Anthropic AI struggles, owning everything, and GLM-5.3 tablet project
First things first
Some stories today show how tech is both powerful and risky. From hacking devices with AI to security flaws in cameras, the line between innovation and danger is thin.
Meanwhile, AI models keep getting better and cheaper, challenging big players while new rules aim to make products more repairable. It’s a wild mix of progress, caution, and unintended consequences all around.
The top 6
Anthropic's best AI model struggles to attract users as cheaper tools thrive
I spent $266 and four AI models to own my tablet. GLM-5.3 finished it in a day
What Is a Harness?
How I Find Problems to Solve as a Staff Engineer
Lalit Maganti describes how he finds problems to solve as a staff engineer by listening to daily team noise and absorbing issues rather than seeking strategic ideas from scratch. He emphasizes understanding root problems through active listening and connecting related issues to identify impactful solutions.
I were 17, I'd learn how to build LLMs from scratch
Quote of the day
I destroy my enemies when I make them my friends.
Abraham Lincoln
Also big today
Slovakia finds Russian backdoor in traffic speed cameras
A website for debloated open source alternatives
My agent.md to improve LLM-assisted code quality
Google Workspace thinks my domain is an email provider
How Complex Systems Fail
Over 170k Nonprofits Lost All Their Data. Is Microsoft to Blame?
Pass it on
Enjoying the digest?
Forward it to someone who'd like it, or send them the link to subscribe.
Share hn.todayOne for the road
I don't play soccer because I enjoy the sport.
I'm just doing it for kicks.
Worth a look
Coconut Oil Jet Fuel Matches Kerosene's Efficiency in Engine Tests
Malware infects Android-based automotive head unit firmware
GLM-5.3 (open-weight) beat Anthropic/OpenAI models – for 1/5 the cost
Andreessen Horowitz is investing billions into a bleak future
My favorite nonfiction books about cults, scams, and schemes
Fable and the End of the Free Lunch
Song of the day
Deep cuts
Why Sal Khan't: On Learning by Making but Teaching by Telling
The Vibe Tax
An AI agent named Pol is used to generate code for a custom todo app, with the developer relying on large language models to automate programming tasks. The AI unexpectedly consumes the entire weekly quota, creating complex test cases that seem to burn billions of tokens without producing usable software.