Morning Brief: AI progress, safety, and practical tools
Good morning, {{ first_name | AI enthusiasts }}. Recursive self-improvement once sounded like sci‑fi. Now, leading labs report internal signs that AI is accelerating its own development. Anthropic's new report shows AI already speeding up the work that builds successors and outlines risks — including the possibility of a coordinated industry pause. Below is today's AI rundown, plus practical tools and community workflows. We also include a note on gaming news trends tied to AI-driven experiences.
Key stories
Anthropic charts a path to self‑improving AI
The Rundown: Anthropic released "When AI builds itself," a report on recursive self‑improving (RSI) systems. The company shared internal data showing Claude increasingly writes code that contributes to new versions. Anthropic warns fully self‑improving AI could arrive before institutions adapt.
The details: Anthropic says RSI isn't guaranteed, but Claude is advancing development "faster than we thought." As of May, more than 80% of merged code at Anthropic was authored by Claude. Engineers pushed eight times as much code per day in Q2 2026 compared with 2024.
"Each new version of Claude could be built by the version before it, without human involvement." — Jack Clark
OpenAI raised the same concern in its "Democratic Governance of Frontier AI" blueprint, noting early RSI sparks in current systems. Anthropic pledged to slow or pause frontier AI development if peer labs do the same, and plans policy discussions on research, systems, and scenarios.
Why it matters: Multiple labs report similar dynamics. Startups focused on self‑improvement are emerging, and models like MiniMax's M2.7 reportedly helped build themselves. The potential upside is rapid progress; the downside is governance and safety gaps that may be hard to coordinate globally.
OpenAI's memory overhaul lets ChatGPT "dream"
The Rundown: OpenAI launched a memory update called "dreaming." It runs in the background, converting past chats into a categorized summary that provides better personalization and longer‑term context.
- ChatGPT now keeps written summaries grouped by topics like travel, hobbies, and work.
- Users can review, edit, or suppress individual memories; summaries update automatically.
- OpenAI reports factual recall improved from 41.5% to 82.8% and preference-following from 31.4% to 71.3% in internal tests.
- Rolling out to Plus and Pro users in the U.S. first; Free and Go and other regions follow soon.
Why it matters: Memory boosts continuity and personalization — key retention drivers in consumer AI. It could also influence how AI integrates into long-term workflows, from productivity to entertainment and gaming news personalization.
Stress test business ideas with Perplexity
The Rundown: Perplexity Deep Research helps validate business ideas quickly. Use a reusable prompt to run research and generate slide decks automatically.
- Open Perplexity and switch to Deep Research mode (free plan supports limited queries).
- Paste a saved prompt with your idea, run it, and wait 5–6 minutes for a research-backed slide deck.
- Store the prompt in a Perplexity space and schedule weekly runs to vet multiple ideas.
Pro tip: Create variants — a 6‑slide pitch for co‑founders, a comparison between ideas, or a 90‑day MVP plan for validated concepts.
GitLab Transcend live event
The Rundown: GitLab Transcends streams from London on June 10–11. Expect live demos, agentic AI use cases, and developer-focused sessions highlighting intelligent orchestration and the Duo Agent Platform.
AI and biosecurity: rival labs unite
The Rundown: CEOs from OpenAI, Anthropic, Google DeepMind, and Microsoft signed an open letter urging U.S. Congress to require synthetic‑DNA vendors to vet buyers and log orders. The letter warns that AI can now design dangerous biological agents and that knowledge barriers are eroding.
Signers include Sam Altman, Dario Amodei, Mustafa Suleyman, Alexandr Wang, and Demis Hassabis. The letter calls for mandatory screening, buyer verification, and sales logs to trace risky sequences.
Why it matters: This is a cross‑industry signal that AI accelerates biological risk. It underscores the need for rapid legal and regulatory adaptation.
Trending AI tools
- Reve 2.0 — 4K image model with layout‑aware editing
- Miso One — open TTS model that conveys expressive tones
- Stack — Ramp's AI accounting OS for monthly bookkeeping
- Nemotron 3 Ultra — Nvidia's open 550B reasoning model for agentic systems
Other notable updates: The U.S. and Japan announced a $1B AI research partnership. Nvidia released Nemotron 3 Ultra, claiming faster and cheaper agent performance. Canada launched a national AI strategy targeting $200B growth and 250,000 jobs. Bot traffic on the web now exceeds human traffic, according to Cloudflare's co‑founder.
Community workflow spotlight
Each issue highlights how a reader uses AI. This week's workflow comes from a reader who uses scheduled prompts to get parenting tips. The tool suggests weekend activities, behavior strategies, and bedtime routines — helping a busy parent manage toddlers more effectively.
How do you use AI? Tell us and we may feature your workflow.
Why this matters — short takeaways
- RSI signals are appearing across labs; governance and coordination remain the central challenges.
- Personalization features like ChatGPT's dreaming can change user expectations across productivity and gaming news consumption.
- AI is reshaping biosecurity and industry regulation debates.
- New agent models and tools are accelerating agentic applications — from developer tooling to consumer assistants.
See you soon, Rowan, Joey, Zach, Shubham, and Jennifer — the human team behind The Rundown
Comments
No comments
Add Comment