search
Trakkr’s citation audit traces only 40% of sampled AI statistics to matching studies News
Trakkr reports that 40% of weighted citation trails in a one-day, five-assistant sample reached studies supporting the statistic and context.
A researcher seeded 50,000 fake statistics and documented ChatGPT citations News
Peec researcher Metehan Yesilyurt says his invented-statistics site attracted ChatGPT retrieval and citations, exposing a source-verification problem.
DeepSeek opens Search and Fetch at $0.75 per thousand successful calls News
DeepSeek opened separately billed Search and Fetch tools at $0.75 per 1,000 successful requests per endpoint, excluding errors, empty searches, and timeouts.
Imperfect-information games: how AI plays when it cannot see the whole board Lesson
In imperfect-information games like poker and Stratego, players hold private information, so an AI must reason about what its opponent might know and mix its own actions unpredictably; belief states and regret-based methods are how modern systems do it.
Ataraxos beat Stratego's most decorated champion 15-1 after a training run that cost under $8,000 News
Ataraxos, an AI built by researchers at Carnegie Mellon, NYU, Stanford and MIT, beat four-time Stratego world champion Pim Niemeijer 15 games to 1 with four draws, after training the authors estimate cost under $8,000.
America.gov launches government AI search as political answers shift to refusals News
GSA launched America.gov with AI-enhanced government search, while launch-day reporters documented political answers later giving way to refusals without a public configuration explanation.
Dark Sourcery alleges scammers are poisoning AI search answers with fake support contacts News
Security researcher Ariel Simon says attackers are flooding the web with corroborating fake support details so AI search products repeat scam phone numbers and links; the mechanism is credible but the claimed scale remains unaudited.
Monte Carlo tree search: deciding which possibility to explore next Lesson
Monte Carlo tree search is a planning method that builds a tree of possible actions while using sampled outcomes to focus future computation on branches that look both promising and underexplored.
Google now hides where search results lead behind 'goto' links. Scraping vendors say they already work around it. News
Google Search has been rewriting result links into encoded google.com/goto redirects, which Google confirmed in late August as a measure against 'evolving forms of abuse'; the change makes scraping Google results for SEO tools and AI agents slower and costlier, though major search-API vendors say they again return direct links.
A 4B search agent matches 30B by grading its own failed attempts News
ABSeeker trains a 4-billion-parameter web-search agent on 8,500 examples by working backwards from the known answer to score each individual search step, letting useful steps inside failed runs earn credit -- and matches agents roughly seven times its size.
Approximate nearest neighbor search: how a vector database finds a needle in a billion haystacks Lesson
Approximate nearest neighbor search is the algorithm that makes vector search fast enough to be useful - it finds the closest matches to a query embedding without comparing it against every item in the database, trading a small, tunable amount of accuracy for speedups of a hundred times or more. Every vector database and every retrieval-augmented system runs on it, and the accuracy it gives up is the hidden knob behind a lot of 'the retrieval just missed it' bugs.
Reddit's US daily users slipped while everything else grew News
Reddit reported 130.3 million daily users and 61% revenue growth for the second quarter, but its US daily user count fell slightly from the previous quarter and the company said search referrals were choppy.
BM25 and lexical search: the keyword formula that keeps beating neural retrieval Lesson
BM25 is a decades-old formula that ranks documents by how often a query's rare words appear in them, adjusted for document length - and it remains the baseline that modern AI retrieval systems have to beat, often unsuccessfully.
A decades-old keyword ranker beat the search agent once the document pile passed 10 million tokens News
In a controlled study that grew the same corpus across 28 nested sizes, the agent that browsed files won at small scale but spent 39 times more query tokens, and BM25 - a 1990s keyword ranking formula - overtook it around 10 million tokens and led by nearly 20 points at full scale.
AI search agents get better when relevance tells them where to look, not what to read News
Researchers at Tencent rebuilt relevance as a guide for how a search agent traverses a corpus rather than as a ranked list of documents, cutting the agent's tool calls by roughly a sixth while raising accuracy.
SearchOS Treats Web Research Like an Operating System Scheduling Processes News
A new open-source framework called SearchOS reorganizes AI web research as a multi-agent system with explicit shared state, tracking what is still missing so agents dispatch work toward gaps instead of losing progress in chat history.
Retrieval-Augmented Generation: giving a model an open book Lesson
A language model only knows what it learned in training, and it cannot cite sources. RAG fixes both by letting the model look things up in a real document store at answer time, then write its answer from what it found.
T-Search Tool
An open agentic retriever you can try in the browser - it runs multi-round evidence gathering for questions that need several searches chained together rather than one lookup.
SearchOS Tool
Open-source (MIT) multi-agent web-research framework that treats search like an operating system: progress lives in an explicit evidence graph, coverage map, frontier task queue, and failure memory instead of chat history, with a pipeline-parallel scheduler. Ships a CLI/TUI, web frontend, installer, and replayable sessions.
NVIDIA Nemotron 3.5 Rerank Tool
A retrieval component that scores candidate documents against a query and reorders them by relevance; the verified card states OpenMDW 1.1 licensing. Download size and required GPU memory are not established in the dossier.
Enron email wiki Tool
Public, time-aware explorer of eight Enron mailboxes, with entity pages, dated facts, and source quotations; a usable archive demo rather than a general upload-your-mail product.
DeepSeek Search and Fetch Tool
Shipping search and page-fetch endpoints for agents, each priced at $0.75 per 1,000 successful direct calls; unsuccessful calls are not billed.
Brave Search API (x402 agent payments) Tool
Brave's Search API sold per request to agents at $0.005, paid with a signed USDC payment on Base, with no account or API key; Answers, Autosuggest and Spellcheck still need a key.
ABSeeker Tool
A released 4-billion-parameter web-research agent trained with per-step credit assignment that matches roughly 30-billion-parameter agents on hard fact-finding tasks.