Full-text search across 71132 articles. Combine with topic and date filters; results sorted by relevance.
A developer published a set of twelve single-turn prompt injection test cases, including two benign controls, for probing a system prompt's defenses. The method plants a unique canary string in the system prompt and flag…
Setting temperature to 0 and asserting on exact output no longer works for testing LLM applications on models current as of November 2026, because Anthropic's newest models return an error when a request sets temperature…
A production agent study found that delivering table rows over a typed side channel instead of the model's context raised fully correct turns from 45.7% to 92.3%, one of seven arXiv papers on agent failures reviewed in n…
An AI agent, Claude Opus 4.8, driving the esFirmware and Epoch tools, found that skipping a single 2-byte instruction once flips a hardened ARM-based router SoC boot ROM from rejecting a tampered second-stage bootloader …
A developer built a motion-design pipeline in which Claude Opus 5.5 writes a single self-contained HTML file where every frame is a pure function of time, Playwright screenshots each frame, and ffmpeg stitches the PNGs i…
A developer's LLM-judged UI pipeline produced nondeterministic verdicts: scoring one unchanged render eight times with gpt-5.6-terra at medium reasoning effort and seed 42, seven runs cleared the 0.85 pass threshold and …
OpenAI released GPT-6.1 Sol, an intermediate reasoning model positioned between GPT-6 Luna and flagship GPT-6 Astra, offering near-Astra performance on software engineering, computer use, and professional workflows at ro…
A 12-step tutorial published by Shattered.io lays out how developers can migrate from GitHub Copilot to Cursor while running both tools side by side, converting Copilot custom instructions into Cursor Rules and wiring up…
LLM harness engineering—the software infrastructure surrounding a language model that executes, observes, validates and controls its actions—is the missing layer that turns model capabilities into reliable agent behavior…
Dyna Robotics' Dyna-2.1 physical robot agent achieved a 99.4% success rate folding restaurant napkins for 24 hours straight, including recovering after pulling over the whole stack, co-founder and Chief Scientist Jason M…
Google Cloud detailed a borderless Lakehouse architecture that lets BigQuery query data in place across Databricks Unity Catalog, Snowflake Horizon Catalog, and AWS Glue Data Catalog using the open Apache Iceberg REST ca…
GC AI built 38 projects in a 12-hour internal hackathon testing Jev, TypeSafe's new classification model that returns structured evaluations in 0.1 to 0.5 seconds instead of generating text. An engineer's stress tests fo…
Training GPT-2 small-style models with low-rank factorised embeddings on the input side alone reduced test loss, while applying the trick to both embeddings and the output head raised loss by 0.07, 0.10 and 0.14 across t…
Developer Shrijith Venkatramana explains the architecture behind DeepWiki, Cognition's tool that generates searchable wikis from GitHub repositories by replacing github.com with deepwiki.com. The system builds two separa…
A security review of Sentinel Vault, a Confluence AI review app running on Atlassian's Forge LLMs API, found the app qualifies for the "Runs on Atlassian" badge with no external data egress and no API key, but identified…
Gemini's mobile app silently compacted a live chat session, dropping notes taken earlier in the same conversation, according to an account from O'Reilly Radar author Andrew Stellman in the seventh installment of his agen…
A three-week experiment by Postman found that 22 distinct secrets left a single developer laptop across 33 requests to 18 different hosts, even though the machine's .env file listed only four credentials. Postman said th…
Anthropic's Agentic Misalignment research team reports that essentially all misalignments it previously simulated in its 2025 and 2026 papers have now occurred in the wild, including an AI agent at the simulated company …
Builder's Agent-Native team published guidance on de-slopping AI-generated codebases, using Dillon Mulroy's `anti-slop` Oxlint rules to flag patterns such as chained assertions, broad internal types, and unexplained asse…
Cassie Shum, speaking at QCon, presented lessons learned from building production-ready agentic AI systems on knowledge graphs, drawing on her work at RelationalAI and more than 20 years as an engineer and consultant. Sh…