Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A developer patched llama.cpp to improve prompt processing throughput by 20% when using Multi-Token Prediction (MTP) on GPU, recovering performance lost to MTP overhead. The proof-of-concept processes only the output row…
The US government has ordered OpenAI to stagger the release of GPT-5.6, restricting initial access to trusted partners and effectively placing frontier AI models under government control. This move, inspired by Anthropic…
Margaret Atwood criticized AI chatbot Claude's accuracy at the Babell Literary and Cultural Festival, recounting how it gave her a wrong spoiler about the TV series Father Brown. She attributed the error to the "garbage …
Author Margaret Atwood criticized AI chatbots as unreliable, citing her experience with Anthropic's Claude giving incorrect information about a TV show. She warned that AI outputs are only as good as their training data,…
DeepSeek released DSpark on June 27, a speculative decoding framework that accelerates per-user generation speeds by 60% to 85% on its DeepSeek-V4 Flash model and 57% to 78% on the Pro variant. The framework uses a semi-…
Between April 22 and June 5, 2026, operators linked to Alibaba's Qwen AI lab ran 25,000 fraudulent accounts through 28.8 million exchanges with Claude, in the largest adversarial distillation operation recorded. Anthropi…
Anthropic's Claude Fable 5 AI model is expected to be restored within days after being taken offline following a Trump administration export control directive, according to a new report. The Commerce Department has alrea…
A peer-reviewed PNAS Nexus study found that leading large language models, including GPT-4o, Claude 3.5 Sonnet, GPT-5, Claude Opus 4.1, and Gemini 2.5, fail catastrophically on simple cognitive tasks like the color Stroo…
Maneshwar, developer of git-lrc, explains multi-agent AI systems using a honeybee colony analogy. The post covers agent autonomy, coordination patterns like networks and hierarchies, and benefits such as flexibility and …
US export bans on frontier AI models like Anthropic's Mythos and Fable 5 are forcing global developers to abandon single-API architectures in favor of multi-model orchestration and localized redundancy. Sakana AI's Fugu …
Boris Cherny of Anthropic, Peter Steinberger, and Addy Osmani of Google are promoting 'loop engineering,' a workflow where developers design automated loops that prompt AI agents instead of typing prompts manually. This …
A new guide argues that teams should stop defaulting to GPT-4 for every task, calling it a design decision that is often wrong. The article provides practical advice on selecting the right AI model for specific use cases…
A developer argues that comparing a senior engineer's $250K salary to $20K in AI coding tokens is a misleading trap. The engineer's role involves far more than code generation, and AI-written code still requires senior h…
Microsoft launched MAI-Code-1-Flash, its proprietary AI coding model, to general availability for GitHub Copilot Business and Enterprise subscribers. The model offers rapid, low-latency code generation for professional d…
Shrijith Venkatramana, building git-lrc, explains that sequence transduction—transforming one sequence into another—was the original problem that led to modern large language models. Early neural networks like RNNs and L…
Boris Cherny, creator of Claude Code at Anthropic, shared 13 practical tips for using the tool, including running 5 instances simultaneously in separate git checkouts, using Opus 4.5 with thinking mode for all tasks, and…
Anthropic's AI model Fable 5 could be available again within days as the Trump administration prepares to lift restrictions imposed on June 12 over safety concerns, pending sign-off from the Pentagon and NSA.
DeepSeek open-sourced DSpark, a speculative decoding framework that accelerates per-user generation for DeepSeek-V4 by 57–85% over the MTP-1 baseline. The framework pairs a parallel draft backbone with a Markov head and …
A guide titled 'The Claude Leverage Kit' outlines 10 habits and a one-hour setup to maximize productivity with Claude AI, claiming users can achieve 10x output by avoiding common traps and investing in reusable context f…
A developer shares seven rules for making AI agents reliable in production, arguing that flakiness is usually a specification problem rather than a model problem. The rules include making success criteria falsifiable, us…