“A year prior LLMs were just starting do correct math. Now ChatGPT could reproduce his hardest paper in the time it takes to get a coffee.”

Some people are going crazy over GPT 5.5. Some people. This is the story of the Jagged Frontier. People who use AI to write emails or even code implementation work find the lift moderate whereas people pushing the limits of the model are figuring out that the limits just moved outwards.

Alex Lupsaska has been tracking this limit for a year and a half now. “When GPT5 came out, it was able to reproduce one of my best papers (that took a very long time to come up with) in 30 minutes.”

But Alex also notes that this shift was mostly invisible.

I remember when GPT-5 came out… on Twitter, the reception was lukewarm. A lot of people were like, well, we expected a lot more, and it’s not better at writing email. And I remember thinking, well, okay, GPT-3 could write email. How much better can it get at writing email? That’s not the point. But at the science frontier, the capabilities were really taking off.

We walk through his paper and more with him in today’s Science pod! Watch here.

Read more

“When the question is contested and the cost of being wrong is high, models smooth away the minority views and uncertainty. This is the human advantage.”

Hi all,

Last week, I had been in China with Hannah Petrovic, meeting AI and robotics teams including Zhipu and MiniMax (the two publicly listed foundation model companies), as well as Kimi, Alibaba, Xiaomi, Bytedance and others.

Demand is booming. Zhipu, for example, is serving 5.5 trillion tokens per day. Developers are rushing to the platform, joining at about ten a minute. Keeping up with inference loads is a struggle in China, as it is in the US. Teams universally acknowledged compute constraints, particularly the shortage of Nvidia chips. But that isn’t stopping innovation under those constraints.

Researchers and developers are free to use whichever model they want. Anthropic’s Claude was the preferred choice for technical teams, but it was clear that “dog-fooding” is commonplace.

Read more

“Associating all of distillation with these attacks, which is to date an industry standard for post-training, from open and closed models alike will be a massive own goal.”

‘Distillation attacks’ is a horrible term for what is happening right now. Yes, some Chinese labs are hacking or jailbreaking APIs to attempt to extract more signal from model APIs — stopping this is important to maintain the U.S.’s lead in AI capabilities. Referring to this as distillation attack is going to irrevocably associate all distillation with this behavior, and distillation generally is a core technique needed to diffuse AI capabilities broadly through academic and economic activities.

We went through this sort of language transition with the open source vs open weight debate. All the terms just reduced to open models – very few people in the large AI community know exactly how open-source differs from open-weights. And terminology matters, as the less informed people who still care about — and influence — the technology are bound by different terms they use. If we’re not careful with the discourse around distillation, many people could associate this broad technique used for research and development of new models as an act at the boundary of corporate manipulation and crime.

Read more

Hi all,

As we wrote three months ago, we are in the midst of a compute crunch – demand is running ahead of supply and our position has stayed the same:

The real risk isn’t that we’ve invested too much in AI. It’s that we haven’t invested nearly enough.

Today’s quick look at the data shows that much of the supply remains latent, waiting on enterprise funding. When firms start spending serious money, the compute crunch will get crunchier.

Read more

Get a Team Subscription

The Week in AI

Microsoft and OpenAI restructured their partnership on April 27, ending Azure’s revenue share from OpenAI while OpenAI continues paying Microsoft a capped 20% through 2030; Andy Jassy confirmed OpenAI models are headed to AWS Bedrock, formalizing the multi-cloud era.

Google Cloud unveiled the “Agentic Enterprise” strategy at the hands of CEO Thomas Kurian, repositioning Gemini from a “system of intelligence” to a “system of action” with the new Gemini Enterprise Agent Platform as its anchor.

Ming-Chi Kuo reported on April 27 that OpenAI is building a smartphone, with MediaTek and Qualcomm developing the custom silicon and Luxshare handling assembly — the strongest signal yet that frontier labs intend to own the device, not just the model.

This Week from SAIL Authors

The Compute Cold War

DeepSeek V4 — V4 lands 3-6 months behind the frontier and was still trained on Nvidia, but its post-training and inference are engineered for Huawei Ascend; talent flight, missed monetization, and a “post-DeepSeek era” inside China. — ChinaTalk

No Jensen, Not All Compute is Created Equal — Aggregate FLOPs miss the point: numerical precision, memory bandwidth, and network bandwidth are what separate a Blackwell cluster from a thousand Ascends, and chip policy should be written around the crown jewels, not the headcount. — ChinaTalk

The Quantum Industrial Base — Constanza Vidal Bustamante joins Jordan and Chris Miller to map the cryogenic supply chain, the helium-3 bottleneck, and how export controls on dilution refrigerators backfired — taking China from zero to more cryo suppliers than the rest of the world combined in two years. — ChinaTalk

AI Hits the Physical World

Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition — The CEO and CTO of Applied Intuition on putting AI into mining rigs, drones, trucks, and warships — the most adversarial physical environments imaginable, where simulation gaps get exposed in seconds. — Latent Space

Human drivers keep crashing into Waymos — A review of 78 serious Waymo crashes between August and March: most were other drivers’ fault, and Waymo’s own mistakes were almost always errors of excessive caution — stopping where it shouldn’t, or for too long. — Understanding AI

How Ukraine solved the hardest problem in defense — Cost-per-kill is down from $60,000 to $1,000 because Ukraine’s 500 manufacturers iterate weekly, engineers talk to operators directly, and the procurement loop closes in seven days instead of seven years. — Exponential View

State of the Labs

AIE Europe Debrief + Agent Labs Thesis: Unsupervised Learning x Latent Space Crossover Special (2026) — A crossover episode debriefing AI Engineer Europe and laying out the Agent Labs thesis — recorded just after AIE Europe and just before the Cursor-xAI deal landed. — Latent Space

Full Library

Access the complete, searchable archive of SAIL Media in our Sitemap.

• The Quantum Industrial Base — Jordan Schneider, Chris Miller & Phoebe Chow | ChinaTalk

• How Ukraine solved the hardest problem in defense — Azeem Azhar, Greg Williams & Nathan Warren | Exponential View

• No Jensen, Not All Compute is Created Equal — Aqib Zakaria & Nick Corvino | ChinaTalk

• Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition — Swyx | Latent Space

• DeepSeek V4 — Jordan Schneider, Aqib Zakaria & Irene Zhang | ChinaTalk

• Human drivers keep crashing into Waymos — Timothy B. Lee & Kai Williams | Understanding AI

• AIE Europe Debrief + Agent Labs Thesis: Unsupervised Learning x Latent Space Crossover Special (2026) — Swyx | Latent Space

Get a Team Subscription

“If China invents an ability to take that from 40 hours to 12 hours, you go from one test a week to one test a day. Your iteration cycle changes completely, and they’ll lock that down and grab that supply chain.”

Constanza Vidal Bustamante, Senior Researcher at CNAS and author of the landmark report Quantum’s Industrial Moment, joins ChinaTalk to map out how supply chains behind quantum computers wind through the US and China. Co-hosting are Chris Miller, author of Chip War, and Zachary Yerushalmi.

Our conversation covers:

• What it takes to build a quantum computer — Inside the cryogenic supply chain, the helium-3 bottleneck, and why mining the moon might actually make sense.

• How export controls backfired — How restrictions on dilution refrigerators helped spur China to go from zero to more cryogenic suppliers than the rest of the world combined in just two years.

• The scaling problem — Simply multiplying dilution refrigerators doesn’t get you to a million-qubit machine. Cooling, cabling, and the chips all have to be rethought — and no country owns that yet.

• Why being first isn’t winning — Why long-term victory isn’t cracking Shor’s algorithm first, but locking in supply chains across multiple modalities.

• The public-private fault line — The high-stakes balancing act between the government stepping in to accelerate innovation and letting the market work on its own.

Plus, what China is getting right, where the US still has an edge, whether the US should ban Chinese components, and why quantum supply chains are a national security priority.

Read more

“The $1,000 cost-per-kill came from a feedback loop of engineers talking directly to operators, designs changing every week, five hundred manufacturers running parallel experiments which are treated as data.”

Along Ukraine’s eastern front, a drone stutters mid-flight. On the operator’s screen a few kilometers away, the grainy view of splintered treelines and trenches, fades into static. The control link drops. He removes his headset, he knows what’s happened. Russia has found a way to jam the frequency. Again.

It’s 2024, near Bakmut, a few kilometers from the then-front line. The commander of the Terra drone unit of the 3rd Separate Assault Brigade Mykola Volokhov, has watched the slow degradation of his command links. A system that had worked at eight kilometers out, started to fail at four, then at two. Russian electronic warfare teams were getting better and better with their jamming. Later, speaking on camera, Mykola described the reversal with a soldier’s understatement. “But this problem was overcome… and now it is not with us. No problems!”

Read more

We’ve recently tried to pin down how much compute China actually has, approaching the question from both the supply and demand sides. We converged on roughly 2.5 to 2.8 million H100-equivalents. But a single aggregate figure only captures part of the picture.

Jensen on China

On Dwarkesh’s podcast last week, Jensen Huang argued that China already has enough compute to build frontier AI.

“They manufacture 60% of the world’s mainstream chips, maybe more.”

When Dwarkesh raised the gap in advanced chips, Jensen responded,

“AI is a parallel computing problem, isn’t it? Why can’t they just put 4x, 10x, as many chips together because energy’s free?”

Read more

“Physical machines today are more akin to the state of the phone market before Android and iOS existed.”

From building Applied Intuition from YC-era autonomy tooling into a $15B physical AI company, Qasar Younis and Peter Ludwig have spent the last decade living through the full arc of autonomy: from simulation and data infrastructure for robotaxi companies, to operating systems for safety-critical machines, to deploying AI onto cars, trucks, mining equipment, construction vehicles, agriculture, defense systems, and driverless L4 trucks running in Japan today. They join us to explain why “physical AI” is not just LLMs on wheels, why the real bottleneck is no longer model intelligence but deployment onto constrained hardware, and why the future of autonomy may look less like one-off demos and more like Android for every moving machine.

Read more

“The compute story probably demonstrates that Chinese models like DeepSeek will fall further and further behind Western counterparts.”

Finally, DeepSeek V4 is here. The Pro and Flash models are available through DeepSeek’s website, mobile apps, and API access as of April 23, and the lab has also released its technical report.

Bucking a recent trend of Chinese AI labs moving away from open source, V4 was released under the highly permissive MIT license. It performs admirably on various benchmarks and leads the pack of Chinese open models, but did not close the gap with closed models from the US, with the authors themselves admitting in the paper that V4 is “3 to 6 months behind” state-of-the-art frontier models (though we think it feels further). And as we will discuss later, while its architecture shows progress towards indigenizing the Chinese stack, the model probably still relied on Nvidia GPUs.

Read more

“When Waymo’s vehicles did make mistakes, they were almost always mistakes of excessive caution.”

Last October, Waymo had begun testing its freeway capability, but the company had not yet rolled it out to all vehicles. On a rainy Saturday morning, a routing error caused a Waymo vehicle not qualified for freeway operation to drive onto US 101 just south of the Golden Gate Bridge. Unable to continue, the vehicle stopped in the right lane about 30 meters past the entrance ramp (there was no shoulder).

For the next two minutes and 18 seconds, nothing bad happened. Four vehicles entered US 101 South and routed around the stopped Waymo without incident, according to a Waymo crash report.

But then a white Honda SUV entered the freeway and tried to drive around the Waymo. Unfortunately, the SUV collided with a pickup truck that was driving by in the next lane. The pickup truck lost control, swerved right, crashed through a steel railing, and fell more than 15 feet onto a road below.

Read more

“The general thesis that I have been pursuing now is that the same way that 2025 was a year coding agents, 2026 is coding agents breaking containment to do everything else.”

Checking in a year after the first Unsupervised Learning x Latent Space Crossover special to discuss everything that has changed (there is a lot) in the world of AI. This episode was recorded just after AIE Europe, but before the Cursor-xAI deal.

Unsupervised Learning is a podcast that interviews the sharpest minds in AI about what’s real today, what will be real in the future and what it means for businesses and the world - helping builders, researchers and founders deconstruct and understand the biggest breakthroughs.

Jacob Effron

• LinkedIn: https://www.linkedin.com/in/jacobeffron/

• X: https://x.com/jacobeffron

Full Episode on Their YouTube

Read more

Read more

“The real unlock is not more agents in parallel—it’s better critique loops and spending more on review than generation.”

From near-universal AI tool adoption inside Shopify to internal systems for ML experimentation, auto-research, customer simulation, and ultra-low-latency search, Mikhail Parakhin joins us for a deep dive into what it actually looks like when a 20-year-old, $200B software company goes all-in on AI. We cover why Shopify has become much more vocal about its internal stack, what changed after the December model-quality inflection, and why the real bottleneck in AI coding is no longer generation, but review, CI/CD, and deployment stability.

Read more

“Without access to foreign tools the US could control, Chinese chip and memory makers would not be in a position to even produce the meager amounts they can today.”

Some notes of what struck me most from the instant classic of a pod.

Are AI chip exports a red herring?

The Trump administration has agency over the two variables most relevant to whether China will have enough compute to really compete with the US: how many chips they can make and how many chips they can buy. But for all the drama we’ve had this administration around whether Trump will allow Jensen to sell chips to China, we’ve had basically zero movement on the tooling side. Without access to foreign tools the US could control, Chinese chip and memory makers would not be in a position to even produce the meager amounts they can today.

This administration teased controls on sub-systems in Trump’s July 2025 AI Action Plan, but absent the headfake around the Affiliates Rule that was wound down after Beijing escalated on rare earths, we’ve had zero movement to close loopholes.

Read more

“If this direct data access becomes the next frontier of training, open models in their current form will be left behind.”

It’s a clear, current equilibrium that open models will be in perpetual catch-up of closed models, but this gap being viewed as a single number, a “distance”, covers up a nuanced and crucial dynamic at what capabilities the models are covering. The most popular benchmark to comment on this gap is the Artificial Analysis Intelligence Index — a composite benchmark of ~10 sub-evals that they maintain over time to capture the “frontier” of current language model capabilities.

Particularly, I spend a lot of time understanding how dynamics that feed into that index are misunderstood by the natural tendency to reduce performance and trends to one number. Examples include:

• How benchmarks evolve over time, becoming more or less correlated with how people actually use models,

• How different models’ real-world performance relates to their benchmark rankings, and

• How training regimes evolve over time to move said benchmarks.

Read more

“With respect to GaN power semiconductors, the U.S. has already lost its lead and is at risk of being pushed out altogether.”

In the semiconductor industry, the Trump administration is striving to bring back critical technologies that slipped out of our hands decades ago. The U.S. has attracted billions of dollars in investment to stimulate cutting-edge logic manufacturing, the development of EUV lithography, and HBM production. However, the semiconductor ecosystem is a lot more than just AI chips. And if the administration wants secure supply chains, it should focus on another rising material: gallium.

Just as Pluto is technically not a planet, gallium is technically not a rare-earth element despite often being discussed in the same context. Like many rare earths, gallium is not directly mined from the Earth’s crust but rather a byproduct of aluminum extraction. Although not classified as a rare earth, the mineral plays a major role in compound semiconductors and has critical importance for the future of AI, defense, robotics, and more.

Read more

“From my research into the guardrails thesis, plus other data points gathered on the fast pace of agent adoption at large especially in software engineering, it is clear that we are kind of winging it and roughing it head first into the AI age.”

Welcome to Interconnected Capital’s Q1 2026 performance update letter.

To new and old readers alike, a friendly reminder: I run a global technology long-only fund focused on investing in both the hardware and software “picks and shovels” of the interconnected global digital AI economy. I draw on my technology business operator’s experience and geopolitical antennas to bring an edge to how I assess a tech company’s rhythm and prospects in a constantly changing world. [1]

As always, first the numbers, then the reflections.

[1] My past experiences include: senior leadership position at GitHub (the world’s largest developer and open source technology platform, now owned by Microsoft), a unicorn database startup, early stage VC, and the White House and Department of Commerce during the Obama administration. I studied law and computer science at Stanford; international relations at Brown.

[2] Includes January 1 - March 31, 2026, gross returns. Unaudited.

[3] Includes January 1 - December 31, 2025, gross returns. Audit in progress.

Portfolio positions in random order (as of March 31, 2026):

• CIENA CORP

• NEBIUS GROUP NV

• APPLIED MATERIALS INC

• OUSTER INC

• WERIDE INC ADR

• ARM HOLDINGS PLC-ADR

• INTEL CORP

So What Happened in Q1?

Read more

This post originally appeared in Understanding AI.

“I am placing Meta in that category of AI labs whose pronouncements about model capabilities are not to be trusted, that cannot be relied upon to follow industry norms, and which are clearly not on the frontier.”

The April 8 release of Meta’s new model Muse Spark got overshadowed by Claude Mythos Preview, which was announced one day earlier. But Meta’s new model family — and the 158-page safety report Meta released about it last week — are still significant for what they tell us about the company’s future role in the AI industry.

Mark Zuckerberg spent billions of dollars to assemble the team that built Muse Spark. The model’s release gives us our first hints about whether Meta will be able to break into the top tier of AI labs.

Meta has all of the advantages of a well-resourced technology company: lots of AI chips, proprietary data, and lavish salaries. Those resources have enabled the Meta team to produce a model with strong benchmark scores. But I suspect that those scores still overstate the model’s real-world utility.

Read more

“With AI on our doorstep, quantum represents the single biggest lever we have to pull as a society for the next couple of decades.”

What exactly is quantum computing? Why does it matter, and what would it actually mean to “win” the quantum race? Zach Yerushalmi, CEO of Elevate Quantum, a Mountain West–based public-private consortium advancing the U.S. quantum ecosystem, and Chris Miller join the podcast to discuss.

Our conversation covers…

• What Quantum Computing Actually Is — A primer on qubits, superposition, and why quantum computers aren’t “faster classical machines” but fundamentally different systems designed to simulate nature and solve specific classes of problems.

• Why Quantum Matters Now — Breakthroughs in error correction and hardware have shifted quantum from theory to an engineering race, with major implications for drug discovery, materials science, artificial intelligence, and cybersecurity.

• The Economic and National Security Stakes — Quantum’s potential impact on cryptography, advanced manufacturing, biotech, and defense makes it a strategic technology with an extremely small margin for error in global competition.

• From Science Project to Industrial Policy Challenge — The bottleneck is no longer just physics but scaling. Talent pipelines, fabrication capacity, supply chains, and the kinds of public-private partnerships needed to move from lab prototypes to deployable systems.

• What Winning Looks Like — Leadership isn’t just building the first powerful machine. It’s shaping standards, securing supply chains, protecting encryption, diffusing capabilities across industry, and sustaining innovation in a tight U.S.–China technological race.

Read more