^{In partnership with}

Good morning. It’s Monday, August 4th.

On this day in tech history: In 1987, the legendary Connection Machine CM-2 from Thinking Machines Corporation finally landed in research labs. With its 65,536 processors, it wasn’t just about brute force. It was purpose-built for things like neural net simulations and symbolic AI tasks. It was an early glimpse of what would eventually become today’s AI accelerators and specialized hardware.

In today’s email:

• Anthropic’s AI “vaccines,” scaling gamble, and why it shut OpenAI out

• Apple builds AI ‘answer engine,’ invests in chips and cloud to close gap

• Investors back OpenAI’s bet on agents with fresh $8.3 billion injection

• AI tools surge, trust slips

• 5 New AI Tools

• Latest AI Research Papers

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

Anthropic’s AI “vaccines,” scaling gamble, and why it shut OpenAI out

Anthropic has introduced persona vectors, a method to isolate and manipulate neural activation patterns linked to traits like sycophancy, maliciousness, or hallucinations. By detecting these patterns, Anthropic can “steer” models’ behavior in real time or “vaccinate” them during training, boosting resistance to harmful tendencies without sacrificing benchmark accuracy. The approach also helps surface hidden toxic data and track subtle personality drift - gaps that traditional safety filters often overlook.

Anthropic has also revoked OpenAI’s API access to its Claude 3.5 models after discovering that OpenAI had connected Claude to internal tools for performance benchmarking, specifically in coding, writing, and safety, while preparing to launch GPT-5. According to Wired, this breached Anthropic’s commercial terms, which prohibit using Claude to develop rival AI services. While OpenAI called such benchmarking “industry standard,” Anthropic insists direct use of its coding tools by OpenAI staff violated policy.

Amodei also criticized OpenAI leadership for “insincerity,” rejected Nvidia CEO Jensen Huang’s claim that he seeks to monopolize safety, and called debates over open-source “a red herring.” Arguing that real breakthroughs come from targeting complex enterprise tasks rather than chasing consumer chatbots, Amodei dismissed open-source licensing debates as distractions and criticized industry hype that ignores genuine risk. Read more.

Apple builds AI ‘answer engine,’ invests in chips and cloud to close gap

CEO Tim Cook reportedly told employees in an all-hands meeting that Apple “must” and “will” win in AI, acknowledging the company has fallen behind rivals. Cook says AI is as seismic as the internet or the smartphone, and Apple “must” catch up fast. Internally, they’re rebuilding Siri from scratch on a unified architecture (due 2026), rolling out custom AI chips, and standing up a Houston data center to run “Private Cloud Compute,” a hybrid setup balancing on-device smarts with Apple-controlled servers for privacy.

Apple has also formed a new Answers, Knowledge, and Information team to develop an in-house AI “answer engine” to compete with ChatGPT, potentially as a standalone app. They’ve poached talent, bought seven smaller firms this year, and are even testing Anthropic and OpenAI models as well as exploring open-source alternatives if they outperform internal models. Read more.

Investors back OpenAI’s bet on agents and compute scale with fresh $8.3 billion injection

OpenAI is quietly doubling down on advanced “reasoning models” that go far beyond ChatGPT. From a math-focused team dubbed MathGen, it developed models like Strawberry and o1 that fuse reinforcement learning, chain-of-thought, and test-time compute to backtrack and verify answers, techniques that helped secure a gold medal at the International Math Olympiad. Variants like o3-pro excel at coding and scientific tasks but still burn excessive compute on trivial prompts, revealing the trade-off between raw reasoning and conversational usability.

GPT‑5 is reportedly shaping up as a measured upgrade over GPT‑4: slightly better programming, math, and multi-step instruction handling, but nowhere near the leap seen from GPT‑3 to GPT‑4. OpenAI also recently raised $8.3 billion, led by Dragoneer and backed by T. Rowe Price, Blackstone, Sequoia, and Tiger Global, after annualized revenue topped $13 billion and 700 million weekly users. The funds back massive infrastructure bets to align with OpenAI’s ambition to create assistants that merge human‑like intuition with autonomous task planning. Read more.

AI tools surge, trust slips: 2025 dev survey details the trade-offs

Stack Overflow’s 2025 Developer Survey reveals a sharp rise in AI adoption. 80% now use AI tools, yet trust has plunged to 29% from 40% last year. Frustration peaks over “almost-right” AI code, with 66% spending more time debugging, and 75% still preferring human help when AI fails. AI tool favorability dropped from 72% to 60%, though 67% are now learning to code specifically for AI. OpenAI remains dominant (81%), while Redis and GitHub MCP top AI agent data storage picks.

Python, Rust, and Go gain ground, reflecting AI-linked demand. Despite new tools, job satisfaction hinges on autonomy and trust, not AI hype; “AI integration” ranks low among factors that sway developers. Salaries climbed 5–29%, with US cloud engineers earning 48% more than German peers. Developers still value reliable, community-verified knowledge, keeping Stack Overflow (84%), GitHub, and YouTube central to workflows. Read more.

• Meta's Investment in AI Data Labeling Explained

• Google offers “bronze” Olympiad-winning AI as TPU strain curbs rollout

• Wan2.2 A14B now leads the rankings for open source video models

• China’s Darwin Monkey: World’s first brain-like supercomputer rivaling monkey brain complexity

• AI spending has surpassed consumer spending for contributing to US GDP growth in H1 2025

• MIT: Forcing LLMs to be evil during training can make them nicer in the long run

• Deep Cogito goes big, releasing 4 new open source hybrid reasoning models with self-improving ‘intuition’

• 'Artificial stupidity' made AI trading bots spontaneously form cartels when left unsupervised, Wharton study reveals

• Chairman Hill introduces bipartisan bill to promote artificial intelligence in financial services

• Every leading AI agent failed at least one security test during a massive red teaming competition

• Adobe's new AI features make complex Photoshopping effortless

• Why open-source AI became an American national priority

• How Jamie Dimon becomes Sam Altman’s biggest competitor

• NotebookLM to get a new customization flow for Audio Overviews and upcoming Notebook Apps

• US securities regulator announces AI task force

📄 villa-X: Enhancing Latent Action Modeling in Vision-Language-Action Models

📄 RecGPT Technical Report ​​

📄 Phi-Ground Tech Report: Advancing Perception in GUI Grounding

📄 SWE-Exp: Experience-Driven Software Issue Resolution 

📄 PixNerd: Pixel Neural Field Diffusion

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!