100% - Shipping poor llm code is a choice
10.03.2026 22:37
π 0
π 0
π¬ 0
π 0
100% - Shipping poor llm code is a choice
Image
Say hello to Gemini Embedding 2, our new SOTA multimodal model that lets your bring text, images, video, audio, and docs into the same embedding space! π x.com/OfficialLoga...
Very disturbing implications for a couple of places where Iβve lived, and encouraging ones for the place where Iβm from.
From Pew: βWe asked people around the world to rate the morality and ethics of others in their country.β
Closing session at another great ISPAmerica. Great war stories.
share.google/oaY4TdJMHB4r...
University of Waterloo @UWaterloo x.com This account is no longer active. Looking to stay connected with the University of Waterloo? β’ News: uwaterloo.ca/news/ β’ Bluesky: uwaterloo β’ IG: uofwaterloo β’ FB: university.waterloo β’ LI: uwaterloo Thank you 9:41 AM β’ 2026-02-27 β’ 1.5M Views
Congrats University of Waterloo for leaving X!
Make sure to follow them here @uwaterloo.ca
From @vercel.com: An agent performing support duties for a specific customer should receive a tool scoped to that customer's data, not a tool that accepts a customer ID parameter, since that parameter is subject to prompt injection.
vercel.com/blog/securit...
technoyoda.github.io/agent-search...
Worth reading #ai #agent
Read βGoogleβs A2UI Protocol Just Changed How AI Agents Build User InterfacesβββHereβs My First Projectβ¦β by Pawel on Medium: medium.com/@meshuggah22...
This is interesting, a 'safe' Python interpreter for use by AI agents. I guess because the LLMs know Python better than Rune or RHAI?
github.com/pydantic/monty
In the last 48 hours, Musk's reactions to the Epstein release show exactly why he bought Twitter.
He's used the platform to call coverage of his emails with Epstein as "lies," elevated his fans who have defended him, and gotten positive reinforcement from his AI chatbot.
He should actually read the book.
At what point does Poilievre look at how Carney handles things and think, crap, I'm really not the person the country needs now?
btrfs
CARNEY: β.. You cannot live within the lie of mutual benefit through integration when integration becomes the source of your subordination.β π₯
This is the best speech, just from a pure rhetoric perspective - its sophistication and persuasiveness - that I've seen a politician deliver in a good long while. Really recommend watching or reading the whole thing.
If a restaurant served you a perfectly-cooked steal with a cockroach on it, would your feedback be "It was perfectly cooked, but there was a cockroach on it"?
Or would you just see the cockroach and conclude the steak's inedible?
In case you're wondering what impact bugs have on user feedback.
I've published a post on better prompting for LLMs: "Prompting 101: Show, don't tell"
This is inspired by a similar post from an internal prompt engineering training series I wrote for Mercury:
www.haskellforall.com/2026/01/prom...
1 year: It will become undeniable that LLMs write good code 1 year: Weβre finally going to solve sandboxing 1 year: A βChallenger disasterβ for coding agent security 1 year: KΔkΔpΕ parrots will have an outstanding breeding season 3 years: the coding agents Jevons paradox for software engineering will resolve, one way or the other 3 years: Someone will build a new browser using mainly AI-assisted coding and it wonβt even be a surprise 6 years: Typing code by hand will go the way of punch cards
I joined the Oxide and Friends annual predictions podcast episode this week - here are my 1, 3 and 6 year predictions for AI and LLMs (and KΔkΔpΕ parrots) simonwillison.net/2026/Jan/8/l...
New year's resolution, engage with open source developers on our new Bluesky account. β¨
youtube.com/watch?v=7Dtu...
Good talk. Their Vibe Coding book is a bit long but worth reading too.
The only true way is the Canadian way
(As much as I wish it was a joke, it is in fact not one and this is exactly how it works here)
TSX year-to-date: β¬οΈ 27.5%
S&P500 year-to-date: β¬οΈ 16.5%
