klein

curious about everything.
building

[klein @moretti__klein]

moat might be amongst the most used words on X

[klein @moretti__klein]

@shoucccc its the infinite reselling loop

[klein @moretti__klein]

@MGH path of exile 2 if you are into that genre

[klein @moretti__klein]

@leostera just what @tsoding is looking for

[klein @moretti__klein]

man believes tools like these are accurate

[Rhys @RhysSullivan]

the fully ai generated comment saying that they deliver value is cracking me up

[klein @moretti__klein]

@theo would you be open to exposing something like a t3 server, for people to build their own ui (web / gui / tui) on top of it?

[klein @moretti__klein]

@KalshiFC mf let out a bark like a dog sighing 😭

[klein @moretti__klein]

mhm

[Cohere @cohere]

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: https://t.co/TZSugXi2ZE

[klein @moretti__klein]

never understood why people come up with weird ways to make walls of text from ai readable, drawing diagrams, html artifacts, and now this. lets be real, if you didn't bother to read the 1st time, all you are gonna do is glance at it the 2nd time, assure yourself that you understand, and move on.

dense information is dense for a reason. either give in to the slop completely, or stop larping

[vogel @ryanvogel]

i started having my agents print off technical (nasa like) review sheets for when they have stuff ready for me

i also got some newsprint and it feels so authentic and good to use

[klein @moretti__klein]

@b1rkh0ff lets do it

[klein @moretti__klein]

@spacedevin https://t.co/28NkPijnDD

[klein @moretti__klein]

@raunofreiberg yep, finally decided to take the shot, more to come

[klein @moretti__klein]

finally have paying customers for https://t.co/28NkPijnDD

waited a day after the first subscription just to make sure they don't cancel immediately :p

[klein @moretti__klein]

> marked as "free to be written to" by the OS

this is what delete does. marks that block of memory free to be used by anything else. it doesnt "reset" data in that block, because that would essentially mean writing data equivalent to that file size. imagine deleting a 50gb file and you have to wait like 10mins

[klein @moretti__klein]

@thorstenball i mean t is already taken, you should drop it and go nuclear

[klein @moretti__klein]

my agent creates branches on its own

mf no one is ever going to work on this except me, push to main and ship it

[Vic 📠 @VicVijayakumar]

If your agent is not directly yeeting code into main, you’re behind

[klein @moretti__klein]

@AidenAllred2 i'd suggest use opus with @impeccable_ai

ask it to suggest a few UIs with differed tastes, mix and match

[klein @moretti__klein]

made opus redesign https://t.co/28NkPijnDD with @impeccable_ai and the redesign is sooooo on point. huge props 👏

[klein @moretti__klein]

i swear all AI generated UI look the same

[Armin Ronacher ⇌ @mitsuhiko]

The poll so far between Twitter, Bluesky and Mastodon. I think it looks less absurd than last time, but still pretty clear that the base is divided. https://t.co/9PtrCLGrbB

[klein @moretti__klein]

I honestly dont get the hype around @bendlang

If I want to formally verify that my code works as I intend it to, I already have tests. Bend laws just seem to be a natural language way of writing tests.

Honestly feel like I am missing something.