This week in AI focuses on AI demos & interviews with AI's top personalities.

This Week in AI

The battle over model weights has moved from developer forums to national security briefings. Does open source democratize AI or hand dangerous tools to bad actors? Let's unpack whether the current regulatory push will cement a fortress AI under big labs or force a sprawling open ecosystem.

The open source vs. closed source AI divide is the defining geopolitical fight of this decade

Pre-training is running into a wall of exhausted human-generated text. Scaling up synthetic data can extend the runway, but model collapse and homogenization are real threats. Is there a sustainable path forward, or are we about to watch AI models eat their own tail?

The real bottleneck in AI is no longer compute — it's data, and synthetic data is a double-edged sword

The old frame of reward hacking has given way to a broader concern: can we build AI that doesn't just comply but genuinely remains open to correction and transparent about its own uncertainty? And why do we still lack agreed-upon benchmarks for these properties?

AI alignment research has quietly shifted from 'follow instructions' to 'be truthful and corrigible'

Every technological revolution has brought both displacement and creation. But AI's ability to reason and act may compress that adjustment period. What do the early signals from coding, design, and customer support tell us about the shape of work in 2030?

Will AI transform labor markets by replacing jobs, augmenting humans, or inventing entirely new roles?

The EU AI Act, US executive orders, and China's own rules are pulling AI governance in different directions. Meanwhile, open-source models move across borders. Could we end up with incompatible AI stacks, and what would that mean for scientific collaboration and global competitiveness?

As the West regulates and China accelerates, will we see a split internet with two distinct AI ecosystems?

As agents book flights, write code, and manage finances, the question of legal responsibility becomes urgent. Should the user, the developer, or the deployer bear the risk? And can current tort law even handle a system that acts independently across many jurisdictions?

Who is liable when an autonomous AI agent causes real-world harm?

Most founder podcasts settle for the polished press tour. Ti Morse is building the opposite: long-form, cinematically staged interviews with the operators actually shipping the future — and the people who don't usually sit for podcasts are saying yes.
In months, Morse has become one of the most trusted interviewers in hard tech, defense, space, and frontier AI. The guest list maps where serious capital and talent are flowing:

Brian Armstrong (Coinbase), Peter Beck (Rocket Lab), Baiju Bhatt (Robinhood, now building orbital data centers at Aetherflux) Apple Podcasts
Sulaiman Ghori on xAI's Colossus, Keller Cliffton on Zipline's 2 million drone deliveries Amazon
Justin Fiaschetti (Inversion's sub-hour cargo from orbit), Clem Delangue (Hugging Face) Spotify for Creators

What lands is the production. For Armstrong, Morse's team built the Iron Bank of Braavos as a set and commissioned a four-foot, 1,000-pound marble Bitcoin. For Beck, he flew to Rocket Lab's HQ in New Zealand. Founders sit because they can tell the work matters to him. LinkedIn
He's documenting the El Segundo hard-tech renaissance — Hadrian, Neros, Castelion, Picogrid — at the moment American reindustrialization and defense tech moved from niche thesis to consensus.
Interviewing the greats is the tagline. The honest version: he's interviewing the people who will be the greats, on the way up.

Get Garrett Scott 🕳 on This Week in AI

The Case for This Conversation

Shape What Gets Discussed

You Might Also Enjoy