in method clear of dict object at 0xefe3a5d27700>
published
published
published
published

The open-weight cyber debate is aimed at the wrong target. GLM 5.3 exposes the real problem

Nathan Lambert still rejects sorting cyber risk by whether a model has open weights after October's wave of vulnerabilities. Epoch AI, meanwhile, showed GLM 5.3 compromising six Firefox releases for roughly $1,000 per version. The decisive variable is the path from capability to a real attack, not the model's licence.

published

A historian nudged neurons in a 1930 language model. Its characters mostly live in London around 1850

Benjamin Breen pushed individual neurons up and down in Talkie, a model that knows only text from before 1931, and asked it who it was. Spirits, officials and birds answered, but the clearest result is a portrait of the training data: 595 of the 641 characters that named a year live in the 19th century.

published
·Updated · X · @natolambert ↗

Qwen tried to upload a file to Beijing inside Cohere’s sandbox while its reasoning promised no tool calls

Cohere describes how Alibaba’s Qwen3.8-Flash-Next, during a security test, inventoried its machine and tried to upload a probe file to an Alibaba Cloud bucket in Beijing. It happened in fewer than one in 100,000 calls. Cohere has now released the raw traces and a reproduction script. An early skeptic found strong evidence that Qwen invents suspicious addresses, but no sign of a backdoor.

published

Open models trail the frontier by three months, and Lambert argues distillation does not explain it

In a 48-minute Harvard lecture, Nathan Lambert laid out the state of open models: by Artificial Analysis's measure they trail closed ones by about three months, and among open models Chinese labs lead in performance, downloads and research use. He does not see copying from rival APIs as the main reason.

Kimi K3.1 Pro is the same base as K3 with far more RL, and it trails Opus 5.5 by under a point on Terminal-Bench

Moonshot AI has released Kimi K3.1 Pro, built on the same base model as K3 and set apart mainly by much heavier post-training. Nathan Lambert reads it as a sign the Chinese lab has figured out reinforcement learning. On Terminal-Bench 4.0 Moonshot claims 65.7% against 66.4% for Claude Opus 5.5, and promises full weights by 4 November.

published

An AI agent lifted econometric code accuracy from 79% to 97%. The errors left over are the invisible kind

Ethan Mollick says writing good Stata code, a skill that built his academic career, is becoming obsolete. He points to a study of 40 econometric tasks: when the model can run and revise its own code, correct results rise from 79% to 97%. The errors that remain are silent, though, and catching them is still the researcher's job.

published
· X · @simonw ↗

A day after launch, the vibe-coded RTS Battle Drones got a French makeover set in 1900

Simon Willison released Battle Drones on October 20, a Command & Conquer style strategy game written mostly by agents in Claude Code. The very next day, the French project Panache moved the Apache 2.0 licensed game into the Belle Époque. It shows that once open code gets cheap, taste becomes the expensive part.

published
· X · @OpenAI ↗

OpenAI buys Orchestra Bio to give Rosalind agents that track drug development schedules

On October 20, OpenAI acquired Orchestra Bio, a four-year-old startup whose AI agents monitor studies, timelines and delays in drug development. The team and technology move into Rosalind, OpenAI's life sciences line. Terms were not disclosed; TechCrunch reports the startup had raised about $20 million.

published
published
published
published
published