TLDR 2026-08-27
Nvidia buys Hugging Face π°, Inside OpenAI π€, making LLMs faster β‘
Nvidia in Talks to Buy AI Startup Hugging Face (2 minute read)
Nvidia is close to finalizing a deal to acquire Hugging Face at a value of roughly $13 billion. Hugging Face makes AI software and hosts it for other companies. Nvidia was already a backer of Hugging Face, which was valued at $4.5 billion in a funding round three years ago.
Inside OpenAI's Reboot (26 minute read)
OpenAI has suffered a series of setbacks, including a spate of leadership departures and several lawsuits. It has also suffered an erosion of public trust. However, executives within OpenAI are painting a more upbeat picture. The company remains in an enviable position, and ChatGPT is still one of the most popular AI products in the world.
π
Science & Futuristic Technology
FDA Approves Drug for Pancreatic Cancer in Milestone for Treatment of Deadly Disease (6 minute read)
Researchers have discovered a drug that disables a stubborn mutation that made pancreatic cancer a death sentence. Rasonque helped subjects in a late-stage study live more than 13 months, nearly twice as long as those on chemotherapy alone. It has side effects, including a harsh rash across the body and gastrointestinal issues, and costs more than $477,000 a year. The drug could generate more than $20 billion in annual sales.
Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha model (1 minute read)
Ox Alpha was created by Z.ai, which also made GLM. The company has confirmed that Ox Alpha is the newest iteration of its GLM series. It will open-source the weights so developers can build on top of it. Z.ai says Ox Alpha is a reasoning model designed for coding, sustained agentic work, and production workloads.
π»
Programming, Design & Data Science
What breaks when your agent has 100,000 tools (Sponsor)
Most AI agents demo well and fall apart in production. Viktor's team wrote up what breaks at scale. Viktor is an AI employee in Slack and Microsoft Teams, working across 3,200+ tools. 60,000+ teams use it.
Start free with up to $100 in credits
How to Make LLMs 3X Faster (18 minute read)
Speculative decoding is a technique that converts a GPU's unused math units into output. It involves using a second, much smaller model that produces several candidate tokens in advance. The large model evaluates all of them in a single forward pass. This results in up to three times faster generation.
I had some free time, so I tried to pwn V8 (19 minute read)
This developer chained three public V8 bugs against the exact Chrome build used in Google's v8CTF. The first bug leaked a compressed object address, the second turned a garbage collection mistake into a fake JavaScript array and read/write access inside the V8 cage, and the third used a JSPI and JS Dispatch Table mismatch to pivot the native stack outside that cage. The developer then reused code already inside Chrome to open, read, and print the flag.
Get the most interesting stories in startups, tech, and programming delivered in a free daily email.
Join 1,600,000 readers for
one daily email