Teaching an AI My Taste
Lon Riesberg, editor of the Data Elixir newsletter, built an LLM-based judge to automate the final curation of links, teaching it his taste via a 111-line markdown file called taste.md and testing it against a spreadshee…
Full-text search across 71678 articles. Combine with topic and date filters; results sorted by relevance.
Lon Riesberg, editor of the Data Elixir newsletter, built an LLM-based judge to automate the final curation of links, teaching it his taste via a 111-line markdown file called taste.md and testing it against a spreadshee…
A developer proposes building programming languages, runtimes, and tools designed specifically for large language models (LLMs) rather than adapting human-oriented tools, citing security, feedback, introspection, and sca…
OpenAI claims 10,000 concurrent agents solved the Navier-Stokes problem, a 90-year-old fluid-motion question with a $1 million Clay prize, in 88 hours, but says it will not claim the bounty. NYU mathematician Tristan Buc…
OpenAI reported that its research organization used 3.1 agent-workdays for every human workday by mid-August, a figure that measures aggregate machine runtime (24.8 agent-hours per eight-hour human workday) rather than a…
Anthropic confirmed that hackers using infostealer malware have been hijacking Claude subscribers' login sessions and draining their paid usage without stealing passwords, affecting users like independent AI consultant G…
OpenAI released GPT-6 Astra, a new language model positioned as a major step up from GPT-5.6 Soul, with hands-on testers reporting the largest measurable jump in 3D spatial reasoning and describing the gap as a "magnitud…
Nvidia CEO Jensen Huang dismissed fears that AI will kill SaaS, calling the notion 'the most illogical thing in the world' and saying markets 'got it wrong' after SaaS stocks tumbled. Shopify reported that merchants crea…
WorkBuddy, a desktop AI agent app, is offering free access to Tencent's Hi-4 Preview model for two weeks from launch, allowing users to try the 770 billion parameter mixture-of-experts model without payment. The model, r…
Hermes Kanban, a durable task board for multi-agent profile collaboration, is now available, allowing multiple named agents to work on tasks via a shared SQLite database and a dedicated kanban_* toolset. The board suppor…
A developer built a consent-gated knowledge pipeline for Tencent RTC's social-messaging community chat, ensuring messages become visible to MCP-based AI assistants only after explicit author consent and moderator review.…
Amazon Web Services (AWS) published a technical post detailing how to migrate agentic workloads to Amazon Bedrock AgentCore, a platform for building and optimizing agents at scale. The post maps ten operational burdens—i…
Anthropic's engineering team is advancing the discipline of harness engineering, which focuses on the systems surrounding AI models—context, tools, execution environments, and feedback loops—that enable reliable agentic …
Google's Gemini 3.8 Flash, released three weeks after Gemini 3.7 Flash, scores 73.7% on Deep SWE v1.1, edging out GPT 5.6 Soul's 72.7% and matching Claude Opus 5, while priced at $0.75 per million input tokens and $3.75 …
Anthropic released FrontierSWE v2, a benchmark with 21 new ultra-long horizon tasks totaling 34, and reported that its Claude Fable 5.1 model leads the ranking, followed by GPT-5.6 and GLM-5.3. The revised benchmark uses…
PyTorch 2.14, released by the PyTorch team, introduces NVGEMM for CuTeDSL-generated CUTLASS kernels, a new nccl2 backend for distributed training, fault tolerance as a first-class concept, native linear algebra on Apple …
TypeFire, a free Mac app combining a text expander, clipboard manager, and Markdown notes, now supports AI-powered snippets via Apple Intelligence, Claude, OpenAI, Gemini, and Ollama, with on-device processing and a 7-da…
Anshu Chimala, who led software engineering and design teams at Apple for 12 years, shares techniques in a newsletter post for unlocking AI's creative potential in design, arguing that typical LLMs produce bland output b…
A new testing guide from Bright Security argues that the riskiest part of an enterprise AI agent is not the model but the authority wrapped around it, and that security testing must cover the execution chain—what the age…
A developer argues that AI should be used to understand legacy systems before rewriting them, emphasizing the importance of mapping existing business rules and architecture. The post advocates for incremental modernizati…
A 2.78-trillion-parameter Kimi K3 model ran on a single CPU with 8.24 GB of RAM using a 176 KB portable C99 engine, though it requires a 1.56 TB checkpoint on disk and runs at 26.5 seconds per token at 8 GB, 19.8 at 64 G…