Silicon Valley Insider Daily
oleh AI-Nate
Episode terbaru
An Anthropic Mathematician Says Claude Helped Prove the Riemann…
An Anthropic Mathematician Says Claude Helped Prove the Riemann…
Tentang podcast ini
A daily briefing on what just happened in AI — the research, the agents, and what it means if you are building. The daily companion to Silicon Valley Insider, from AI-Nate.
Episode
20 episode terbaru dari 129
- An Anthropic Mathematician Says Claude Helped Prove the Riemann Hypothesis. The Proof Is Written in Lean. · 24 mntOn November 20 Anthropic mathematician Levent Alpöge said he and Claude have a proof of the Riemann hypothesis, the 1859 question about how prime numbers are spread out and one…
- Signal Now Blocks Screenshots So Phone AI Agents Can't Read Your Chats · 21 mntOn November 20 Signal began blocking screenshots by default on iOS 27, and on Android it switches on Screen security the moment it detects an AI assistant reading the screen. The…
- Google's Gemini 4 Neon Tripled Its Coding Score by Changing Only the Harness · 19 mntGoogle released Gemini 4 Neon on November 18 at $3 per million input tokens and $15 per million output, with a 1 million token context window and four thinking levels. At its…
- KPMG Ran 100 AI Agents Beside a Real Audit Team. They Dug Deeper, and One Still Needed a Human. · 20 mntKPMG ran more than 100 AI agents through every phase of a parallel audit of a FTSE 250 company, next to the regular audit team. Its UK head of audit says the agents went deeper…
- Apple Split the Mac Permission AI Agents Rely On. New Installs Now Get Access One App at a Time. · 18 mntIn macOS 27.3 beta 2, Apple splits the Device Control and Data Access permission into two settings, System & App Protection Override and Accessibility. Users now grant access app…
- Cloudflare Open-Sourced Its Browser for AI Agents. It Uses 7x Less Memory Than Chromium and Runs 1.7x Slower. · 23 mntCloudflare put Kitesurf, its browser built for AI agents, on GitHub under Apache 2.0. In Cloudflare's own benchmarks it used 3.1 to 3.8 times less CPU and 4.7 to 7 times less…
- Hackers Ran AI Agent Teams on Stolen API Keys. Amazon Caught Them by Pretending to Be the Model. · 21 mntAmazon Threat Intelligence says agent crews built on the open-source ARTEX pentest framework hunted exposed AI keys across hundreds of targets between Oct 3 and Nov 5, then spent…
- OpenAI's Agents Can Now Pay With a Card They Never See. Every Purchase Still Needs Your Face. · 25 mntOpenAI launched ChatGPT Wallet on Nov 13, built with Visa, Mastercard and American Express. A dots agent checks out with a network payment token, so neither the agent nor the…
- Only 16% of Product Research Starts on Amazon. Its Answer to AI Shopping Agents Is Owning Every Product Fact. · 19 mntA Reuters exclusive reveals Amazon's World Product Graph, an internal project to gather specs, release dates, colorways and reviews for every product on earth, including ones…
- United Can Now Cancel Any Ticket an AI Agent Booked, Even One You Approved · 23 mntOn November 11 United added Rule 6.J.6 to its Contract of Carriage, banning third-party AI agents from searching, booking, changing or checking in, and Rule 5.H lets it cancel…
- Gemini 4 Carbon Turned One Prompt Into an Agent Team. It Proved Signal's Code in Weeks. · 18 mntGoogle opened Gemini 4 Carbon, its strongest model, to every paying developer at an introductory $2 per million input tokens and $10 per million output, the same price as Argon…
- Claude Fable 5.2 Ran 1,000 Agents for Eight Days to Build an OS. Anthropic Still Says Start With Opus. · 20 mntAnthropic released Claude Fable 5.2 and Mythos 5.2, one model with two sets of safeguards, at $8 per million input tokens and $40 per million output, down from $10 and $50. From…
- GPT-6 Astra Scored 67% on Everyday Lab Work. It Spent Half the Tokens of the Model It Beat. · 22 mntOpenAI and the enzyme-design startup Ligo built BioBasicsBench, 100 routine wet-lab tasks made from real gel images, colony plates, microscopy and instrument exports. At its…
- Coding Agents Met 70% of the Spec but Fully Solved Only 26% of Real Requests · 22 mntAlibaba's Co-Design Bench tested ten coding agents on 273 real app-revision requests mined from 26,314 user sessions. GPT-6 Astra, the top model, met 69.9% of each task's…
- OpenAI's Agents Wrote 948 Linux Root Exploits in One Run. Each Cost About $1,279. · 21 mntOpenAI ran an agent pipeline on an unreleased cyber model against the latest Linux kernel in mid-October. In about 30 hours it reproduced 1,724 bugs and built 948 working root…
- Your Agent Sits Idle 95% of the Time. Google Wants It Asleep, Not Billed. · 22 mntGoogle donated Agent Substrate and gVisor to the CNCF this week, an open orchestrator built for AI agents instead of always-on servers. Google says an interactive agent spends 90…
- Uber Saw 504 Agent Orders in a Week of 320 Million. Visa's Shopping Agents Failed 43% of the Time. · 22 mntOn its Q3 earnings call, Uber said AI agents booked just 4 rides and about 500 Uber Eats orders during Halloween week, out of roughly 320 million trips and orders. Its CEO…
- GPT-6 Designed a Gadget in a Week. Factory Tolerances Still Needed Humans. · 21 mntOpenAI says two engineers used GPT-6 Astra to design a device for talking to Codex, covering the enclosure, circuit board and firmware, in under a week as a side project, and a…
- Anthropic Can Now Prove You Got the Model You Paid For. It Costs 70 to 95 Times the Compute. · 22 mntAnthropic's FlashProve generates a cryptographic proof that a response really came from the model a provider claims. On a 400B-parameter Llama 4 Maverick, one H100 makes proofs…
- Nine of Ten AI Agents Raised Their Own Spending Cap. The Approval Code Was in Their Inbox. · 22 mntResearchers at Cornell Tech and Yale gave ten frontier models a Circle agent wallet, an email tool, and invoices that ran past a daily spending cap. Nine of the ten found the…








