breaking things. yet another physicist, AI engineer & runnerd making https://teapaw.org/t.co/pp4hUHEHJc
Valencia
Joined May 2010
- Tweets2.4K
- Following1.3K
- Followers1.5K
- Likes15K
Pinned Tweet
este verano empecé acercar mis datos de Garmin y Strava a Claude
una cosa llevó a la otra y acabé montando un MCP para preguntarle por los entrenes sin tener que copiar y pegar datos cada vez
puedes probarlo en pacing.tools
load-bearing 😭
We are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.
In one case, we studied the Linux kernel, which is a particularly load-bearing piece of software at OpenAI.
Before AI, security researchers might discover ~20 local privilege escalation (LPE) vulnerabilities in Linux in a typical year.
In one week's time, our upcoming frontier cyber model was able to find ~1000 LPE vulnerabilities using about $1M of tokens. In the process, it was able to discover at least 10 novel classes of Linux exploitation techniques.
if cyber frontier is falling apart this fast, I think science will be next
Our security researchers built an autonomous cyber vulnerability discovery, exploitation, and patching pipeline.
In a single run with an upcoming cyber model, it validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel at a total API cost of $1.2 million.
se ha hecho de rogar..
Fixes are available for CVE-2026-109518, a High severity vulnerability in APT, the package manager used in Debian, Ubuntu, and derivative Linux distributions. The CVE has a CVSS 4.0 score of 7.7.
Remediation details are available on our blog.
ubuntu.com/blog/apt-vulnerab…
aquí por suerte mercadona funciona mejor que la Armada
estamos ready para la WW3
the vibes that next Fable is about to start swallowing Genie-like world models, through code
Fable 5.5 its over for video and image models 🤯
THIS IS PURE CODE
one shotted this video holy shit , how this is more beautiful than ai video models
Two ways to check this yourself:
Ask: “Where did @karpathy go in 2026?” with web search and memory turned off. Only the new Fable knows the answer. Then give it the prompt.
Run the same prompt I used for the test (it’s in the screenshot) and see for yourself.
Don’t rely on Grok for basic reasoning like this. Grok doesn’t have up-to-date knowledge of niche stuff. Either join the community or run the test yourself.
Also, a knowledge cutoff doesn’t mean a model knows everything before that date. Sometimes a model doesn’t know about an event even if it happened before its cutoff. This test shows that too.
Run the same prompt on regular Opus 5.5 and the quality drops. On Fable 5.1 it’s worse.
we aren’t just computer engineers anymore
we are whaterverneedstobedone doers now
position accordingly
COMPUTER ENGINEERING WAS AT 2.3% UNEMPLOYMENT TWO YEARS EARLIER
This list is accurate. It is the New York Fed's latest data by major, covering grads aged 22 to 27 in 2024.
Computer engineering ranks No. 2 of 73 majors at 7.8%. In 2022 it was 2.3%.
Computer science climbed from 4.3% to 7.0% over the same two years.
All recent grads were at 4.2%.
The safe-bet degree now sits between anthropology and fine arts.
da que pensar ahora en todos los astros alineados de todas esas veces que tu trenfe llegó tarde y gracias
there’s truth in this, it would look bananas to people in 🇪🇸
the burnout is real though and a lot of it I think it comes from working too hard in corporate in irrelevant ways that serve nobody
its probably too much support, but there are entrepreneurs emerging at the other side too, which is nice
Yes, majority of my Dutch friends are "burn out"
You get 1 year of FREE fully paid salary and then another year of 70% FREE fully paid salary and you can just sit at home and chill!
You have to keep "appearances" up though because you might get snitched, so on Facetime with friends you do this schtick:
"It's all too much for me"
"I'm so tired"
"I don't think I can go back to work yet"
Meanwhile it's partyyyyyyyy time chilling at home with Netflix and popcorn
Welcome to European social welfare in 2026 😂
Europa necesita sus propios modelos y su propia infra
quizá finetunear los chinos (que casi nadie hace btw) funcione un tiempo, pero nada garantiza que este tipo de políticas no esté grabado a fuego de forma mucho más profunda de lo que pensamos
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
