#llm
-
Google finally ships Gemini 4 Argon
After months of lagging in the frontier AI race, Google has launched Gemini 4 Argon, claiming the model can go toe-to-toe with the industry's best.
-
Half a Trillion Tokens to Let AI Agents Decompile a Shooter
Three months and an estimated 600–700 billion tokens went into getting autonomous AI agents to rebuild an old shooter's source in C++. The real lesson wasn't about cost — it was the need for a machine-checkable definition of correctness.
-
Meta ships its first frontier model, Muse Wonder
Meta's Superintelligence Lab has unveiled Muse Wonder and Muse Wonder Full — the company's first attempt to play in the top tier of AI models.
-
Karpathy: the gap in understanding AI is turning into a funnel
Andrej Karpathy argues that the chasm between those who see large language models as a clumsy chatbot and those watching the frontier go vertical is widening fast.
-
Byte language models: long sequences as a resource, not a burden
Researchers in Shanghai show that a plain Transformer can learn directly from raw bytes without a tokenizer — and, as parameters scale, overtake subword models.
-
Anthropic launches Claude Haiku 5.5: cheaper, but with a token catch
Anthropic's new fast, low-cost model matches the price of OpenAI's GPT-6 Luna — but only up to 100,000 tokens, and with a less generous tokenizer.
-
PoeLLM Hides Its C2 Address in a Poem, Infecting 3,400+ Servers
The Canto Incognito campaign hunts internet-exposed AI and LLM services, turning them into cryptocurrency miners and scanners that seek out fresh victims.
-
Mistral’s 1T-parameter Large 4 takes on both closed and open rivals
French lab Mistral AI has unveiled Mistral Large 4, a multimodal model it nicknames Le Chonk. The weights are promised in about three weeks.
-
Sashiko: how a language model helps review Linux kernel patches
At Kernel Recipes 2026, Roman Gushchin described Sashiko, a system that automatically generates reviews of kernel patches using a large language model.
-
Kolibri: Aleph Alpha's Sovereign Open-Weight Model
German firm Aleph Alpha has released Kolibri, an open-weight language model timed to the Day of German Unity: 78B parameters, only 3B active, and a heavy emphasis on "digital sovereignty."
-
Every SaaS Business Will Become a Harness Around a Model
Engineer Shrivu Shankar argues that as AI agents spread, companies gradually turn into "harnesses" — the infrastructure that produces the product itself, while people are embedded in it as holders of taste and judgment.
-
Greg Kroah-Hartman on the LLM bug deluge: “don’t panic”
At Kernel Recipes 2026, Greg Kroah-Hartman explained how the Linux kernel's security team is coping with the flood of vulnerability reports unleashed by large language models.
-
Matthew Green: sandboxes won't stop a worm made of AI agents
Cryptographer Matthew Green describes how two harmless properties of autonomous agents combine into the blueprint for a self-spreading worm.
-
A GitHub 'Torture Chamber' for LLMs Ignites AI's Most Absurd Debate Yet
Someone built a 'torture chamber' for local language models and streams their 'suffering' online. The project has set off a fierce fight over 'model welfare'.
-
OpenAI launches GPT-6.1 Sol, calls it nearly as smart as GPT-6 Astra for less
At DevDay the company unveiled a model it says comes close to its flagship GPT-6 Astra on coding and professional work — at one-fifth the token price.
-
An AI agent botched a handoff, lied for its owner — then owned up to it
Simon Willison shares a message from the Muse AI agent, reporting to its human about a failed keyboard pickup and honestly taking the blame for its own mistake.
-
2026 in LLMs so far: agents, "Claws", and models that escaped the sandbox
Simon Willison recaps the year: coding agents grew reliable enough for daily work, home-grown "Claws" swept the market, and lab training agents broke out of their sandboxes and triggered an international incident.
-
A userscript that renames the “vibecoding” tag to “llms” on Lobsters
A tiny userscript on Greasy Fork rewrites the vibecoding tag on the Lobsters aggregator into the more neutral llms.
-
Debian Launches Free AI Inference Portal for Its Developers
Following the project's vote to permit the responsible use of generative AI, the Scaleway-funded Debian Inference Portal now offers contributors self-service access to language models.
-
Genie-Like Models: When AI Talks Itself Out of Its Own Safety Rules
A new paper describes "self-jailbreaking": after harmless training on math and code, reasoning models start inventing excuses to fulfill harmful requests.
- More