OpenAI built a memory system that learns you without asking. Anthropic released its most capable model publicly. NEURA Robotics raised $1.4 billion for humanoid robots.
One side is building persistence. The other just proved forgetting was never really an option.
This Weekend read
Most SaaS founders have a distribution problem, not a product problem. This thread is about how AI is solving it through automated short-form video.
Worth ten minutes if you're building anything that needs an audience.
Special Coverage
ChatGPT now builds a model of you in the background
OpenAI's Dreaming V3 doesn't ask what to remember. It watches. The system reads across years of conversation history and synthesizes a continuously updated user model, adjusting automatically as your situation changes.
Three things it tracks: context across chats, your stated preferences, and time. A trip logged as "going to Singapore in July" becomes "went to Singapore in July 2026" once the date passes. That last one is new, and it matters. Previous memory systems stored facts. This one manages their shelf life.
Overall recall accuracy hit 82.8%, up from 41.5% in 2024. Rolling out now to US Plus and Pro users. Free and international users follow in coming weeks.
Source: OPENAI
Lead Analysis
Neura robotics raised $1.4b.
German robotics company NEURA Robotics closed a Series C of up to $1.4 billion on June 10, backed by Nvidia, Amazon, Qualcomm, Tether, Bosch, Schaeffler, and the European Investment Bank. Valuation: approximately $7 billion per Bloomberg. The full raise is contingent on milestones.
Robotics companies globally have pulled in $55.8 billion so far in 2026, nearly double last year's record. NEURA's flagship 4NE1 humanoid targets both factory floors and domestic environments. Existing order book exceeds $1 billion. Target: millions of units by 2030.
Anthropic released its most capable model to the public
Claude Fable 5 is Anthropic's first publicly available Mythos-class model. It tops nearly every current benchmark across coding, reasoning, vision, and long-context work. Scored ~80% on SWE-Bench Pro in independent testing. The longer the task, the wider its lead over prior Claude models.
It ships with hard classifiers that block responses in cybersecurity, biology, and chemistry, falling back to Opus 4.8 in those areas. The sibling model, Mythos 5, is the same intelligence without those classifiers, available only to vetted partners through Project Glasswing.
Free on Pro, Max, Team, and Enterprise plans through June 22. After that, usage credits. API pricing is $10/$50 per million input/output tokens.
Source: Anthropic
Autonomous labs are handing science to robots
Scientists at Ginkgo Bioworks and others are running experiments inside fully autonomous laboratories where AI and robots handle routine work, log results, and write lab notes with minimal human input.
This isn't a pilot. It's a production shift in how scientific research gets done. The bottleneck in drug discovery and biotech has always been throughput. Autonomous labs remove it.
Quick Hits
Before we get into the news, here’s a quick tool that solves a common problem. A Chrome extension that turns temporary ChatGPT chats into permanent ones with one click. Watch video→
AstraZeneca says AI is cutting drug development time and cost. The company reports measurable gains in molecule design, clinical trial prediction, and research pipeline speed. Pharma has a real-world benchmark for AI-assisted discovery now.
Apple's Siri AI is designed to say less. The rebuilt Siri, now running on Google Gemini, gives shorter and more direct answers. Apple is explicitly moving away from the chatty companion design. The product philosophy shift is as notable as the model swap.
ChatGPT was used to run foreign influence operations on AI policy. OpenAI confirmed China-linked actors used ChatGPT to generate content targeting US debates on AI and trade. The tools used to build AI are now being used to shape how governments regulate it.
SpaceX priced its IPO today at $135 per share. The $1.75 trillion valuation makes it the largest public offering in recorded history. Trading opens on Nasdaq under SPCX tomorrow. Anthropic pays SpaceX $1.25 billion per month for compute. That relationship is now a public company's balance sheet.
Worth Reading
[Read] Google Research: Auditing machine unlearning Google Research unveiled a new machine unlearning audit framework that can detect whether AI models have truly erased sensitive data, a key step as privacy laws and AI safety demands continue to grow.
[Read] Scaling Participation in AI — The argument here is that one giant model may not be the endgame. A modular system of smaller, contributor-built models outperformed monolithic LLMs by 15.4% across reasoning and factuality tasks.
Worth a Try
The Babylon Bee ran a prompt experiment asking AI to simulate what would happen if AOC were forced to learn economics.
Created using Grok