Gemini 2.5 Flash

Headlines · 4

1. 11h
   AI-Enhanced Infostealer Uses LLMs to Profile Victims

   Cisco Talos researchers identified Volta-Autodoxer, an infostealer that collects credentials, authentication tokens, cookies, payment data, crypto wallet seed phrases, and active sessions. It compiles stolen information into a dossier and sends it through OpenRouter to language models for structured, citation-backed victim assessments, defaulting to Claude Opus 4.6 with fallbacks to Kimi K2 and Gemini 2.5 Flash. The researchers found the malware with CAIRN, a framework for detecting hacking tools that incorporate AI capabilities.

   Hacker News 首页（100+ 分） · 🔥 24
2. 3d
   Anthropic adds abuse protections for Claude as research questions rudeness

   Anthropic updated Claude’s usage policy on October 8 to prohibit sustained and unnecessary abuse of AI, according to 36Kr. The company says Claude may end a conversation when a user treats it as an outlet for abuse, while ordinary anger, questioning and model testing remain allowed. The accompanying review cites multiple studies in which politeness, rudeness and threats produced inconsistent results across models and tasks. It concludes that abusive prompting is not a reliable way to improve AI performance.

   36氪 人工智能 · 🔥 0
3. 4d
   Hill-Climbing Agent Quality with Agents CLI Evaluation

   Google Developers Blog walks through an Agents CLI evaluation loop for AI agents, covering test-case generation, grading, run comparison, failure analysis, prompt optimization, regression tracking, model-backend experiments, and automated alerting. Using a LangChain template, the tutorial demonstrates how per-case results can reveal regressions and model differences that summary scores alone obscure. It also argues that evaluation helps only when teams act on the findings.

   Google Developers Blog · 🔥 0
4. Oct 15
   Apple study compares plan-first vs retrieve-first tool discovery for LLM agents

   Apple Machine Learning Research published an EMNLP paper by Mingfei Zhang and co-authors comparing two ways tool-using LLM agents find APIs before planning: Retrieve-then-Plan (RtP), which embeds the query and retrieves similar tools, and Plan-then-Retrieve (PtR), in which the model first generates abstract tool stubs that become retrieval queries. On ToolBench's 46,984 API endpoints across 50 categories, RtP beat PtR on single-tool queries while PtR beat RtP on multi-tool queries; the crossover held at retrieval budgets k=1–20, registry sizes from 1k to 47k APIs, and with Gemini 2.5 Flash and gpt-oss-120b. The authors also describe a stub-hallucination failure mode, note that PtR's extra discovery call roughly doubles inference cost, and suggest complexity-adaptive routing using RtP's top-1 retrieval score.

   Apple Machine Learning Research · 🔥 0

Experience and discussion from the community

Share my Gemini 2.5 Flash experienceAsk about Gemini 2.5 Flash