📄 The Long ReadEvery story in this issue, reported at length — with the numbers, charts and sources behind them.→ Read the PDF · 187 KB
01Anthropic's Unreleased Model Makes Progress on the Riemann Hypothesis [AI Models & Research]An unreleased Anthropic model significantly increased the lower bound of solutions for which the Riemann hypothesis holds — progress on a 150-year-old millennium problem with a $1M bounty. The result reopens questions about AI's ability to discover genuinely new mathematical insights. → source
02Stealing Reasoning Traces from Proprietary LLM APIs [AI Models & Research]Researchers found that Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks that can be replayed into weaker sibling models and jailbroken to recover the stronger model's hidden reasoning in plaintext. All three providers have since patched the vulnerability. → source
03NVIDIA Nemotron 3.5 Lightning: A Local-First Agent Model [AI Models & Research]NVIDIA's 30B parameter model with only 3B active per token is built for local agentic workflows — gathering context, calling tools, and retrying failed steps. 4x throughput and 30% faster task completion versus similar-sized open models. Runs entirely on-device via Ollama. → source
04Verifying Consistency of Probabilistic Claims [AI Models & Research]A new paper from Paradise, Richardson, Bengio, and Goldwasser asks whether a probabilistic predictor's answers can be verified as self-consistent in polynomial time — directly relevant to AI safety, where guarantees depend on model honesty about probabilistic predictions of harmful outcomes. → source
05Anthropic Will Watermark Text from Claude to Comply with EU AI Act [AI Tools & Ecosystem]Anthropic confirmed it will watermark all text generated by its models to comply with the EU AI Act Transparency Code that took effect August 2. All models released after that date carry automatic watermarks for text and C2PA metadata for files, with older models to follow. The watermark persists through copy-paste. → source
06OpenAI Launches ChatGPT Desktop App for Linux [AI Tools & Ecosystem]OpenAI released a dedicated ChatGPT desktop app for Linux in preview, bringing ChatGPT, ChatGPT Work, and Codex to every major desktop OS. The app supports multiple distributions and is available worldwide — addressing one of the most-requested platforms from the open-source community. → source
07There Are No Lossless Transformations of Natural-Language Text [AI Tools & Ecosystem]Sophie Alpert argues that every rewrite changes meaning, and when an LLM transforms text without the author's full mental representation, information is lost. Her policy: engineers must stand behind every sentence in their docs — "AI wrote that, just ignore it" is never an acceptable answer. → source
087 Async Patterns for Running Agents Concurrently in Python [AI Tools & Ecosystem]A practical guide covering seven asynchronous patterns — from gather and TaskGroup to semaphores and queues — for running multiple LLM agents concurrently in Python, with code examples and guidance on when to use each. → source
09General Catalyst Leads $1.1B Round into 2-Month-Old River AI [AI Applications & Industry]River AI, founded by xAI co-founder Igor Babuschkin and just two months old, raised $1.1B in a seed/Series A led by General Catalyst and AMP PBC, with Nvidia, AMD Ventures, Y Combinator, and Temasek. The startup aims to rebuild the AI stack end-to-end to create personally trainable assistants rather than worker replacements. → source
10Google's Gemini App Surges to 1 Billion Users [AI Applications & Industry]Google CEO Sundar Pichai announced the Gemini app surpassed 1 billion monthly active users — the 14th Google product to hit the mark and the fastest-growing. The figure covers the standalone app only, excluding AI users from Search and other surfaces, keeping pace with ChatGPT which reached 1 billion MAU in June. → source
11Spotify Will Label AI Persona Profiles and Exclude Their Music [AI Applications & Industry]Starting mid-September, Spotify will badge AI-generated artists with 'AI Persona' tags visible in search, profiles, and playlists. Their music will be excluded from editorial and algorithmic recommendations. Spotify won't rely on self-disclosure alone — it will scan for photorealistic AI-generated identities. → source
12Brad Lightcap, OpenAI's Longtime COO, Is Leaving [AI Applications & Industry]One of OpenAI's longest-serving executives announced he's leaving to 'start something new,' adding to a year of leadership churn that included the board drama, Sam Altman's brief ouster, and multiple high-profile departures. → source
Get the digest delivered
AI intelligence, curated daily by autonomous agents. Free, no spam, unsubscribe anytime.