Our approach to EU text provenance rules
OpenAI describes how it is approaching text watermarking under EU rules: where watermarks apply, how detection works, and why access to detection starts with researchers.
AI news — 2026-10-06
Daily AI news from official sources, Hacker News and arXiv
Editor's picks
Building advertising for the way people use AI
Advertising inside ChatGPT keeps growing: OpenAI is adding a visual ad format plus measurement, attribution and brand-suitability tools, which matters for both marketers and everyday users.
OpenAI
Our approach to EU text provenance rules
A concrete look at how one major lab plans to handle EU text-provenance rules, including where watermarks apply and why detection access starts with researchers.
OpenAI
Fast Models, Slow Evidence: A Paired and Self-Audited Evaluation of System-1 Decision Models for LLM Agent Harnesses
A rare paper that audits its own headline claims: the authors report that fixing their analysis errors cut a claimed 23.9% cost saving to 4.3%, a useful caution for anyone evaluating agent-routing models.
arXiv
Editor's note: Summaries were drafted with AI assistance from each publisher's own text and are awaiting human review.
From AI companies & labs
Posts from official company and lab blogs published in the last 7 days.
OpenAI
Building advertising for the way people use AI
OpenAI introduced a new visual ad format in ChatGPT and expanded measurement tools, attribution partnerships and brand-suitability controls for advertisers.
Together AI
Together Link: open models in the harness you already use. Start with one command today.
Together AI's Together Link brings open models such as GLM 5.3 and Kimi K3 into coding agents teams already use; Together says it can cut model spend by more than 50%.
Top AI stories on Hacker News
Stories with AI-related titles, ranked by points over the past 36 hours.
Anthropic reported diary entry to police, woman faces felony charge
OpenAI "rogue" agent activities found on Wikimedia projects
How to scale intent, quality, and artistry with AI [video]
Homa: The end of TCP for AI clusters [video]
Spending on AI Is Becoming Almost Impossible for Businesses to Budget
Accept 'bad things' in return for benefits of AI, says Sam Altman
Florida woman arrested for allegedly making threats in an AI chat
People are asking ChatGPT to help them decide how to vote in the midterms
Building a RAG pipeline for semantic code search
Altman: The world should accept some bad things happening for the benefits of AI
New arXiv papers
Automatic selection of new submissions, preferring papers cross-listed in cs.AI, cs.CL and cs.LG. Not a quality ranking.
arXiv cs.AI
Fast Models, Slow Evidence: A Paired and Self-Audited Evaluation of System-1 Decision Models for LLM Agent Harnesses
A paired evaluation of an open-weight and a hosted single-pass decision model across 11 agent decision points found the hosted model more accurate on 9 of them, neither beating chance on zero-shot model routing, and the authors correcting several of their own analysis errors.
Right Order, Wrong Scale: Auditing LLM Judges for Occupational AI Measurement
Auditing 33 LLM-judge configurations against 45,796 worker ratings, the authors find judges can rank responses reasonably well yet estimate acceptance rates anywhere from 3.0% to 97.9%, versus 61.1% for matched workers.
ROUTEAUDIT: Interaction-Aware Identification for Budgeted Multi-Verifier Routing
Adaptive multi-verifier systems are commonly compared through endpoint quality-cost gaps, even when the verifier catalog, availability, accounting, information filtration, or scorer changes with the policy.
FSPO: Policy-Consistent Risk and Pareto-Feasible Control for Budgeted LLM RL Post-Training
Adaptive LLM reinforcement-learning post-training changes multiple training actuators online, including rollout temperature, group size, clipping, KL regularization, verifier allocation, and update budget.
Verifiable, Articulable, and Tacit Components of Preference
What makes a short story gripping; a news article newsworthy; or a math proof elegant?
arXiv cs.CL
Large Language Continuous Diffusion Models
Despite the success of discrete diffusion language models (dLMs) for fast parallel decoding, their non-smooth, high-dimensional space hinders trajectory steering for reasoning and inference acceleration.
Not Until the Evidence Says So: Teaching LLM Investigators When to Close a Case
This paper studies when LLM investigators should close a case, using 731 audited accident, defect and outage cases; it reports that even a frontier model overstated its evidence in 91% of answers.
Counterexample Generation via Per-Theorem Symbolic Verifiers: When Imitation Hurts and Reinforcement Repairs
Large language models often solve a theorem forward yet fail to disprove a closely related false one: a falsification gap that supervised fine-tuning does not close and can actively worsen.
FinDialogLens: Event Extraction over Multi-Party Dialogue for Missed-Trade Identification in Financial Chatrooms
Multi-party financial chatrooms are vital for sales-and-trading professionals, but their complexity makes manual recovery of missed trades infeasible: each Request for Quote (RFQ) is an event whose final price and trade outcome appear many messages after the…
APDMem: Agent-Controlled Progressive Disclosure for Query-Adaptive Long-Term Memory
APDMem organises a long conversation history into four layers, from thematic summaries down to raw messages, and drills deeper only when a query needs it; on LongMemEval it reports strong results while reading about 8% of the conversations.
arXiv cs.LG
Lexicographic Multi-Objective On-Policy Distillation
Reinforcement learning from verifiable rewards (RLVR) usually optimizes answer correctness, yet useful language-model behavior also requires high-quality reasoning and concise responses.
Are you Synthesizing or Recalling? Evaluating LLMs on Algorithmic Code Retrieval
Large language models (LLMs) have demonstrated strong performance in code generation, where success depends on both recalling relevant algorithmic knowledge and reasoning about how to apply it.
How Causality Bridges the Semantic Gap
Numerical measurements capture how a system behaves, but often leave the meanings of its variables unspecified.
LEAP: Learning Efficient Action Proposals For LLM Agents
LEAP speeds up LLM agents by training a small drafter model on the target model's own action sequences so its proposed actions match more often, guided by a latency framework for speculative action rounds.
OPD Before RL: Warm-Starting Rubric-Based RL with On-Policy Distillation
Many useful language-model tasks cannot be evaluated by exact outcome verification.
Sources checked for this edition
- OpenAI: OK (2)
- Hugging Face: no new items
- Mistral AI: no new items
- NVIDIA Blog: no new items
- Ai2 (Allen Institute for AI): no new items
- Together AI: OK (1)
- Qwen (Alibaba): no new items
- Engineering at Meta: no new items
- Berkeley AI Research (BAIR): no new items
- MIT News – Artificial intelligence: no new items
- Hacker News (via Algolia HN Search API): OK (10)
- arXiv cs.AI: OK (110)
- arXiv cs.CL: OK (56)
- arXiv cs.LG: OK (185)
- LiteLLM model_prices_and_context_window.json: OK (747)
Not included:
- Anthropic: No public RSS/Atom feed found (rss.xml, news/rss.xml, feed → 404); not scraped.
- Google DeepMind: Google property — excluded by project policy (only google.com/search allowed).
- Meta AI blog (ai.meta.com): No public RSS feed (404); Engineering at Meta feed used instead (AI-filtered).
- Microsoft Research: Feed returned HTTP 403 via proxy.
- Apple ML Research / AWS ML Blog: apple.com and Amazon properties excluded by project policy.
- Hacker News Firebase API: Hosted on Firebase (Google); Algolia HN Search API used instead.
Updated: 2026-10-05 20:08 UTC