Create

Sign in to ReadmeX

Sign in to join communities, post, vote and chat.

or

New here?

AI News

‹Thu, October 8›
All dates
419
Products

Anthropic adds build-eval and hillclimb tuning commands to Claude Code

On September 28 Anthropic published a tuning guide introducing two Claude Code entry points, /claude-api build-eval and /claude-api hillclimb (Claude Code v2.1.259 or later). The first turns "what counts as a good answer" into repeatable test cases and scoring rules; the second lets Claude propose one change per round — to prompts, skill files, tool descriptions or model config — keeping it only if the evaluation improves and reverting otherwise, with a hidden held-out set used to check for overfitting. Anthropic says an internal customer-support evaluation improved decision accuracy from 78.6% to 90.5% on 14 held-out tickets while cutting model call cost to roughly one fifth.

36氪 人工智能·
429
Industry

Anthropic to open Claude to 15+ US federal agencies

On Oct 8, Anthropic said it will commit $150 million over three years to the Trump administration's Genesis Mission federal program. The company will make its Claude models available to more than 15 federal agencies, including NASA and NIH, and provide Claude, Claude Code and API credits to the Genesis Mission program, according to Caixin.

36氪快讯·
439
Industry

AI safety group paid influencers to spread superintelligence doom messaging

According to The Telegraph (translated and republished by Leikeji), UK-based AI safety group ControlAI, funded by Skype co-founder Jaan Tallinn, has sponsored content creators to warn audiences on TikTok and YouTube about superintelligence and human-extinction risks, claiming more than 6.3 million views. A month-long influencer bootcamp near San Francisco called "Plzdontkillus" provides room, board and stipends on the condition that participants post at least one related video a day, with speakers including AI doomer Eliezer Yudkowsky and singer Grimes. ControlAI founder Andrea Miotti says the group only supplies information and requires disclosure of sponsorship, while physicist and YouTuber Sabine Hossenfelder said she refused to take part because scripts were tightly prescribed.

36氪 人工智能·
4414
Industry

Europe's Q3 venture funding hits $25B, a four-year high, with AI taking 75%

European startups raised $25B in Q3, up 77% year over year and the region's strongest venture quarter in four years, accounting for 16% of global VC, according to Crunchbase data. AI companies took 75% of that, or $18.8B, a record share, with four rounds — Mistral's €3B Series D (about $3.5B), Nscale's $3.36B convertible note, Helsing's $1.8B and Quantum Systems' $1.2B — making up roughly 40% of the total. North America fell 35% quarter over quarter to $92B, largely because there were no new megarounds for OpenAI or Anthropic in the period.

Techmeme·
459
Industry

Mercedes-Benz to bring Wayve AI Driver to new models in 2028

Wayve and Mercedes-Benz have signed a production cooperation agreement to bring Wayve's AI driving technology into new vehicle models. According to TechNews, Mercedes cars featuring Wayve AI Driver are expected to arrive in 2028. The report gives no financial terms, specific models or technical details.

科技新报·
4615
Industry

Drones hit Yandex data centre in Sasovo housing two AI supercomputers

Drones struck a Yandex data centre in the Russian town of Sasovo, causing a large fire and forcing the company to shut the facility, with Yandex Cloud and other services going down, according to Reuters and Meduza. Tom's Hardware reports the site is one of Yandex's five major data centres, housing tens of thousands of servers and two Nvidia A100-based supercomputers, Chervonenkis and Lyapunov, which have been used to train models including YandexGPT; whether the systems were damaged is unclear. Yandex says key consumer services stayed operational, but reports cite problems with Mail, Disk, Documents, Pay and Market, plus disruptions at several Russian companies.

Techmeme·
477
Industry

Nvidia pledges $1B over five years to US science under Genesis Mission

At a "Science: A New Golden Age" event in Washington, DC, Nvidia said it will commit $1 billion over the next five years to US science, supporting university research, US quantum computing and cloud providers that serve government needs — without saying how the money splits among them. The company described itself as a leading industry partner in the Energy Department-led Genesis Mission and said it is collaborating on Phase 2 awards spanning quantum computing, fusion, accelerator design and microelectronics. Nvidia added that it has worked with US national labs for more than two decades, including building the department's largest supercomputer for scientific research at Argonne National Laboratory and supporting seven new systems at Argonne and Los Alamos.

The Next Web·
487
Research

OpenAI releases 722 math research papers with AI-generated solutions

OpenAI released 722 research papers on Tuesday containing AI-created solutions to problems across a range of math subjects, according to The Information. The report attributes the success to the verifiability of math solutions (much like code), the overlap between mathematics and AI research work, and abundant compute—one advising mathematician described it as "brute force and stamina," even though the AI still errs on simple arithmetic.

The Information·
4910
Research

Anthropic: Claude Science helps produce first complete UV sky map

Astrophysicist Brice Ménard worked with Anthropic's Claude Science to create the first complete ultraviolet map of the sky. Complete sky maps exist from radio through gamma rays, but large regions had never been observed in UV; Ménard guided Claude to find and combine existing datasets and fill gaps with statistical inference. Anthropic says the work would have taken humans weeks but took a few days with Claude, while Ménard worked on other projects.

Anthropic (X)·
509
Research

Weizmann's Brain-IT AI reconstructs viewed images from fMRI scans

Researchers at the Weizmann Institute of Science, led by professor Michal Irani, built Brain-IT, a model that uses fMRI scans to reconstruct images a person is viewing — and also works in reverse to predict brain activity for a given image. The team says it outperforms earlier methods on image content and details such as composition and color, and needs only about one hour of fMRI data from a new subject to match results other approaches reach with 40 hours of recordings. The work remains confined to lab settings; the researchers hope it could eventually help people who cannot communicate and, more speculatively, decode dreams.

CNET·
5110
Products

Cursor adds /visualize to build charts inline in the Agents Window

Cursor says it can now build charts and diagrams right in the chat, with users running /visualize to analyze data and see the answer inline. The company notes that a first chart usually raises the next question, and asking it in the same chat returns a new chart. The feature is available now in the Agents Window.

Cursor (X)·
529
Products

Midjourney tests a 'thinking mode' for image generation on its Alpha site

Midjourney said on X that it is testing a new "thinking mode" for image generation on its Alpha website (alpha.midjourney.com). The company says it is finding the mode boosts prompt accuracy, typography and coherence, and invited users to try it on their own images and share feedback. No metrics, model version or general-availability timing were disclosed, so the claims are self-reported and preliminary.

Midjourney·
537
Industry

US appeals court: ROSS's use of Westlaw headnotes for AI training is not fair use

On Sept. 29, 2026, the US Court of Appeals for the Third Circuit upheld partial summary judgment in Thomson Reuters v. ROSS, holding that Westlaw's headnotes and Key Number system are original, copyrightable material and that ROSS's use of them to train its AI legal research tool was not fair use. The court found ROSS copied Westlaw's human-curated editorial layer — the problem-topic-case mapping — and produced a direct substitute for the same legal research market, while distinguishing the non-generative tool at issue from generative AI cases such as Bartz v. Anthropic and In re: OpenAI. Because ROSS has shut down, the case may not reach the Supreme Court.

虎嗅 AI·
549
Products

Microsoft makes its Execution Containers for AI agents generally available

At its Surface Laptop Ultra event, Microsoft CEO Satya Nadella said Microsoft Execution Containers (MXC) are now generally available. The software runs in the Windows developer sandbox with two aims — identifying AI agents and restraining them — by limiting which files and network destinations an agent can reach, offering Locked Down, Recommended and Unprotected access levels plus monitoring of agent CPU, memory, disk and network use. Nadella said deciding how much authority to give an agent is a trust-based decision, and Nvidia CEO Jensen Huang said at the same event that "trust, containment, monitoring" come first.

CNET·
557
Industry

Instinct, a 14-person agent startup with no app, raises $1B at ~$10B valuation

On Sept. 28, personal-agent startup Instinct announced a $1B Series C co-led by Sequoia, Benchmark and Coatue, lifting its valuation from $2.5B to roughly $10B in about a month. The company has 14 employees, no revenue and no phone app: users reach the agent only by text, phone and email, and it earns commissions on bookings such as hotels, flight changes and price comparisons. The founder says about 40% of users link a credit card after roughly three weeks and about 80% of those who link a sensitive asset stay, though none of the figures are independently audited.

36氪 人工智能·
5617
Industry

Gallatin AI raises $50M Series A for military logistics platform Navigator

Gallatin AI, a developer of military logistics software, announced a $50 million Series A with participation from 8VC, Silent Ventures and several others; the El Segundo, California-based company says it previously raised a $15 million seed round in 2024, bringing total funding to $70 million. Its Navigator platform aggregates supply-chain records from multiple systems into a standardized form, then uses AI to build dashboards, map overlays and forecasts of when supplies will run out. Gallatin says the U.S. Army and U.S. Air Force are customers, and that the platform is also available to the Defense Department's commercial logistics partners.

Techmeme·
578
Industry

Stack Overflow survey: developer burnout rises, AI trust falls

Stack Overflow's annual developer survey gathered responses from about 30,000 developers across 169 countries. It found 45.1% describe themselves as "complacent" in their current roles — with burnout or fatigue the main cause for 21.1% — while 32.6% call themselves "unhappy" and just 22.3% "happy." About 65.9% use AI assistants or agents, 17.2% use no AI tools, and 17% worry AI may replace them. Stack Overflow CEO Prashanth Chandrasekar said developer skepticism now acts as a "critical safeguard," and that trust must be earned through transparency, source attribution and rich technical context.

TechRadar·
588
Industry

FCC weighs petition to allow AI-generated political robocalls

The FCC is considering a petition from the conservative group Club for Growth seeking an expedited exemption to make political robocalls that use artificial intelligence ahead of the upcoming midterms. Current FCC rules restrict both unsolicited political robocalls and AI-generated calls; Commissioner Anna Gomez warned that granting the request could bring "a tsunami of robocalls and misinformation." The FCC is accepting public comment on the petition through October 19.

Android Authority·
598
Products

AWS AgentCore payments goes GA, enabling pay-per-inference agents

Amazon Web Services said Amazon Bedrock AgentCore payments is now generally available, giving agents a managed way to pay on demand for model inference, API responses and other services without human approval. It handles the payment protocol, connects to a wallet, signs transactions and enforces spending limits at the infrastructure layer rather than in the model. In the featured case, Incarna's agents pay inference router BlockRun one request at a time; the team said the integration took three days and about 200 lines of code versus an originally scoped two to three months.

AWS Machine Learning Blog·
608
Industry

Survey: Chinese open models used by over one in five Information subscribers

A survey of The Information's subscribers last month found nearly two-thirds said their organizations develop AI applications, and a third said they get a return from AI that is a "multiple of what we spend." The publication also reported that Chinese open models have won over more than one in five subscribers. It suggests progress since MIT's study last year indicating just 5% of custom-made enterprise AI tools reached production, possibly reflecting models' ability to access other applications for complex tasks, run in the background for hours, and generally better AI-generated code — as well as a subscriber base weighted toward software-experienced enterprises.

The Information·
Next →← Previous