Tweet fixado
Carlos Garrido 🍿
1.3K posts
Carlos Garrido 🍿
@shouldomythesis
Drinking coffee. One deadline at a time. Lacking coherence
Lisbon, Portugal Entrou em Aralık 2015
2.1K Seguindo215 Seguidores
English
I want to give you a full set of @Stanford course materials on how to measure AI.
I'm on the teaching team for a new class: CS 321M - AI Measurement Science.
Comment or reshare and I'll send you everything: slides, readings, textbook.
Why does this class exist? Because most AI benchmarks are already broken. Saturated, partly memorized, or disconnected from real use.
Yet we still use them to pick models, justify budgets, and shape policy. AI moves fast and measurement hasn't kept up.
This class covers what goes wrong with benchmarks, how to design evaluations that hold up, and what valid measurement looks like.
If we can't measure AI we can't improve it. I think this material matters way beyond one classroom. Help me get it out there.
English
@Dorialexander @V4ldeLund Something similar to a variational auto encoder but for math problem space? Or do you recon this problem space isn't compatible with this genre of architectures? The paper by @ClementBonnet16 comes to mind:
arxiv.org/abs/2411.08706
English
@V4ldeLund No don’t really exist yet. My immediate problem right now is math problem discovery but most synth pipelines are broadly combining constraints/digging through a search space and we don’t have the right language for it yet.
English
Starting to suspect most synthetic pipelines will be about topology long term (first with math, but not only).
English
@TheAhmadOsman @CryptoElite007 Exited to fill the form 👀✅✅✅
English
@CryptoElite007 Any, as long as you take a screenshot for proof for the form you’ll receive later
x.com/theahmadosman/…
Ahmad@TheAhmadOsman
RTX PRO 6000 (96GB VRAM, ~$15K) GIVEAWAY FAQ Q: Cost to enter? A: $0. Free. Q: Do I have to register for GTC? A: Yes, virtual attendance is COMPLETELY FREE Q: Where do I enter? A: Tap the link in my bio, there’s a clear button on the page Q: How do I increase my chances? A: Earn bonus entries: • +150 for signing up for GTC 2026 • +75 per referral when someone uses your code • Follow / subscribe on socials for extra entries Q: Is this officially sponsored? A: Yes, sponsored by NVIDIA Q: When do entries close? A: March 19 Q: What happens after I enter? A: After GTC, you’ll receive a form by email Q: What do I need to submit? A: Proof of attendance: • Virtual → screenshot • In-person → selfie at GTC Q: When is the proof deadline? A: April 1 (preliminary date, may change based on response rates) Multiple reminders will be sent Q: How is the winner chosen? A: Random draw among verified entries Q: When is the winner announced? A: TBD I need time to verify all valid submissions Depends on verification volume Q: When does the GPU ship? A: TBD Q: Where will updates be posted? A: Email + my socials Q: Didn’t get the verification email? A: Scroll down and hit “Submit” on the Giveaway Entry page Q: Are there location restrictions? A: No, there was a bug, now fixed. Try again Q: Who can enter? A: Anyone who can attend GTC and provide valid proof Q: Is registering enough? A: No, you must attend and submit proof Q: Do I need to watch sessions or just register? A: You must attend and provide proof Q: Do I need to attend live? A: Yes, you must attend live and provide proof Replay views don’t qualify Q: Is registering enough? A: No, you must attend and submit proof
English
started from the basement now we’re here
Ahmad@TheAhmadOsman
LocalLLaMA Home Server Final Boss 😎
English
Carlos Garrido 🍿 retweetou
Prover correctness is becoming a central question as AI enters mathematics and software verification. New essay on why Lean's architecture is designed to survive AI pressure.
leodemoura.github.io/blog/2026-3-16…
English
@3lfares Only had it in tea xD learning new things every day 🙂
English
@shouldomythesis Bro kinda disappointed that someone with the name carlos from portugal is unaware of this basic fact
English
The ultimate combo:
Watermelon, mint and white cheese 😋🍉
English
YouTube
QME
@SAIRfoundation @grok what's the original link of the video
English
OK, THERE'S NO FIRE DOG THAT STANDS UP AND IT'S GONNA STAY THAT WAY, RIGHT?!, RIGHT??!!!
MAGIKARP🦈💦@UniteVids
🔥 Fire pups 🔥
English
@Afinetheorem Tablets, computers and digital education?
lectera.com/magazine/artic…
English
I didn't realize this: Finland, which used to be the absolute star in the West on education, is now on math roughly at the OECD average, only a bit higher than the US (meaning way behind New England), and fell 60 pts in 20 years, worst in the world. What happened?!
English
@Leonard41111588 Great article :)
"What would a verification platform for the AI era require? A small, trusted kernel"
Would love to hear your thoughts about how trust is built checking this kernel and also how trust builds around the kernel!
x.com/JFPuget/status…
JFPuget 🇺🇦🇨🇦🇬🇱@JFPuget
How is Lean code proved correct? Is Lean written in Lean? It all stemmed form a wild thought I had: what is there is a bug in Lean? How would it impact al the proofs created with lean? Reason for the above is I never saw a bug free software.
English
AI is writing a growing share of the world's software. No one is formally verifying any of it.
New essay: "When AI Writes the World's Software, Who Verifies It?"
leodemoura.github.io/blog/2026/02/2…
English
@Alonso_GD Hello, I'm not a lawyer but I enjoyed this conversation between 2 lawyers on a similar topic: pca.st/episode/b06eea…
Maybe in some scenarios, with the required guardrails, it could be helpful 🤔
English
@moonstar24689 @RepThomasMassie That was likely Martin Nowak and is mostly unredacted already justice.gov/epstein/files/…
English
@RepThomasMassie I don’t have a screenshot but the one that says “what a night, glad I didn’t kill anyone” or something along those lines. Hopefully someone has it and will post
English
Tomorrow I will go to DOJ to view the unredacted Epstein files.
Which docs should I view?
Include EFTA link in reply
I will sort the replies to this post by “number of likes,” so instead of making redundant posts, please “like” replies that contain docs you think are important.
English
Carlos Garrido 🍿 retweetou
#Tempo: De 2 a 8/Fev prevemos uma semana muito chuvosa, na generalidade do país, especialmente no Norte e Centro. Episódios de trovoada com possível granizo. Esperada neve na Estrela e extremo Norte. Vento moderado a forte e agitação marítima forte👉 tinyurl.com/3jbfes7x
Português
@tufpraise Good age for the marathon if that's your jam 😅
English
30 is old only if you are trying to become an athlete.
𝙍𝙞𝙘𝙠 🥊@RickCombatTV
The older i become the more i realise how young 30 is
English
@Dorialexander This might be just anecdotal evidence, but I've been somewhat successful in doing simple scrips with Gemini. I like the explanations and it's easy to follow and debug. Never tried opus, but 5.2 sometimes overcomplicates v0
English
am I the only one to see a big gemini problem rn? just had an eval for a conversational use case on flash and results are plain bad, significantly worse than self-hosted gemma.
Alexander Doria@Dorialexander
sorry to say that gemini looks previous generation llm vs. opus/5.2
English
@ChShersh Escaping poverty sounds like a Great plan tho...
English
So, is anyone still interested in becoming a better SWE?
Or is everyone obsessed with AI agent orchestrators to build and sell your next SaaS and escape poverty as soon as possible?
English
@Aaronontheweb I feel like I'm hearing am episode of the vergecast through this comment.... The dead internet theory coming to fruition:( @reckless
English
@MKBHD I fell like it really depends if those 24 reviews are from a trusted reviewer or just random users. Also depends on the 24 reviews themselves, especially the worse ones. Any common failure modes? Common complaints? Etc.
English
You're about to spend $100 on a product you've never tried. Which version do you pick?
English
@yacineMTB @yacinelearning you don't get it. I have over half a year played
English
believe it or not this one little game would butterfly-wing-flap it’s way into a high performance reinforcement learning library with a puffer fish as logo
2001 Live@25YearsAgoLive
“RuneScape,” a new MMORPG, is launched.
English