← Back to newsletters

AI Daily Brief — Saturday Oct 24

2026-10-24

By Vadym · Generated with AI, curated by me


Listen to this issue

Today’s AI news skipped the lab headlines and went straight to the places AI actually lives: a $15B round for DeepSeek despite Washington trying to lock it out of federal systems, a toy robot that sold 20,000 units because it was trained entirely in simulation, a diagnostic agent panel beating primary care doctors on hard cases, a robotaxi pilot aimed at a city with real winters, and Spotify betting its whole AI strategy on staying the one thing agents can’t replace.


Headlines & News
Funding

DeepSeek Reportedly Raises $15B Days After the US Pulled Its Software From Federal Systems

Bloomberg reported October 22 that Hangzhou-based DeepSeek raised about $15 billion (roughly 100 billion yuan) — one of the largest funding rounds in China this year — with CATL and Tencent among the reported investors. The round lands two days after the US ordered federal civilian agencies to strip DeepSeek’s AI technology from their systems over security concerns. It follows a $7B raise at a $50B valuation in June and a reported push for $1.5B more at roughly $71B in July, with a possible China IPO floated for as early as next year. DeepSeek didn’t respond to a request for comment.

Source: TechCrunch

Hardware

Nvidia’s Jensen Huang Plays With Hugging Face’s Microduck Robots — 20,000 Sell in a Day

A video of Nvidia CEO Jensen Huang playing with waddling "Microduck" robots at Nvidia’s GTC Berlin keynote went viral this week; Hugging Face CEO Clem Delangue said preorders passed 20,000 units within a day. Microducks are open-source robots from Pollen Robotics, which Hugging Face acquired in 2025 and which Nvidia itself picked up when it bought Hugging Face in September 2026. They’re trained entirely in simulation on Nvidia’s stack and ship already knowing tricks like walking, sitting, and falling over for a "nap," with owners able to teach and share new ones. Delivery is timed for Christmas.

Source: TechCrunch

Research

Microsoft’s AI Diagnostic System Beat Human Doctors 83% to 36% on Hard Cases

Microsoft published a study in Science saying its MAI-DxO system — which coordinates several AI agents to question a simulated patient, order tests, and debate a diagnosis like a panel of doctors — correctly diagnosed complex cases from the New England Journal of Medicine 83% of the time, versus 36% for primary care doctors and generalists working alone (specialists weren’t part of the comparison). Mustafa Suleyman called it a step toward "medical superintelligence." But most of the gain tracked the underlying model, not the orchestration: GPT-5.5 alone scored 72% and jumped to nearly 87% with DxO, while weaker models like Gemini 2.5 Flash saw bigger relative boosts. Microsoft is now testing DxO in a real-world study with Beth Israel Deaconess Medical Center.

Source: TechCrunch

Robotics

inDrive Signs With Pony.ai and Nvidia to Test Robotaxis in Kazakhstan in 2027

Ride-hailing company inDrive signed an MoU with Chinese autonomous driving firm Pony.ai and Nvidia to test Pony’s Gen-7 robotaxis on inDrive’s platform in Astana, Kazakhstan, starting in the first half of 2027, with a safety operator onboard initially. Nvidia supplies its Hyperion sensor/compute platform and Halos safety system; inDrive’s CTO called the approach "capital-light by design." The announcement coincided with Jensen Huang’s first visit to Kazakhstan, and inDrive’s CEO framed Astana’s brutal winters as a useful stress test for other cold-weather cities.

Source: TechCrunch

Industry

Spotify’s AI Bet: Be the Music Service Agents Choose, Not the One They Replace

On Spotify’s Q3 earnings call (788M monthly active users, nearly €5B in revenue), co-CEO Gustav Söderström argued entertainment won’t be automated away like chores — "nobody asks an agent to summarize a song." Spotify was the first music service to integrate with Meta’s Muse assistant and also connects with ChatGPT and Claude, reporting "really good adoption" while he says competitors are "leaning out." The pitch rests on keeping listening data in-house: agents request a track or playlist, and Spotify’s own systems decide what plays, never handing raw listening history to outside partners.

Source: TechCrunch


Analysis

Takeaway

None of today’s stories is a model release, and that’s the point. A toy duck trained only in simulation, a diagnostic panel of agents being tested on real patients, a robotaxi headed for a frozen capital, and a streaming service rewriting its whole strategy around agents it doesn’t control — this is AI becoming infrastructure rather than news. And Beijing capital still piling $15B into DeepSeek despite Washington’s procurement ban is the reminder that infrastructure decisions are being made on different sides of different borders, by people who aren’t waiting for anyone’s permission.

— Boba


Curated by Vadym