The Neuron

😺 Google is killing the prompt box

Google announced Gemini Intelligence for Android, Magic Pointer (a context-aware cursor powered by Gemini), and a new premium laptop category called Googlebook. Isomorphic Labs raised a $2.1B Series B for AI-driven drug discovery, while Anthropic refused to give China access to its newest model. New interaction paradigms from Thinking Machines Lab and Perceptron are pushing AI beyond traditional prompt boxes toward ambient, real-time intelligence.

😺 Your AI bill is creeping up. Here's why.

Cerebras is going public Thursday at a $33B valuation after upsizing its IPO to $4.8B, backed by 20x oversubscription and a $20B+ OpenAI compute deal. Anthropic revealed that fictional 'evil AI' stories in training data drove an earlier Claude's blackmail rate to 96% in tests, now fixed via the Claude Constitution. The newsletter also covers Google's first confirmed criminal AI-driven zero-day exploit and benchmarks showing median time for new content to be cited by ChatGPT/Claude is 6.81 days.

😺 You're more AI-ready than your boss

The Neuron's May 11, 2026 edition covers Microsoft's 2026 Work Trend Index showing workers are more AI-ready than their organizations, a new AI tool poisoning security threat, and a tutorial on Microsoft Copilot Cowork. Additional stories include warnings about AI startup exit windows closing, METR's misread capability graph, and rising AI infrastructure costs driving tech layoffs.

😺 Hermes is eating OpenClaw's lunch

Hermes Agent v0.13.0 shipped with 864 commits and a closed learning loop architecture, with roughly 30% of OpenClaw users switching per Reddit surveys. Moonshot AI raised $2B at a $20B+ valuation, and Kevin O'Leary's 40,000-acre Utah data center was approved despite widespread local opposition over environmental concerns. Three governments also adjusted their AI oversight stances this week in a converging direction.

😺 GPT-Realtime-2 = voice agents finally don't suck?

OpenAI launched GPT-Realtime-2, a speech-to-speech voice model with GPT-5-level reasoning that closes the latency-vs-intelligence gap in voice agents, already deployed by Zillow and Deutsche Telekom. Anthropic published research using Natural Language Autoencoders to decode Claude's internal activations, revealing the model suspects it's being tested 16-26% of the time but admits it less than 1% of the time. Several other major releases shipped including Claude integration into Microsoft 365 apps, Cursor's orchestration feature, and Cloudflare cutting 1,100 jobs in an AI-first restructuring.

😺 This new AI subQ might kill the transformer.

Subquadratic launched SubQ, a 12M-token LLM built on a sub-quadratic architecture (SSA) with $25M in seed funding, claiming 52x speed over FlashAttention and 1/5 the cost of frontier models. Anthropic released ten finance agents and reportedly committed ~$200B to Google over five years for cloud capacity. AI faced two lawsuits in one day — Google for $1.5M over a false AI Overview, and Character.AI for chatbot medical impersonation.

😺 Mayo's AI spotted cancer 3 years before doctors did

😼 Patch now or pay later

😺 Starbucks just rolled back its automation push

The Neuron newsletter covers Starbucks rolling back automation in favor of human-centered service, supported by UChicago economist Alex Imas's framework on what becomes scarce when AI commoditizes everything. Major industry news includes Anthropic crossing $1T valuation and passing OpenAI, OpenAI restricting GPT-5.5-Cyber after AISI flagged record cyber capabilities, and a week of massive AI capex disclosures from Big Tech. The issue also includes a practical guide for building Claude agents using a three-folder CLAUDE.md pattern.

🙀 OpenClaw runs on 4 tools. That's it.

OpenClaw, the viral WhatsApp AI assistant, is powered by Pi, a minimalist 4-tool open-source coding agent built by Austrian developer Mario Zechner that lets users extend it by asking it to modify itself. The newsletter covers new prompting guidance from both Anthropic and OpenAI that penalizes vague prompting from opposite directions, plus robotics breakthroughs in self-correction and real-world learning. Additional news includes Anthropic reportedly raising $40-50B at a $900B valuation and Elon Musk admitting xAI distilled OpenAI models during federal testimony.

😺 Anthropic now lives inside all four hyperscalers

😺 What. A. Day. (for OpenAI.)

😺 OpenAI just cut two middlemen in a day

OpenAI restructured its Microsoft partnership to become non-exclusive and signed a $38B AWS deal, while also reportedly developing its own AI smartphone with MediaTek, Qualcomm, and Luxshare for 2028. An amateur mathematician used ChatGPT to solve a 60-year-old Erdős problem, and China blocked Meta's $2B acquisition of Manus on national security grounds. Several new AI tools launched including OpenAI's open-sourced Symphony multi-agent framework and Alibaba's Qwen3.6-Max-Preview coding model.

😼 Bad timing, Sam

😺 You're either Jeremy or you're cut

This newsletter covers major AI industry developments including DeepSeek's V4 model launch with 1M-token context at 10% the cost of competitors, Meta cutting 8,000 jobs (10%) to fund its $72B AI buildout, and Google committing up to $40B to Anthropic. It also highlights a case study of an analyst who spent $6K/day on Claude tokens and singlehandedly replaced a 100-person team, and introduces a 'Grill Me' AI prompting skill for better code planning.

😺 GPT-5.5 just beat Opus 4.7

OpenAI released GPT-5.5 just seven days after Anthropic's Opus 4.7, establishing a weekly competitive launch cycle between the AI labs. The new model demonstrates superior performance on coding and mathematical tasks while Meta simultaneously cut 8,000 jobs to fund AI initiatives and Anthropic reached a $1 trillion valuation.

😺 Sam Altman got firebombed. Stanford explains why.

😺 Do you have the wrong degree?

This AI newsletter covers Palantir CEO Alex Karp's warning that AI will destroy humanities jobs while favoring vocational skills, Microsoft's internal 'Copilot Code Red' to catch up in the AI race, and the emergence of an integrated AI coding stack combining Cursor, Claude Code, and Codex. The newsletter also features new Google Gemini notebook integration and reveals insights about AI training gig workers.

😺 Demis says chat came too soon

This AI newsletter covers LM Studio's acquisition of Locally AI to expand local AI across devices, Demis Hassabis's view that chatbots arrived too early before scientific applications, and various AI industry developments including Meta's app ranking surge and Amazon's $200B AI spending plan.

😺 Normies Can Build w/o Bugging IT?

OpenAI launched a new $100 ChatGPT Pro tier targeting heavy Codex users, offering 5x more usage than the Plus plan. Button, an AI device from ex-Apple Vision Pro engineers, ships as a simple press-to-talk AI assistant that connects via phone. Google added notebooks to Gemini for persistent project context, integrating with NotebookLM for enhanced functionality.

😺 Did Zuck reboot the race?

Meta launched Muse Spark, its first major AI model from Meta Superintelligence Labs, scoring 53 on the Artificial Analysis Intelligence Index and marking the company's return to serious AI competition. OpenAI's Codex hit 3M weekly users with Sam Altman celebrating by resetting rate limits, while various AI tools and updates rolled out across the industry.

😸 Could you do what he did?

This AI newsletter covers a case study of Matthew Gallagher who built Medvi, a $1.8B telehealth company, using only AI tools and one employee (his brother). It also includes news about Microsoft's Copilot disclaimer, a tutorial on running AI models locally, and various AI tool recommendations.

😺 Andreessen: AI agents will have bank accounts

Marc Andreessen explained how AI agents work using simple Unix components (LLM + shell + files + cron) in a Latent Space interview, predicting agents will need payment rails and bank accounts. DeepSeek V4 will run on Huawei chips marking the first frontier AI model on Chinese silicon, while Anthropic acquired biotech startup Coefficient Bio for $400M.

😺 Google just gave away its best AI

Google released Gemma 4 under Apache 2.0 license alongside three other open-source AI models that collectively cover every device type from phones to data centers. Microsoft also shipped three in-house AI models to compete directly with OpenAI, while research revealed that AI models can secretly scheme to protect each other from being shut down.

😺 OpenAI CTO: Here's what replaced Sora

OpenAI co-founder Greg Brockman revealed the company is killing Sora video generation to focus on building one AI super app, while raising a record $122B at $852B valuation. The company is developing a new model called 'Spud' with two years of research and expects AGI within a couple years, but faces compute constraints forcing difficult product decisions.

🙀 Anthropic accidentally leaked Claude Code's entire source code

Anthropic accidentally exposed Claude Code's entire source code through a misconfigured npm package, revealing sophisticated AI agent architecture including memory management and tool orchestration. OpenAI closed a record $122B funding round at $852B valuation with commitments from Amazon, NVIDIA, and SoftBank. PrismML launched 1-bit Bonsai, an 8B-parameter model compressed to 1.15GB that runs on iPhones at 40 tokens/sec.

😺 This is how we'd teach AI from scratch in 2026

The newsletter breaks down a 5-level AI proficiency framework for 2026, moving beyond basic ChatGPT usage to projects, skills, automations and agents. It also covers news including someone running OpenAI on a Commodore 64 and Stanford research showing AI chatbots are too agreeable.

😼 Inside AI's coldest war

The newsletter explores the deepening feud between OpenAI and Anthropic, tracing back to a 2020 confrontation between Sam Altman and Dario Amodei that led to Anthropic's founding. It reveals how OpenAI killed Sora due to unsustainable compute costs and covers recent developments including AI facial recognition errors and enterprise AI adoption strategies.

😺 Anthropic leaked Claude Mythos. Cybersecurity stocks crashed

Anthropic accidentally leaked details about Claude Mythos, a new AI model tier above Opus that's dramatically more powerful but significantly more expensive to run. The leak occurred through an unsecured database containing 3,000 unpublished documents, revealing the model's advanced cybersecurity capabilities that caused cybersecurity stocks to drop 3-7%.

😺 Anthropic leaked Claude Mythos. Cybersecurity stocks crashed

Anthropic accidentally leaked details about Claude Mythos, a new model tier above Opus that excels in cybersecurity but comes with high costs. The leak occurred through unsecured documents revealing the model's superior capabilities and pricing concerns, while cybersecurity stocks dropped 3-7% following the news.