AI The Creator; ChatGPT The Consultant; Google The CarBot

Today's AI Outlook: 🌤️

AI Learns To Write... Life

Stanford and Arc Institute’s Evo work landed in Nature with a milestone that sounds like science fiction with a lab coat: AI was used to design full bacteriophage genomes. Researchers synthesized 302 fully AI-written designs, and 16 became working viruses that infected, copied themselves and killed E. coli.

This is a leap from AI predicting biology to executing it. AlphaFold helped scientists understand protein shapes. Evo points at something bigger and weirder: genomes as prompts, outputs and products.

Why it matters

AI is starting to move from “understands life” to “builds biological machines.” That could speed up pharma, synthetic biology and drug discovery, while also making biosecurity people reach for stronger coffee.

The Deets

• Researchers created fully AI-designed bacteriophage genomes.
• 16 designs worked in wet-lab tests.
• One AI-designed phage included a capsid protein with no known natural relative.
• The model appears to be exploring biological design space beyond known nature.

Key takeaway

The “creator mode” metaphor is getting uncomfortably literal. Biology is becoming programmable, and the GPU curve is coming for the petri dish.

🧩 Jargon Buster - Bacteriophage: A virus that infects bacteria, often used in research because it can target specific bacterial cells.

Compute Goes Nautical Because... NIMBY

Oregon startup Panthalassa raised a $140M Series B led by Peter Thiel to build autonomous floating compute structures powered by ocean waves. The company’s idea: take AI infrastructure offshore, use wave motion for electricity, cool chips with seawater and beam results back via Starlink.

Each planned node is an 85-meter steel structure that can move to remote waters using hull design rather than engines. The new funding will help finish a pilot factory near Portland and send the first wave-powered compute nodes into the Pacific, with commercial rollout planned for 2027.

Why it matters

AI’s hunger for compute has collided with public backlash over data centers, power draw and local disruption. Floating infrastructure is still early, but compared with space-based compute ideas, the ocean looks like the more practical frontier.

The Deets

• Panthalassa is reportedly valued near $1B.
• Its offshore nodes use wave energy and seawater cooling.
• Starlink would handle communications back to shore.
• The startup is aiming for commercial deployment in 2027.

Key takeaway

AI infrastructure may be getting pushed off land, literally. When local zoning gets spicy, the Pacific starts looking like a data center with better views.

🧩 Jargon Buster - Wave-Powered Compute: Data-processing infrastructure that uses ocean wave motion to generate electricity for chips and servers.

🏛️ Power Plays

Frontier Labs Discover Consulting

Anthropic and OpenAI are both moving deeper into enterprise deployment through private equity-style ventures. Anthropic announced a $1.5B Claude services company with Blackstone, Hellman & Friedman and Goldman Sachs, while OpenAI is reportedly raising for its own Deployment Company with $4B from investors at a $10B valuation.

The goal is not just selling model access. It is helping companies actually install AI inside messy business workflows, where spreadsheets, compliance rules and ancient software go to multiply.

Why it matters

The AI bottleneck for many companies is no longer whether models are powerful enough. It is whether anyone can wire them into real operations without breaking everything or launching another “innovation task force” that produces a deck and a migraine.

The Deets

• Anthropic’s venture focuses on mid-sized companies using custom Claude workflows.
• OpenAI’s reported venture includes investors such as TPG, Brookfield, Bain and SoftBank.
• Both strategies could give AI labs access to private equity portfolio companies.
• This turns frontier labs into something closer to AI-native consulting firms.

Key takeaway

The enterprise AI race is moving from model leaderboard to implementation muscle. Whoever owns deployment may own the customer.

🧩 Jargon Buster - Deployment Company: A services business built to help companies integrate AI systems into actual workflows, software and operations.

🛠️ Tools & Products

Google Building 'KITT'

Google is bringing Gemini into cars with Google built-in, replacing the older Assistant layer with a more conversational AI experience. The rollout starts in English in the U.S. and will expand through software updates.

Gemini will handle natural conversation, Maps-aware routing, message summaries, music control, EV charging help, vehicle manual answers and cabin commands. GM alone says around 4 million eligible U.S. vehicles can get Gemini.

Why it matters

The car interface is becoming prime AI real estate. Tesla owns more of its in-car stack, but Google has Maps, Gmail, Calendar, Home and deep automaker distribution. That gives Google a strong shot at becoming the default AI layer for non-Tesla vehicles.

The Deets

• Gemini will replace the old Google Assistant experience in supported cars.
• Features include routing, summaries, EV charging help and cabin controls.
• The first rollout starts in English in the U.S.
• Google’s automaker footprint gives it a large installed base.

Key takeaway

The next platform fight is beyond phones and browsers, and into the cockpit... and Google wants drivers to talk to Gemini before they touch anything else.

Your iPhone Can Run A Local AI Sidekick

The Rundown AI shared a guide for replacing Siri-style interactions with a free local AI model on an iPhone using Locally AI. The setup lets users bind a local model to the iPhone Action Button, so prompts run on-device without internet access or sending data out.

The suggested starting point is Google’s open-source Gemma model, with users able to test larger models for better answers at the cost of speed and storage.

Why it matters

Local AI is becoming more useful for everyday tasks like translation, math and explanations. It also gives privacy-conscious users a simple way to try AI without pushing every question through the cloud.

The Deets

• Download Locally AI from the App Store.
• Pick a model, such as Gemma.
• Bind Voice Mode to the iPhone Action Button through Shortcuts.
• The first use may require a speech-to-text model download.

Key takeaway

Cloud AI is still more powerful, but local AI is getting good enough for pocket-sized utility.

🧩 Jargon Buster - Local Model: An AI model that runs directly on your device rather than sending prompts to a remote server.

🧪 Research & Models

AI May Start Building Its Own Replacements

Anthropic co-founder Jack Clark published a forecast putting 60%+ odds on AI systems training their own successors before 2029. His argument draws on public papers and benchmark data showing AI moving rapidly across core research and development tasks.

One cited measure from METR shows AI’s independent work capability growing from 30-second tasks in 2022 to 12-hour tasks in 2026, with 100-hour runs projected by year-end. Clark also pointed to SWE-Bench progress, where coding performance rose sharply from Claude 2 at 2% to Mythos Preview at 93.9% in under three years.

Why it matters

Self-improving AI would change the pace of model development. Instead of humans building each next generation, AI systems could increasingly automate research, coding, training and evaluation loops.

The Deets

• Clark gives 60%+ odds that AI trains successor systems before 2029.
• AI’s independent task horizon has expanded dramatically since 2022.
• SWE-Bench scores suggest fast gains on real-world coding tasks.
• OpenAI is reportedly targeting an automated research intern by September 2026.

Key takeaway

The AI race may soon become all about whose AI can help create the next AI fastest.

🧩 Jargon Buster - Self-Improving AI: An AI system that can help design, train or improve future AI systems with limited human involvement.

Hollywood Gives AI A Red Carpet Wristband

The Academy is opening the door for films that use AI tools to remain eligible for Oscars, while still requiring acting and writing to be human-led. Synthetic actors and AI-authored scripts will not count as real creative contributions for the 2027 awards cycle, but AI-assisted filmmaking can still compete.

That creates a new line for Hollywood: AI can help make the movie, but humans still need to be credited for the core artistry.

Why it matters

Award eligibility matters because it gives studios permission to use AI without losing prestige. That could accelerate AI adoption across editing, VFX, dubbing, localization, concept art, cleanup, sound and trailer workflows.

The Deets

• AI-assisted films can remain Oscar eligible.
• Acting and writing must remain human-led.
• Synthetic performers and AI-written scripts will not qualify as creative contributions.
• Smaller film-service vendors may feel pricing pressure as studios adopt AI tools.

Key takeaway

The Academy tried to protect human creativity, but it also gave studios a green light to bring AI deeper into production.

🧩 Jargon Buster - Previs: Short for previsualization, an early visual mockup of scenes used before full production or expensive VFX work begins.

Today’s Sources: AI Secret, The Rundown AI