SunBrief#83: ️AI Is Now Building Itself.

Welcome to the SunBrief

Today in SunBrief 🌞

• Turn Your AI Idea Into a Working Product

• Anthropic warns that AI will soon be able to improve itself

• Trump administration eyes stake in OpenAI

• Stock Updates

• OpenAI Upgrades ChatGPT Memory With “Dreaming” System

• AI Highlights of the Week

• Too Important to Miss

You Can’t Scale If You’re Still Doing Everything

As your business grows, your role should evolve with it.

But for many leaders, it doesn’t.

They stay buried in tasks, decisions, and responsibilities they’ve already outgrown.

BELAY created the free resource From Operator to Owner to show leaders how to step out of the day-to-day without losing control and what it takes to operate at the next level.

At BELAY, we match you with U.S.-based Assistants who take on the operational work that’s keeping you stuck.

Anthropic warns that AI will soon be able to improve itself

New report outlines progress toward recursive self-improvement and warns institutions may not be ready

Anthropic says AI is already accelerating AI development, with Claude writing code, running experiments, reviewing work, and helping researchers move faster. The company says this could eventually lead to recursive self-improvement, where AI helps build its own successors.

Key Points:

• AI-built AI is becoming real: Anthropic says it is delegating more AI development work to Claude, including coding, debugging, experiment execution, and code review.

• Major coding productivity gains: As of May 2026, Anthropic says more than 80% of code merged into its codebase was authored by Claude, and engineers now ship about 8x more code per quarter than before.

• Longer autonomous tasks: Public benchmarks show AI task length is improving fast, with models moving from completing minutes-long tasks to hours-long and even 12-hour tasks.

• Research acceleration: Claude can now run well-defined experiments, optimize training code, and propose research steps, though humans still lead on high-level judgment and choosing what problems matter.

• Recursive self-improvement risk: Anthropic warns that if AI systems become capable of fully building their successors, safety, monitoring, alignment, and oversight become much more urgent.

• Possible futures: The report lays out three paths: progress stalls, AI labs gain compounding productivity, or AI reaches full recursive self-improvement and humans shift mostly to oversight.

Why It Matters:
This is one of the clearest warnings yet that AI development could become self-accelerating. If AI can write code, run experiments, and improve future systems, progress may speed up fast, bringing big gains but also harder safety and alignment challenges.

Stock Updates

OpenAI and Trump Administration Discuss Possible U.S. Government Stake

Talks explore equity donation to seed a public wealth fund tied to AI growth

OpenAI CEO Sam Altman and the White House are reportedly discussing a possible U.S. government stake in OpenAI, with the company potentially donating equity to support a public wealth fund that lets Americans share in the economic upside of AI.

Key Points:

• Government stake talks: OpenAI and the Trump administration have been discussing a possible government equity stake for more than a year.

• Public Wealth Fund idea: OpenAI could donate equity to seed a fund that invests in long-term assets and potentially distributes returns to citizens.

• No final terms yet: The structure, size, and official terms of any government stake have not been finalized and could still change.

• Trump’s support: President Trump said there are concepts where “pieces could be given to the American public,” making citizens partners in AI growth.

• Broader government strategy: The Trump administration has already taken stakes in companies like Intel and IBM, and has pushed for a federal sovereign wealth fund.

• AI policy connection: The talks come as the White House accelerates AI adoption across national security agencies and asks AI companies to provide model access before release.

Why It Matters:
This could mark a major shift in how governments participate in the AI economy. If the U.S. takes an equity stake in OpenAI, it would blur the line between public infrastructure, private AI companies, and national economic strategy while giving citizens a potential share in AI’s financial upside.

OpenAI Upgrades ChatGPT Memory With “Dreaming” System

New memory architecture improves freshness, continuity, and personalization across long-term conversations

OpenAI is rolling out a stronger ChatGPT memory system called dreaming, designed to pull useful context from past chats in the background. The goal is to remember preferences, projects, and constraints more accurately while reducing outdated memory over time.

Key Points:

• Better long-term memory: Dreaming helps ChatGPT automatically synthesize context from many past chats instead of relying only on explicit “remember this” instructions.

• Memory summary page: Users can review what ChatGPT knows about them, correct details, add updates, or guide when certain topics should be used.

• Improved factual recall: OpenAI says the new system improves context recall from 41.5% in 2024 to 82.8% in 2026 in internal evaluations.

• Stronger preference adherence: ChatGPT is better at applying user preferences and constraints, such as dietary needs, travel interests, location, or response style.

• Less stale memory: Dreaming updates memories as time passes, helping ChatGPT avoid outdated assumptions like thinking a past trip is still happening.

• Broader rollout: The update is available to Plus and Pro users in the U.S. first, with rollout to more countries and Free and Go users in the coming weeks.

• More scalable system: OpenAI says recent improvements reduced the compute required to serve dreaming by about 5x, making it practical to bring better memory to more users.

Why It Matters:
Memory is making ChatGPT feel less like a one-off chatbot and more like a long-term assistant. Better recall and personalization could help it support ongoing projects, adapt to preferences, and give more relevant answers without users repeating themselves.

AI Highlights of the Week

• Apple’s WWDC 2026 Could Be Siri’s AI Comeback
  Apple is expected to make Siri and Apple Intelligence the main focus of WWDC 2026, with smarter AI features and more useful actions across apps.

  The update could bring chatbot-style Siri, better AI image editing, smarter shortcuts, and deeper AI support across iOS 27.

• Google’s Gemma 4 12B Brings Multimodal AI to Laptops
  Google DeepMind introduced Gemma 4 12B, a new multimodal model that can handle text, vision, and native audio while running locally on laptops.

  The model uses an encoder-free architecture, needs only 16GB of memory, and brings faster local AI agents with strong reasoning and offline multimodal support.

• Microsoft Opens Copilot Health Preview in the US
  Microsoft launched Copilot Health in preview, helping users connect health records, wearable data, and health questions in one secure space.

  The tool can give personalized health insights, explain lab results, and help users find the right healthcare provider.

• Reve 2.0 Makes AI Images More Editable and Controllable
  Reve launched Reve 2.0, a new image model that plans visuals as code before rendering them, making AI images easier to edit and control.

  The model can create native 4K images, improve text rendering, reduce editing degradation, and give creators more precise control over layouts.

Too Important to Miss

Microsoft Launches Seven MAI Models to Build Its Own AI Frontier

Perplexity Computer Will Run AI Tasks Locally and in the Cloud

NVIDIA Launches Its Most Efficient Reasoning AI Model Yet

Google Just Gave Android a Deepfake Call Detector

Last Week’s Poll Result

• Will AI agents soon handle entire projects with minimal human involvement?

  Maybe, but humans will stay involved → 51.22%

  Yes, we’re getting close → 39.02%

  No, AI isn’t ready for that → 9.76%

• Will AI-native PCs become the new standard within the next 5 years?

  Yes, they’re the future → 66.67%

  Maybe, but adoption will be gradual → 16.67%

  No, traditional PCs will remain dominant → 16.67%

• Could space become the next major frontier of military competition?

  Maybe, but slowly → 45.45%

  Yes, it’s already happening → 36.36%

  No, conflicts will stay Earth-based → 18.18%

Feedback

Know someone who may be interested?