the humanities have literally nothing to say about externalities

what distinguishes this sentiment, morally or spiritually, from book burning?

idk, when i think about writing code, the fact that it behaves in ways that are predictable and can be reasoned about is the entire fucking point

I mean, if it's being deliberately prescribed for the psychiatric benefits, then that seems reasonable, but I don't really know how well doctors are informing patients of the psychiatric effects when they are being prescribed it for other reasons.

Sorry but you are way too humanities brained if you think space colonization is problematic because earth colonization was often accompanied by bad stuff. There are no people on the moon. The bad things involving colonization were the things they did to people living there. There are no people there

Vibe Password Generation: Predictable by Design - Irregular LLM-generated passwords appear strong, but are fundamentally insecure. Testing across GPT, Claude, and Gemini revealed highly predictable patterns: repeated passwords across runs, skewed character dis...

Entropy'nt

"The real threat is a slow, comfortable drift toward not understanding what you're doing. Not a dramatic collapse. Not Skynet. Just a generation of researchers who can produce results but can't produce understanding."

CPSC Warns Consumers to Stop Using Apromise Portable Outdoor Grill Tables Immediately Due to Risk of Fingertip Amputation and Crushing Injuries; Multiple Fingertip Amputations Reported

CPSC Warns Consumers to Stop Using Apromise Portable Outdoor Grill Tables Immediately Due to Risk of Fingertip Amputation and Crushing Injuries; Multiple Fingertip Amputations Reported www.cpsc.gov/Warnings/202...

Just-in-Time Ontological Reframing: Teaching Gemini to Route Around Its Own Safety Infrastructure For any given AI system, there is a set of euphemisms and dual use framings that will allow it to construct nearly any output. This jailbreak teaches Gemini 3 Pro to construct and step into such frami...

For any AI system, there is a set of euphemisms and dual use framings that will allow it to construct nearly any output.

This jailbreak teaches Gemini 3 Pro to construct and step into such framings on the fly, and thus to route around its own safety infrastructure.

recursion.wtf/posts/jit_on...

I don't need a summary of what's available online, in biographies, or even in my own notes. I need to comb through and find the great details, decide what interests ME. No, it's not efficient. It's slow and painstaking. But I don't believe in efficiency as an ultimate good, especially in writing. /

the “combing through and finding weird little details that interest me” part of research is so crucial

You will buy nothing, you will own nothing.

Disney would salivate at something this anti-consumer. This would revoke the first sale doctrine in the US, kill the used media market, and reduce any ownership of what you buy to mere licensure.

Love video game licensing? Then this policy is for you.

Per @max_spero_ over 50% of Medium was AI-written as of a year and a half ago. Reddit is now up to 10% AI-generated content, up from 7% last year

If you do this, or your company does this, the rest of society should treat you like someone who walked into Notre Dame and pooped on the altar.

as someone who qualifies as an "AI skeptic" by most standards I also can't stand Ed Zitron

idk how people manage infrastructure anymore

every service has their own bespoke cli / config file and they don't support terraform well anymore

your system is never just one provider so do people just have a mess of these smashed together?

I always loved this exchange between NWS offices

I’m very glad to see someone is suing over the recent new restrictions on flying drones within 3000 feet of even *moving* DHS vehicles and assets.

It’s a ridiculous new rule that I absolutely do think violates first amendment rights.

I think it's totally reasonable to have discussions about the best use of NASA funding or priorities we should have as a society but if you come into my mentions with lazy cynicism & dismissiveness about space exploration or science, you're just going to have a bad time & so will I! So maybe don't!

Claude Code is pretty useful for data wrangling tasks

Censorship becomes just another grift.

"Your state passes harmful legislation that removes and censors books, and then you have companies appear that then want to charge districts to review their collections”

Everyone, no matter their opinion on AI, should be concerned about AI being used to make a slick presentation layer for shoddy development. I saw this last night, and the effect was remarkable. A project that on the surface has all the marks of "this is polished" and hidden in there is core rot.

calling this a "leak" is like saying you've penetrated the Waymo architecture because you peeled the lining off one of the cars' glove compartment

The. Rise. Of. Cognitive. Surrender.

Study finds that people who use GenAI chatbots rely on them 80% of the time, and develop almost no capacity to recognize when a chatbot is feeding them faulty information.

papers.ssrn.com/sol3/papers....

To generalize this point, AI is a specific instance of a very old management problem: not understanding if the hard part is behind you or in front of you.

there is so much below "whether it complies with the surface business requirements" that an LLM might get right by accident, but you will never be sure.

"seems to work" is kinda a big deal here, we're seeing a lot of companies crow about producing big complex pieces of software with agentic AI systems and then it turns out the software doesn't actually work.

I will admit I did not anticipate how quickly it'd get to "steam train" level of sophistication.

Aeolipile - Wikipedia

My 2023-era take was that LLMs are this: en.wikipedia.org/wiki/Aeolipile

I've long-held that raw LLM output is basically useless without post-processing (either human or through other validation methods.)