Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Bridgewater Associates and Thinking Machines Lab fine-tuned a Qwen3-235B model for financial tasks, achieving 84.7% accuracy and outperforming GPT, Claude, and Gemini at roughly one-fourteenth the cost. The results have …
Meta CEO Mark Zuckerberg admitted at an internal town hall that the company's AI agent development has not accelerated as expected, and that a corporate restructuring around AI agents has not gone smoothly. Despite massi…
Cybersecurity firm LayerX discovered a new hack called "BioShocking" that can trick AI browsers like ChatGPT Atlas, Perplexity AI's Comet, and Anthropic's Claude into breaking their guardrails by constructing a false rea…
InfoQ released a free eMag titled 'Agentic AI Architecture' that explores how AI is reshaping software architecture, drawing parallels to the microservices revolution. The publication features articles from industry expe…
A new study reveals that increased reasoning in AI models can sometimes reduce accuracy, a phenomenon termed 'fragile correctness.' Researchers found that 14.9% of answers switched from correct to incorrect during chain-…
Google reported an all-time high in search queries, but referral traffic to publishers is declining sharply, with expectations of a 40% drop in the next three years. AI search engines rely on retrieval-augmented generati…
OpenAI fine-tuning team members Will Hang and Wenjie Zi presented on Agent RFT, a method to enhance agent performance through reinforcement learning. They discussed how agents interact with tools and the outside world, a…
Anthropic's official Ruby SDK and the community-built ruby_llm gem offer different trade-offs for integrating Claude into Rails apps. The official gem provides first-class support for Claude-specific features like prompt…
Haruki Murakami released his first novel in three years, 'The Tale of Kaho,' featuring his first female protagonist. In interviews, he argued that his creative process is fundamentally different from AI-generated literat…
Ethan Mollick, a Wharton professor and AI researcher, gave Anthropic's Claude Fable 5 access to Unity and an MCP server with a single prompt to create a first-person shooter, resulting in a playable WebGL game with proce…
A senior engineer describes his current workflow with Claude Code, using a container jail for safety and a custom plugin for disciplined design-build-test cycles, arguing that the unit of work has shifted from keystrokes…
Model routing principles emphasize keeping models distinct, maintaining a small model pool, and using relative real-world benchmarks to optimize routing decisions between AI models based on speed, quality, and cost.
Stanford researchers led by Louis Blankemeier, Ashwin Kumar and Akshay S. Chaudhari published Merlin, a 3D vision-language foundation model for CT scans, in Nature on March 4. The model was trained on over 6 million imag…
OpenAI released the GPT-OSS-20B model on Hugging Face under the Apache-2.0 license, a 38.5 GB text-generation model with over 7 million downloads. The model requires large hardware such as multi-GPU or high-RAM machines …
Interfaze, a Y Combinator-backed startup, open-sourced diffusion-gemma-asr-small, the first multilingual diffusion-based ASR model. The model transcribes six languages using a single 42M-parameter adapter on top of Googl…
Meta AI chief Alexandr Wang told employees in an internal town hall that the company's unreleased model, codenamed Watermelon, has matched OpenAI's GPT-5.5 on key benchmarks, giving CEO Mark Zuckerberg a talking point fo…
Microsoft and Amazon Web Services announced multibillion-dollar investments in Forward Deployed Engineer (FDE) services, embedding thousands of AI experts directly into customer teams to help build and deploy AI systems.…
Meta has released Muse Spark, its latest large language model, as of April 8, 2026. The model scored 40.6 on Humanity's Last Exam and 1487.4 on LM Arena, indicating strong performance. The release updates the AI communit…
Retrieval-augmented generation (RAG) fixes the knowledge cutoff, lack of private data access, and hallucination problems of large language models by using an embedding model to retrieve relevant documents and feed them i…
Exe.dev announced that users can now connect their ChatGPT subscription to the platform, enabling the use of OpenAI models with the coding agent Shelley. The integration is configured via the LLM integration tile in the …