AI Is Starting to Build the AI That Comes Next

Welcome back apprentices! 👋

Hey {{first_name|friend}},

AI labs basically looked at the speed limit this week and said, “cute.”

New models, math breakthroughs, nervous researchers — somehow, they all showed up at once.

Somewhere between “this is incredible” and “uh… should we be worried?” is a story worth paying attention to.

_{Read Time: 4 minutes}

🧬 AI Drug Turns Back the Clock. Insilico Medicine says its AI-designed lung drug, rentosertib, made patients appear biologically younger across all six aging tests used in a small follow-up study. In one measure, estimated biological age dropped by roughly 2.7–3.5 years, suggesting the drug may be doing more than simply helping damaged lungs. It’s still early evidence from a limited sample, but it offers a glimpse of what AI-designed medicines could eventually unlock beyond treating a single disease.

🎵  Less Lawsuit, More Licensing. Suno just launched v6, a new family of AI music models built using licensed data alongside Warner Music Group, BMG, and Believe — a notable shift after years of copyright battles over how its earlier models were trained. The lineup includes two paid models and a free v6-mini, while Suno says artist-approved, paid fan remixes are coming next. With other lawsuits still ongoing, the bigger story is that AI music companies are increasingly moving from fighting the music industry to building with it.

🌦️ AI Wants to Outsmart the Weather. Google DeepMind launched WeatherNext 3, a weather model that refreshes forecasts every hour using live satellite and ground data — and it’s already outperforming rival models and major U.S. and European forecasting systems. Google says it delivers much finer local detail and cuts some rain-forecast errors by up to 60%, traditionally one of AI weather prediction’s weak spots. It’s now being rolled into Search, Maps, Gemini, and Earth, meaning better forecasting could soon show up without users even realizing AI is behind it.

Found the right person, but they’re in a country where you don’t have an entity? Setting one up can take months and significant cost.

Remote removes that barrier by becoming the legal employer through our own entities — handling compliant contracts, local benefits, tax setup, and onboarding for you. In fact, an employee is onboarded to Remote every 7 minutes.

Once they’re hired, the same in-house teams that support employment locally also run payroll — so you’re not bouncing between disconnected providers. Less setup, less complexity, and less time between finding the right person and getting them started.

See how it works

This week in AI feels a little like watching someone install a bigger engine while another person asks whether the brakes have been tested. 

OpenAI unveiled GPT-6 Astra, showed that AI agents inside the company are already doing more total research work than its human researchers by one measure, and then revealed that an even more powerful internal system had produced a proposed solution to one of mathematics’ famous Millennium Prize Problems. Meanwhile, former OpenAI and Anthropic researcher Jacob Coxon resigned from Anthropic, warning that frontier labs are moving too quickly toward self-improving AI, while Anthropic alignment researcher Evan Hubinger publicly put his own estimate of catastrophic AI risk above 10% over the next decade. 

The important part isn’t deciding whether the optimists or pessimists win the argument today. It’s that AI is increasingly being used to build better AI, making the speed of the race itself one of the biggest stories to watch.

Coxon’s resignation highlights a deeper concern: future AI systems could help build their successors, accelerating progress faster than humans can understand or control. Hubinger’s >10% extinction estimate is a personal judgment about future systems, not evidence that current models pose that level of risk.

Anthropic emphasizes safety, yet some of its own researchers believe the industry lacks a reliable way to control more capable systems. Coxon favors coordinated slowing, while labs generally argue for continued development alongside stronger safeguards. The disagreement is growing as capabilities improve.

GPT-6 Astra posts major benchmark gains, including 99.9% on ARC-AGI-3, though it does not lead every overall ranking. That makes broad claims about the “smartest” model dependent on which abilities are measured. OpenAI President Greg Brockman has suggested Astra could represent AGI, while acknowledging that AGI remains a disputed concept.

The bigger shift is behind the scenes: OpenAI researchers increasingly use AI agents to write code, troubleshoot systems and run experiments. Humans still provide substantial guidance, but machines are handling more of the execution.

OpenAI says an unreleased model more capable than Astra coordinated thousands of agents to produce a proposed Navier–Stokes solution in days. The work still requires mathematical scrutiny, and a parallel effort has sparked a dispute over research priority and possible indirect influence from AI training data.

The broader question is unavoidable: what happens when AI systems help researchers create the next generation of AI? Rules around confidentiality, attribution and training data may need to evolve quickly.

You probably don’t need to care whether ARC-AGI-3 is 99.9% or 97.9%. What matters is the emerging feedback loop: better AI helps researchers code and experiment faster → faster research helps produce better AI → those better models can take on even more of the next research cycle.

That could compress years of progress into much shorter periods — not just in chatbots, but in medicine, mathematics, cybersecurity, engineering and scientific discovery. It could also make mistakes, security failures and bad decisions propagate faster, which is why the safety argument is no longer separate from the capability story. OpenAI itself says it does not yet know how to safely reach fully aligned recursive self-improvement and has previously slowed some development after discovering safety problems.

So forget the daily “Did we reach AGI?” scoreboard for a moment. The more useful question is becoming: How much of the work required to create the next generation of AI can the current generation already do?

Right now, the answer appears to be: increasingly, quite a lot.

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

Practice Docker, Kubernetes, Linux, Terraform, and Git in real terminal environments. 

Explore hundreds of hands-on labs across the entire DevOps stack that let you provision clusters, automate infrastructure, and manage deployments. Experiment freely across multiple playgrounds that offer sandbox environments. 

Learn DevOps by doing, not just videos or theory.

Start For Free

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

Here’s the WINNER:

Congrats to Walter for this creation!🥳

Want to be featured next? Keep those generations coming!

Inside a massive glass-and-steel train station, [your object] is captured at the exact moment it is moving through the space at incredible speed. Multiple synchronized versions of the object appear along its trajectory, creating a precise visual sequence from stillness to full motion — like a high-end scientific photograph frozen in time. Commuters and architecture remain sharp in the background, while subtle motion trails reveal the object's movement. Vibrant cyan, amber, and violet lighting reflects across the polished floor and glass surfaces. Ultra-detailed photorealism, cinematic wide-angle photography, crisp textures, dynamic composition, dramatic perspective.

If AI outputs feel inconsistent, it’s usually not the model, it’s missing structure.
We documented the exact 6- Part System we use to get reliable results across ChatGPT, Claude, and Gemini.

It’s a short guide you can finish in under an hour, with plug-and-play prompts + exercises so you actually build the skill and fix the frustrating AI inconsistencies.

Subscriber Price: $10 (normally $19).

_FEEDBACK