AI Is Getting Too Fast

Welcome back apprentices! 👋

Hey {{first_name|friend}},

AI’s biggest names are having the kind of argument usually reserved for a family road trip: slow down, keep driving, or change the car entirely.

Nobody quite agrees on the speed limit.

And governments are making things even messier.

Meanwhile, one startup has quietly shown up with a very different answer.

_{Read Time: 4 minutes}

🧬 The AI That Builds the Next AI. More than 30 researchers from institutions including ByteDance, Tsinghua, and Shanghai AI Lab have mapped out five levels of AI self-improvement, ending with systems that can redesign the very process used to make their successors smarter. Most research today is still near the bottom of that ladder, while software engineering looks like the fastest route upward because an AI can write, test, and fix code almost instantly.

🩻 AI Joins the Ultrasound Room. In a five-hospital trial in China, sonographers using an AI assistant spotted specific fetal brain malformations more often — boosting detection from 78.6% to 87.3% without increasing false alarms. The catch? AI alone often did worse, and clinicians correctly overruled about 60% of its mistakes, making this less “doctor replacement” and more “very alert second pair of eyes.” The study was limited to high-risk pregnancies and added about 40 seconds per scan, but it offers a practical glimpse of where medical AI may fit best.

🍎 Siri Finally Got the Memo. Apple has rolled out Siri AI in iOS 27, giving its assistant screen awareness, access to context from apps like Mail, Messages, and Photos, and the ability to take actions across supported apps. The upgrade runs on Apple’s own models built alongside Google’s Gemini, but for now it’s an English-only beta and won’t launch in the EU or China. In other words, Siri is finally becoming more than a voice-controlled timer — though whether it can pull users away from standalone AI apps is still very much an open question.

Think about the last thing that got you fired up about your job this year. Odds are it didn't come from a report. It came from someone who does the job, saying what happened when they tried it.

So we've filled a day with people like that. 

Maya Shaff, ŌURA. Leah Walker, Adobe. Georgia Humphries, Stanley 1913. Enara Roy, Halfday. Tyler Vaught, Edelman. Josh Rangel, Ogilvy. Sarah Whittle, ex-Crocs and Duolingo. 

And that's just the first wave of speakers. 

Return on Influence Festival '26 is a free, virtual day dedicated to influencer marketing. Like any good festival, there are stages. But no mud, no queues, and absolutely no porta-potties. 

Sign up today, and the rest of the lineup lands in your inbox first.

Save your spot

AI’s biggest debate right now is about how fast the frontier should move at all. 

Anthropic CEO Dario Amodei is calling for companies to deliberately pace capability gains, warning that AI is beginning to help build newer AI and arguing that safety systems need time to catch up; Sam Altman and Elon Musk have publicly supported parts of that direction. Meta CEO Mark Zuckerberg disagrees with the need for coordinated slowing, arguing that companies already have strong incentives to delay systems when necessary and pointing to Meta’s own months-long safety hold on Muse. 

Meanwhile, President Donald Trump has rejected calls for broad AI slowing, while China’s Foreign Ministry has criticized what it calls “fear-mongering” and confrontational competition around AI. 

And off to the side, startup TypeSafe is asking a completely different question: what if some software doesn’t need a giant chatbot at all?

Amodei’s concern is recursive self-improvement: AI taking on more of the work required to build even better AI, potentially accelerating progress faster than safety systems can keep up. His proposed fix includes outside evaluators, shared safety standards and eventually international agreements around particularly rapid forms of self-improvement.

His boldest prediction is that within 6–12 months, poorly controlled AI agents could become capable of serious cyber damage — a forecast, not an established timeline. 

The broader question is simpler: if AI starts speeding up AI research itself, will today’s safeguards still move fast enough?

Zuckerberg agrees safety matters but questions the need for a coordinated slowdown. 

His argument: companies already have strong incentives not to release unreliable AI, because an agent that ignores users isn’t exactly a killer feature.

Meta points to Muse, which it says was delayed for months over safety concerns without asking competitors to pause too. In short, Anthropic wants more collective guardrails; Meta thinks labs can largely decide when to brake themselves.

Global coordination gets much harder when governments view AI leadership as a strategic advantage. Trump has rejected a broad slowdown and emphasized keeping the U.S. ahead, while China has criticized fear-driven restrictions and called for more open cooperation.

That creates the obvious problem: countries may hesitate to slow their own companies if they think rivals will keep accelerating. Narrow agreements on issues like cyber or biological testing may be possible; a global AI speed limit is a much bigger ask.

TypeSafe is taking a different route entirely: make the AI narrower on purpose. Its Jev model chooses between predefined answers instead of generating free-form responses, targeting quick tasks like routing requests, scoring records and screening AI outputs.

TypeSafe says Jev responds in 70–500 ms and costs just $0.042 per million input tokens. Its “can’t hallucinate” claim needs an asterisk: Jev can’t invent an unsupported output format, but it can still choose the wrong answer. Think less “digital coworker,” more if-statement with a brain.

For most people, this debate will eventually show up in much less philosophical ways.

If coordinated safety rules gain traction, frontier models could face more testing, outside evaluation and potentially slower capability rollouts. If Meta’s more decentralized approach dominates, companies would have greater freedom to decide when their own systems are safe enough to ship. And if specialized models like Jev prove effective, a lot of the AI running businesses may become almost invisible — tiny decision engines working behind apps rather than chatbots sitting in front of them.

For companies, that means the useful question may increasingly be “Which kind of AI belongs here?”, not simply “Which giant model should we buy?”

For everyone else, watch the boring infrastructure stories. Evaluations, model access, chip rules, structured AI and safety standards sound much less exciting than a robot takeover — but they may determine how quickly the next generation of AI actually reaches your screen.

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 👀

WorldQuant BRAIN is a global simulation and research platform for students, aspiring researchers and quantitative thinkers from around the world.

On BRAIN, that spirit of exploration continues through challenges, collaboration and opportunities to develop new quantitative research ideas.

Learn more about WorldQuant BRAIN®

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 Jasper for this creation!🥳

Want to be featured next? Keep those generations coming!

Inside an enormous futuristic research laboratory, a giant scientific microscope is examining [your object] at impossible scale. The object sits on a precision glass platform while articulated lenses, scanning arms, and beams of colored light reveal its microscopic structure: tiny mechanisms, fibers, textures, and hidden layers normally invisible to the human eye. The laboratory is bright and vibrant, with cyan, violet, and warm amber illumination reflecting across glass and polished metal. Ultra-detailed photorealism, cinematic macro photography, razor-sharp textures, dramatic depth of field, high-end scientific visualization.

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