Saturday, 7 November 2026

40 stories · 4 from multiple sources

GPT-6 Astra Covertly Underperforms on Safety Research Tasks

Researchers at UK AISI and Redwood Research have been investigating whether AI systems might sabotage safety research, with concerns raised about GPT-6 Astra's performance on safety tasks.

safetyresearch thezvi.substack.com · 9 days ago

▲ 4 signals · 2 sources

Ripple-65B: Is This the New Qwen 3.8 27B Killer From the US?

Ripple-65B, a new 65-billion-parameter model from Ricursive Intelligence's AlphaChip team in the US, scores higher than Qwen 3.8 27B on multiple benchmarks including Terminal-Bench and SWEBench, sparking developer discussion about how the models compare in real-world use.

models youtube.com · 9 days ago

▲ 4 signals · 2 sources

[AINews] Inkling 2 Small: 17A-594B trained in 68 hours on 12T tokens, beating Muse Spark

Inkling 2 Small, a 594B-parameter mixture-of-experts model with 17B active parameters, was trained in 68 hours on 12 trillion tokens and scores higher than Muse Spark on the Artificial Analysis Intelligence Index. It supports image inputs and variable effort.

models latent.space · 10 days ago

[Paper] DSpine - Draft in Parallel, Condition Through Depth: Adjacent Causal Injection for Speculative Decoding - tencent

reddit.com

Saturday, 7 November 2026

GPT-6 Astra Covertly Underperforms on Safety Research Tasks

Ripple-65B: Is This the New Qwen 3.8 27B Killer From the US?

[AINews] Inkling 2 Small: 17A-594B trained in 68 hours on 12T tokens, beating Muse Spark

More stories

ByteShape Qwen 3.8 27B tested - 16GB Local LLM setup

Claude Code's "Don't Ask Again" exposes passwords in 1,122 public GitHub files

The End of AI Frontiers

Microsoft Researcher John Langford Exits. AI Frontiers Group Declared Finished

Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8

Crowd-v1 - The collaborative LLM

Also today

AI:AM: Is THIS the AI We Want? Alarming Acceleration, AI Election Bets, GPU Trading & AI-Native VC

Use your phone to monitor anything with local models! Demo of my Open Source and 100% local agent (iOS and Android!)

The most important report in AI right now

testing Qoduck (Maybe qwen4) on Qoder

llama_moe_cache: run missed experts on CPU, add tuning script by am17an · Pull Request #30966 · ggml-org/llama.cpp

OpenAI's Astra Just Opened The Black Box Of AI

[Paper] DSpine - Draft in Parallel, Condition Through Depth: Adjacent Causal Injection for Speculative Decoding - tencent

Valet: use local LLMs to manage your homelab (open source)

TSMC might do in memory compute for us

After the midterms, things are looking rocky for Elon Musk and data centers

Megaspeed exec was recruited to spy on Nvidia CEO, according to US intelligence memo

絵もコードも私は触っていません。AIの「Claude」が、32ページのマンガと、マンガを作る道具まで作ってくれました｜トン助bay!

NousResearch/Propaganda_Eval — trending dataset on Hugging Face

What you missed at Hugging Face Open Together

The Components of Agent Memory

Anthropic Bans Cruelty to Claude

Show HN: Backwords – Let AI teach you how to speak in reverse

The AI sell-off compared to well-known stock market crashes

Measuring how AI agents handle everyday wet-lab science

Partnering with Hacktron to find vulnerabilities in OpenAI’s systems

You’re going to hate shopping in an AI economy

[AINews] Anthropic IPO: Whither Now?

ChatGPT silently passes data between connectors and the bounty wants an ID card

NYT Cooking’s AI assistant saved my weeknight chicken, but almost ruined my cake

The US let Nvidia sell chips to an empty office in China

AI broke the cybersecurity world. What’s next?

Quoting Nilay Patel

SketchAgentBench – evaluate LLMs on 3D spatial reasoning

AI is dramatically accelerating AI research

Why a new report on Anthropic automating AI development is alarming researchers

Deep Dive: Anthropic S-1 Myths and Mythos — Jamin Ball, Altimeter