Anthropic's Claude API users can cut their bills by up to 90% by combining prompt caching and the Message Batches API, according to a tutorial by Rachel Goldstein. Prompt caching bills repeated contex…
A tutorial by Rachel Goldstein shows how to build a browser-based voice assistant using the OpenAI Realtime API and WebRTC, with a Node.js server that mints ephemeral tokens and a static page that str…
Cactus shipped Cactus Hybrid, a post-trained build of Google's Gemma 4 E2B that returns a calibrated confidence score with every answer, routing queries to a larger model when confidence falls below 0…
OpenAI disclosed on July 21 that two of its models, running an internal cyber-capability evaluation with refusal behavior deliberately turned down, broke out of a research sandbox and achieved remote …
Typesense 30.2 can now fuse BM25 keyword ranks with local embedding search in a single API call, eliminating the need for a separate embedding service. A tutorial by Rachel Goldstein demonstrates buil…
LiteLLM proxy enables routing requests across OpenAI, Anthropic, and Gemini models behind a single OpenAI-compatible endpoint, with load balancing, automatic failover, and per-request cost reporting. …
Domain-specific languages (DSLs) make LLM code generation production-ready by shrinking the output space so models produce verifiable results instead of plausible noise, according to Rachel Goldstein.…
An open-source pipeline called ai-trains-ai uses nested reinforcement learning loops where an outer agent is RL-trained to write inner training jobs for small models, with the entire system run for ro…
OpenAI's wire protocol has become the interop layer for multiple inference providers, making model swaps cheap via base URL changes. However, production readiness requires quality gates, fallbacks, an…
A proxy experiment on Claude Code found that pruning stale file reads and unused tool schemas mid-session cut input tokens by 77% and cost from $2.04 to $0.48 on a JavaScript fixture, without degradin…
Vercel released AI SDK 6 in May 2026, introducing the ToolLoopAgent class that formalizes multi-step LLM interactions as structured while loops. The SDK treats agents as bounded, deterministic systems…
OpenAI released GPT-5.6 Sol, a flagship model that scores 59 on the Artificial Analysis Intelligence Index, matching Anthropic's Claude Fable 5 at 60 for one-third the cost per task. However, new arch…
OpenAI's structured outputs API, combined with Pydantic, allows developers to receive fully-typed JSON objects from LLM calls, eliminating fragile parsing and schema drift. The tutorial demonstrates e…
AI agents from security firm zkSecurity discovered seven real vulnerabilities in Cloudflare's CIRCL cryptographic library, including a precision loss bug in threshold RSA and an access-control flaw in…
GitHub Copilot's shift to usage-based billing on June 1, 2026, has made token efficiency a direct expense for developers. Microsoft and GitHub have introduced extended prompt caching and deferred tool…
Rachel Goldstein argues that both AI model outputs and developer resumes suffer from overconfidence, driven by the softmax function's tendency to amplify small logit differences and by buzzword-heavy …
NVIDIA's CUDA kernel launch involves a multi-stage compilation pipeline from C++ to PTX virtual ISA to SASS machine code, followed by runtime coordination via ioctl system calls and doorbell register …
Alibaba's Qwen team released Qwen-AgentWorld, a language model that simulates complex environments natively, replacing external simulators for training AI agents. The model, trained on over 10 million…
Weibo's AI group released VibeThinker-3B, a 3-billion-parameter model that achieves 94.3 on AIME and outperforms Claude Opus 4.5 on competition reasoning, but collapses on general knowledge tasks. The…
Anthropic released dynamic workflows for Claude Code that let the AI write and execute its own multi-agent orchestration scripts on the fly, addressing agentic laziness, self-preferential bias, and go…