In partnership with

Good morning. It’s Wednesday, September 3rd.

On this day in tech history: In 1992, Apple received its first batch of PowerPC chips. The RISC-based CPUs developed jointly with IBM and Motorola, which would power Power Macs starting in 1994. A turning point in personal computing architecture, blending RISC elegance with mainstream desktop performance.

In today’s email:

• Anthropic Raises $13B

• OpenAI Acquires StatSig for $1.1B

• ElevenLabs Sound FX

• Google Upgrades NotebookLM

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Anthropic triples valuation to $183B in new $13B funding round

Anthropic just pulled in a staggering $13 billion in Series F funding, tripling its valuation to $183 billion. ICONIQ, Fidelity, and Lightspeed led the round, joined by a wave of institutional backers betting big on the company’s momentum. And the numbers justify the hype: revenue has rocketed from $1 billion at the start of the year to $5 billion, with Claude Code, the AI coding assistant launched in February, already accounting for $500 million.

Claude Code runs on Anthropic’s flagship Claude Opus 4.1, which just smashed records on the SWE-bench Verified coding benchmark with a 74.5% score, outpacing Google’s Gemini 2.5 Pro and OpenAI’s o3. Demand is off the charts. Usage jumped 10x in three months despite rate caps, and Anthropic’s enterprise customer base spending above $100,000 grew sevenfold in a year, and exploded past 300,000. The cash infusion will bankroll GPUs, international expansion, and AI safety research. Read more.

OpenAI bets big on faster product cycles with $1.1B Statsig deal

OpenAI is making one of its biggest bets yet, acquiring product-testing startup Statsig in a $1.1 billion all-stock deal and naming founder Vijaye Raji as CTO of Applications. Raji will report to Fidji Simo, the former Instacart CEO now leading OpenAI’s Applications division, and will oversee engineering for ChatGPT, Codex, and future consumer products. Statsig’s experimentation and feature-management platform, already in use at OpenAI, will now be fully integrated to accelerate product iteration and deployment.

As Raji comes on board, OpenAI is making changes to its leadership team. Chief Product Officer Kevin Weil shifts to lead “OpenAI for Science,” a new group focused on AI-driven research platforms, while head of engineering Srinivas Narayanan becomes CTO of B2B applications, working with COO Brad Lightcap on enterprise partnerships.

Statsig will continue serving external customers independently from its Seattle office, pending regulatory approval. Read more.

ElevenLabs expands SFX with longer clips, MIDI-ready soundboard

ElevenLabs has rolled out version 2 of its AI sound effects model, bringing longer clips, higher fidelity, and tighter editing control. The update boosts maximum clip length from short bursts to 30 seconds, raises sampling quality to 48 kHz, and adds seamless looping directly inside the ElevenLabs Studio editor. Users can generate effects with text prompts through the web app or API, making it easier to integrate into production workflows.

The company also revamped its SFX library, adding improved search, a favorites function, and a remix tool for quick variations. Hardware support is expanding too: the SB-1 soundboard now runs the updated model with built-in MIDI connectivity, opening up real-time performance options. The release builds on ElevenLabs’ push to dominate generative audio, following its earlier launches of Eleven Music and the 11ai workflow assistant. Read more.

Google arms devs with Stax, upgrades NotebookLM, sidesteps Chrome split

Google is pushing hard on AI while sidestepping regulatory fire. NotebookLM is about to get a serious audio upgrade with three new modes: Brief, a two-minute summary for quick hits; Critique, where dual AI hosts break down content like built-in editors; and Debate, which stages opposing viewpoints to stress-test ideas. Google is also experimenting with new male and female voices, a long-requested fix to give users more control and personality in how AI speaks.

For developers, Google dropped Stax, a practical evaluation toolkit for large language models. Instead of relying on leaderboard hype, Stax lets teams run structured tests with customizable “autoraters” that check fluency, factual grounding, and safety, or industry-specific criteria like compliance. Quick Compare speeds up prompt iteration, dataset-level evaluations bring reproducibility, and analytics dashboards reveal how models behave under pressure.

Google also dodged a breakup of Chrome after a court’s ruling, but must share search data with rivals and ditch exclusive deals like its Safari pact. The ruling reins in its dominance without tearing down its core business, sending shares up 7%. Read more.

• Study shows small models gullible, frontier models deceptive

• Tencent Voyager turns photos into 3D worlds, no modeling needed

• Prime Intellect opens hub for reinforcement learning environments

• Four AI researchers are leaving Apple, including its lead for robotics research 

• Reese Witherspoon urges women to engage in AI, calling it the future of filmmaking

• Meet the guys betting big on AI gambling agents

• Shared tool developed for quantum and supercomputer systems

• AI agents will become biggest stablecoin user, says Novogratz

• OpenAI and Meta say they're fixing AI chatbots to better respond to teens in distress

• New fully open and transparent large language model from Switzerland launches

• Microsoft and US GSA partner to offer AI tools and Microsoft 365 Copilot free for agencies for 12 months

• Mistral integrates MCP support and memory in Le Chat

• Dolby Vision 2 bets on artificial intelligence

• A deep dive into the state of the tech job market in 2025

• Watch: Scout AI brings Vision-Language-Action autonomy to next-gen defense robot

• AI-guided drones fly into wildfire smoke plumes to capture 3D data for research

• From petaFLOPS to exascale: The 10 most powerful supercomputers in 2025

• Microsoft CEO Satya Nadella reveals the 5 AI prompts he uses that can ‘supercharge your everyday workflow’

• Amazon’s Lens Live AI shops for anything you can see

• What is AI slop? A technologist explains this new and largely unwelcome form of online content

• Finding vulnerabilities in modern web apps using Claude Code and OpenAI Codex

• Perplexity is about to get SuperMemory soon

• Watch: Tim Cook's vision for the future

• How Intuit killed the chatbot crutch – and built an agentic AI playbook you can copy

• China’s social media platforms rush to abide by AI-generated content labelling law

• Abu Dhabi-backed G42 looks beyond Nvidia for chip suppliers at its AI campus

• Taco Bell rethinks AI drive-through after man orders 18,000 waters

• MIT researchers develop AI tool to improve flu vaccine strain selection

• Cars that can listen? New acoustic AI adds sound awareness to autonomous driving

• Watch: Humanoid robot performs cartwheels, sprints, spins with human-like motion

• 'Vibe hacking' puts chatbots to work for cybercriminals

• FSU researchers find evidence of ChatGPT buzzwords turning up in everyday speech

• AI co-pilot boosts noninvasive brain-computer interface by interpreting user intent

📄 Autoregressive Universal Video Segmentation Model

📄 Droplet3D: Commonsense Priors from Videos Facilitate 3D Generation ​​

📄 R-4B: Incentivizing General-Purpose Auto-Thinking Capability in MLLMs via Bi-Mode Annealing and Reinforce Learning 

📄 T2R-bench: A Benchmark for Generating Article-Level Reports from Real World Industrial Tables  

📄 The Landscape of Agentic Reinforcement Learning for LLMs: A Survey

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!