Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Anthropic disabled access to its Fable 5 and Mythos 5 AI models for all customers after the Trump administration issued an export control directive citing national security concerns. The order, received Friday, requires …
A developer building a conversational analytics engine found that language models produce confident wrong answers when querying raw database schemas. To solve this, they created a domain graph—a metadata map built offlin…
Anthropic benefits from the US government's export ban on Fable/Mythos, validating its claim that the AI model was too dangerous to release. The ban provides a free pass to miss revenue targets, creates compute breathing…
Anthropic is treating the possibility of AI welfare seriously, testing its Claude models for signs of morally relevant internal states like negative self-image, but critics argue the tests may conflate conversational tac…
Apple announced a new server-side large language model running on Private Cloud Compute (PCC) at WWDC 2026, accessible via the same Swift API as the on-device model. The PCC model offers 32K context, reasoning capabiliti…
A researcher trained a cheap Gemma 2B judge to detect misalignment in AI agents, but testing against Anthropic's AuditBench showed the judge failed to reduce audit costs or reliably distinguish misaligned behavior, with …
Anthropic disabled two of its most advanced AI models, Claude Mythos Preview and related models, after a U.S. government export control directive cited national security concerns. The company had previously hyped the mod…
OpenRouter's Fusion model outperformed DeepSeek-V4-Pro in a head-to-head coding test, delivering a complete, compilable Go rate limit parser with correct edge cases while DeepSeek V4 Pro made basic errors including missi…
Anthropic released its Mythos-class AI model as Fable with guardrails, but a trusted partner discovered a jailbreak exposing advanced cyber capabilities. The US government asked Anthropic to fix the vulnerability or de-d…
The US government issued an export-control directive suspending foreign-national access to Anthropic's Fable 5 and Mythos 5 models, citing national security concerns over a reported jailbreaking method. Anthropic must di…
A developer expresses anger over the takedown of a capable AI model called Fable, arguing that powerful institutions are centralizing AI capability and gating access behind clearances and vetting. The author warns that t…
Researchers Wenfang Sun, Xinyuan Song, Pengxiang Li, Lu Yin, Yefeng Zheng, and Shiwei Liu published a paper on arXiv titled 'The Curse of Depth in Large Language Models,' investigating performance degradation in deep LLM…
TensorZero, an open-source LLMOps platform used by Fortune 10 companies, archived its GitHub repository on June 12, 2026, after failing to achieve commercial product-market fit. The founders returned unused capital to in…
A user on Hacker News asked how to prevent large language models from getting stuck in local minima during conversations, noting that LLMs often cling to ideas despite counterarguments. The user seeks techniques to break…
Malware authors are embedding nuclear and biological weapons language into malicious code to evade AI-powered security scanners, as flagged by Citizen Lab researcher John Scott Railton. The tactic triggers safety-tuned l…
Software developers Johannes Link and Jakob Schnell examine the ethical implications of generative AI, weighing its benefits against energy consumption, e-waste, misinformation, and intellectual property concerns. They a…
Xiaomi's MiMo team open-sourced MiMo Code, a terminal-based coding agent designed for long-horizon automated programming tasks. The agent addresses challenges in decision quality and state continuity over dozens or hundr…
Meera, a GenAI engineer at AcmeAI, builds a customer-support agent using LangGraph that leadership loves, but risk head Sanjay halts the launch with demands for replayable interactions, tamper-evident audit logs, and reg…
Anthropic disabled its Claude Fable 5 and Claude Mythos 5 models after a U.S. export control directive, but CEO Dario Amodei disputed the shutdown's technical basis, saying the government's evidence was verbal and center…
A developer built a system running three coding agents non-stop for three days using a headless mode, task queue, worker isolation, and a local Qwen model to avoid API costs. The system orchestrates multiple AI coders th…