The Token Reporter

A day in AI

Monday, 5 October 2026

40 stories · 1 from multiple sources

▲ 2 signals · 2 sources

Qwen3.8 27B addition in words

Research: Qwen3.8 27B addition in words Colin Frasier posted on Bluesky about an experiment he ran over two years ago using GPT-4o to see how well it could "compute the sum but return the answer in words" across increasingly large numbers. Here's the chart he shared of those results: I'm confident GPT-4o didn't cheat and use a calculator, especially since it got so many of the calculations wrong,…

simonwillison.net · 5 Oct 2026

Open and Emergent Problems in Agentic Privacy and Security: A Contextual Angle

research.google · 5 Oct 2026

Opus 5.5 agents discover two room-temperature magnetic semiconductor candidates

vals.ai · 5 Oct 2026

More stories

pwasm 0.2a0

simonwillison.net · 5 Oct 2026

llm-anthropic 0.30

simonwillison.net · 5 Oct 2026

AI Companies Are Parasites

coryd.dev · 5 Oct 2026

Questions for Believers in AI Consciousness

endsdontjustifythemeans.com · 5 Oct 2026

OpenAI "rogue" agent activities found on Wikimedia projects

diff.wikimedia.org · 5 Oct 2026

When Redditors come in here and ask why we run LLMs, this is why: Big AI is watching.

reddit.com · 5 Oct 2026

Also today

We swapped AdamW's optimizer states for a Fast Fourier Transform (FFT) to cut VRAM in half. Anyone else trying non-quantization methods?

llama.cpp v0.6.0 released with MTP speculative decoding for Qwen4Exp and lots more

Y'all this is a sexy paper; context language models

Whistle: speech to text in a 16.9MB file

Supercharge regulated workloads with Claude Code and Amazon Bedrock

New agent skill: Amazon SageMaker optimized generative AI inference for your coding agent

How is it possible that qwen 27b is so good? When GPT 4o had a trillion parameters and was worse?

Make no mistake, selling 64 GB DGX Spark variants at the same cost as the original 128 GB is straight drug dealer behavior.

SkyIsNotGreen/Scion-35B-A3B · Hugging Face - Ternary MoE

North 2: Enterprise AI without compromises

Cohere and PwC partner to accelerate secure enterprise AI adoption

Speakrail - a low-latency fully-local voice assistant that runs on a single RTX 4090

Reflection AI's Beam: a highly efficient agentic open model with 501B total parameters and 23B active. - Frontier reasoning efficiency - Advances the Western op

Mythos 5.1, Fable 5.1 and Opus 5.5: Model Welfare

thezvi.substack.com

OpenAI will start watermarking ChatGPT’s text in the EU

This startup is issuing AI-generated acne prescriptions

Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost

All the drama around AI’s takeover of mathematics

Wikipedia operator says OpenAI’s ‘rogue’ bots may be linked to a May outage

Instinct brings its AI agent to group chats, even for friends without an account

TikTok rolls out an AI shopping assistant and one-click checkout

OpenAI is adding text watermarking in ChatGPT and Codex

Hot Girl Hotline is like ‘Dear Abby’ for the AI era

OpenAI PR tells journalist to ‘move on’ while asking Sam Altman about a ChatGPT user’s suicide

Sam Altman says ‘some bad things’ will happen, but AI is totally worth it

HackerRank’s AI interviewer offers a glimpse into what job interviews could become

Our approach to EU text provenance rules

Building advertising for the way people use AI

Florida woman arrested for allegedly making threats in an AI chat

Altman: The world should accept some bad things happening for the benefits of AI

Jonathan Haidt: AI Is the 'Neutron Bomb for Education' [video]

← All editions