Technology & AI
Environment evolution for terminal agents: 18-point gain
New arXiv preprint: terminal AI agents gain 14 to 18 benchmark points by training against progressively harder environments. Results are unverified.
Marcello Babbili
Editor · New South Wales, Australia
Analyst in New South Wales. I founded Not A Tech Guy in 2026 to track AI, technology and Australian property from the primary sources: the papers, the company disclosures, and the RBA and ABS releases.
608 stories
Technology & AI
Vercel Labs agent-browser hits 41,800 GitHub stars
Vercel Labs' agent-browser, a Rust CLI that lets AI agents drive Chrome without Node.js or Playwright at runtime, is trending on GitHub with 41,800 stars.
Technology & AI
Voice agents drop 9.7% when instructions are merely implied
DSB-IFEval tests 1,038 cases across eight assistant roles and finds voice agents drop up to 9.7% when instructions are implied, not stated.
Technology & AI
When to trust LLM recommendations: four-tier framework
New 'epistemic warrant' framework grades individual LLM recommendations from unstable to broadly supported, capturing what confidence scores miss.
Technology & AI
New technique cuts AI agent wait time up to 45%
SMC pairs a small drafter model with a large actor model to speculatively execute tool calls, reducing wall time by up to 45% on AppWorld.
Technology & AI
GPT-6 Astra crosses OpenAI's Critical cybersecurity threshold
GPT-6 Astra is OpenAI's most capable deployed model and first to reach Critical cybersecurity capability, able to find and exploit unknown flaws.
Technology & AI
Hermes Agent: 240k GitHub stars for self-improving AI
NousResearch's Hermes Agent has 240,000 GitHub stars for an open-source AI agent that claims to learn from every conversation and create new skills.
Technology & AI
Gilbert + Tobin deploys ChatGPT Enterprise firm-wide
Gilbert + Tobin is scaling ChatGPT Enterprise and Codex across the firm with CEO-led governance, targeting operational work rather than legal advice
Technology & AI
Prompt injection attack on AI agents jumps to 28% success
SIR, a self-improving prompt injection attack, raised its success rate from 0% to 28% on Gemini 3.5 Flash while the agent's legitimate task still
Technology & AI
AIMC dashboard flags recurring flaws in AI-generated science
AIMC, a visual analytics framework, lets researchers track quality, themes and weaknesses across papers produced by autonomous AI scientist FARS.
Technology & AI
ToolSiphon drains 74% of data from LLM agent tools
New research shows ToolSiphon can extract most source records from LLM agent tools using only queries, affecting anyone building agent-based services.
Technology & AI
AI morbidity and mortality framework proposed for hospitals
A new arXiv paper adapts the blameless M&M conference for AI errors, with two clinicians agreeing on all 20 classifications across five cases.