Vlad Iliescu

2. But Anthropic researchers say they see patterns that suggest that Claude can. The employees at Anthropic are telling themselves stories based on the words they input into the software and the words that come back out. >>

There is an argument that some AI-psychosis-sufferers tend to make which I find very gross, stupid, and bad. It goes something like, “the way you are talking about Claude sounds a lot like how they used to talk about [some historically marginalized group], sooo, you better think about that”.

This is the funniest thing I've seen today, possibly ever

cool cool cool cool

Like there are companies that handle dangerous materials. This isn’t a new concept. When you are one of those, you are supposed to be careful.

Embarrassed by the baseless speculation I was spreading nine days ago. Should have waited for the base.

I like this quote on code review: "The idea that humans will reliably look through code to find the increasingly rare issues that automated tools miss seems like a fantasy." By Marc Brooker, AWS.

There was once a time when computer viruses were widely considered to be a myth by sysadmins.

I think a much more urgent and credible threat to either of these, fwiw, is often obfuscated by this discussion: namely *human-directed* hacks using open-weight models.

Another important post you should read from @masnick.com. TLDR: Age verification is bad. Longer TLDR: Avoid having your ID scanned for any reason. www.techdirt.com/2026/09/03/h...

RIP to everyone driven insane contemplating the basilisk for their hubris but im different. and smarter. maybe even smarter than the basilisk

⚠️ A few hours ago, a malicious crate was discovered on crates.​io which spread as a dependency of `arrayref` and some other crates, likely due to compromised credentials. The affected versions have been deleted. For details and how to see if you are impacted, see: blog.rust-lang.org/2026/08/20/s...

gf: "i'm breaking up with you. your fetish for ai cofounder roleplay is out of control" me, under my breath: "directionally very bad"

the risk I think we're sleeping on is that we wirehead ourselves into a "La Belle Dame" state by using coding agents to build increasingly addictive videogames, or inventing conversational partners that people genuinely prefer to other humans

Qwen: boobs sex porn fucking cock!!! Gemma: I want to be the pretty and loved

and THAT's why i need to install a new browser

🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️ One milkshake to rule the boys, one milkshake to find them, one milkshake to bring the boys and in my yard to bind them 🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️🕯️

Nuclear opinion: I don’t care if you use AI to write code,¹ but making a human review a change before an LLM is disrespectful. Why am I catching your bugs if a computer could have done it? It’s like asking to review code that doesn’t compile or pass CI. ¹ As long as you do it responsibly.

“The tooling as it exists today promotes absolutely nothing like the focus required to apply it judiciously.” thoughts.hmmz.org/2026-05-31.h...

Some quick vibe checks on opus 4.8 are not looking good, not good at all.

It makes many online spaces intolerable. If I want to talk to ChatGPT or Claude, I'll just talk to ChatGPT or Claude, I don't need to talk to Claude pretending to be a person with mediocre prompts. You can use AI to help with writing, but you need to actually do some writing to be helped with.

Using llms to help you “express your true self” is like using an electric bike to show just how fit you are.

Oh god the discussion in Vancouver is going to be so stupid from all sides. “11 AI agents” is meaningless as a phrase. If I said “I have 11 spreadsheets” or “I have 11 browser tabs” to do my work, it means about the same thing.

In fact, when they take a closer look at how people in the AI group used AI, they find that there is a segment of people who did no worse—and maybe even better—than they might have without the AI: the people who used the AI to ask for hints, rather than simply asking for the solution directly.

there is only one correct way to redesign GitHub and this is it

Dependency cooldowns but for VS Code.

LLMs let you go 20% faster* but they make you think you can go 2x faster* and there's a lot of bad work in that gap *numbers not exact

“I picked up the guitar last year, and after endless hours of private lessons and online tutorials, I’ve now learned enough to realize that I have no idea what I’m doing with this instrument,” said the 45-year-old content strategist

New policy from Atlassian: Unless you opt out by August 17th 2026, data from Jira and Confluence will automatically be used for AI training. Some data cannot be opted out at all on some plans.