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 released NanoEuler, a GPT-2-scale language model built entirely from scratch in C/CUDA without any machine learning libraries. The project includes a hand-written byte-level BPE tokenizer, pretraining on book…
A developer fixed a broken fallback chain in their OpenClaw agent that was causing request timeouts during peak hours. The new chain includes seven entries: two local Ollama models, three OpenRouter free models, and two …
Researchers from Peking University developed ECG-R1, a multimodal vision-language model trained with reinforcement learning to differentiate ischemic from nonischemic T-wave inversion on ECGs. The model, accepted at ICML…
Subquadratic, a Miami AI startup, claims to have broken through the transformer attention bottleneck that limits large language models, introducing Subquadratic Sparse Attention (SSA) to enable efficient long-context pro…
A cautious optimist argues that AI safety, interpretability, bias, and alignment are as critical as raw capability, warning that the same technology enabling vaccine research could also design pathogens. The author empha…
Google's AI Overview is treating fictional entities from the SCP Foundation horror wiki as real, with no disclaimers. The AI's pattern-matching mistakes the wiki's deadpan bureaucratic style for factual documentation, le…
A new approach to memory systems for long-running AI agents transitions from episodic to procedural memory, enabling agents to accumulate knowledge and skills over days. This advancement addresses the limitations of flat…
IEEE launched a five-course online program, Large Language Models Demystified, to help technical professionals master LLMs for engineering tasks. The program covers APIs, retrieval-augmented generation, data security, an…
Strands Labs released Simple Strands Agent (SSA), a minimal autonomous coding harness that achieves state-of-the-art results on SWE-Bench Verified, SWE-Bench Pro, and Terminal Bench 2 across 21 models. The open-source to…
VladTV published an article on June 19, 2026, summarizing a members-only video interview in which AI safety researcher Roman Yampolskiy repeated his longstanding claim that there is a 99.9% chance artificial intelligence…
Huawei's Ascend 910C chips successfully completed post-training for DeepSeek's V4-Pro model, a 1.6-trillion-parameter AI system, using a cluster of at least 1,000 chips. This marks a major advance for China's semiconduct…
Engineering teams are discovering that expanding AI context windows degrades agent reliability due to the 'lost in the middle' problem, where models fail to attend to information in the middle of long prompts, causing si…
Apple's visionOS 27 update brings Siri AI and other features to all Vision Pro models, but the M2 Vision Pro misses out on voice customization and the most advanced on-device AI model, which are exclusive to the M5 model…
Generative AI coding assistants are making programming language popularity less relevant, as tools like Claude Code can work with older languages like Perl if sufficient documentation exists. The author's experience with…
A new open-source plugin called Hallucinating Canary provides early warnings of response degradation in Claude Code sessions by detecting context loss through planted marker tokens. The plugin monitors whether these 'can…
OpenTelemetry's graduation to stable status within the CNCF provides a standardized observability framework for production LLM pipelines, addressing challenges in tracing inference requests across vector databases and em…
Google's Gemini API now supports streaming text-to-speech, enabling developers to reduce latency in AI voice applications by generating audio from text chunks before the full response is complete. This approach improves …
The Trump administration imposed export controls on Anthropic's Fable model, limiting access to U.S. citizens and forcing Anthropic to make the model unavailable. Ben Thompson argues that the incident highlights Anthropi…
Cohere introduced a new scheduling system for its multi-tenant LLM platform that uses rate limiting, performance tiers, and deficit round robin to prevent one customer's traffic burst from degrading latency for others, e…
A new open-source Python toolkit called PhD_fleet allows researchers to manage a virtual lab of AI PhD students via Slack, using Claude Code agents for research tasks and a coach agent for mentoring feedback. The tool ru…