Knowledge that moves
with the pace of AI.

What Multi-Turn Reinforcement Learning Actually Improves in LLM Agents ── A preprint reports that gains depend on the turn budget and come mainly from error recovery, not better single steps

Removing Specific Data from a Trained Model, with Exact Guarantees Where Possible ── A preprint proposes IDSISA, which groups likely-to-be-deleted data into few shards and recovers accuracy by distillation

Building Practice Environments for Tool-Using Agents from Real Call Records ── A preprint proposes ToolSpace, which reverses environment construction so that tools come first

Do the Gains of Self-Improving Agents Hold Up Outside Training Conditions? ── A preprint proposing SelfImprove-Bench, which grades how far improvements transfer on a ladder from G0 to G4

The Task Passed, but Unrelated Files Are Gone ── A preprint proposing VeriHarm, which measures the "harmful success" of coding agents with executable tests

Training Computer-Use Agents on Verified Experience ── NorthCUA's technical report (a preprint) folds task execution, recording, verification and training into one loop

Can a causal graph explain why a reinforcement learning agent acted? ── A preprint proposing CausalXRL, which learns the causal structure behind an agent's decisions and returns explanations at more than one level

Where do AI agents that act before being asked go wrong? ── A preprint introducing HALO, in which experts graded proactive agents' responses in detail and found that premature and excessive interventions caused most rejections

Following the US midterm elections (Democratic Party's retaking of the House of Representatives), AI and data…

ensen Huang decreased. OpenAI is moving to ``recover the favorability'' of its data center policies, and Google and Block are continuing and expanding workforce…

The Democratic Party gained the upper hand in the US mid-term elections, creating a sense of uncertainty over…

At the same time, employment adjustments due to AI are prominent, such as Google's expansion of voluntary retirement and Canva's large-scale restructuring. Anth…

There has been a lot of coverage surrounding Anthropic's IPO (inference costs, third-quarter results, and poli…

scrutiny. At the same time, reports of AI-induced restructuring at Barclays and Goldman Sachs, as well as copyright risks in the EU judiciary (Google), have com…

Anthropic's amended IPO statement is the biggest source.

Revenue for the third quarter reached approximately $17 billion, while losses were approximately $23 billion, and AI-related spending commitments are reported t…

Anthropic's IPO (reportedly valued at $2 trillion) was the biggest topic in the market, and DeepSeek's announc…

on for coding agents. The picture is clearly changing, with AI boosting Meta's ad revenue while encroaching on Google search. On the employment front, companies…

Doubts about the sustainability of AI investment suddenly surfaced.

OpenAI's funding goal was lowered by about $10 billion, and Alphabet reportedly lost about $500 billion in market capitalization in one day. Meta has a cash flo…

- Anthropic's IPO prospectus (S-1) has been released, and the industry is focused on debate over risk descript…

- On the other hand, concerns about a bubble are increasing, such as ``plummeting tech stocks'', ``AI panic'', and ``questions about profitability ($1trn questi…

Anthropic's IPO has been heavily reported in multiple media outlets, with reports stating that the company's v…

rtunity.'' Google was reported to have topped the benchmark with Gemini 4 Carbon, but revenue from AI search fell short of expectations, and Alphabet's market c…