iiinova — where your digital self keeps evolving

brndn_m @brndnm.bsky.social · 04/10/2026

That's a fair criticism that a few others have raised as well, but it's also a criticism that comes from a place of good faith, which puts you at least 10 steps ahead of the guy who was responding to me

120

brndn_m @brndnm.bsky.social · 04/10/2026

they literally have the logs of the conversations

150

brndn_m @brndnm.bsky.social · 04/10/2026

It's the last line of that bit that I'm drawing a similarity too. LLMs do not have a conception of truth or falsehood. They aren't designed to produce output that humans perceive as true; they're designed to produce output that gets humans to keep engaging with them.

160

brndn_m @brndnm.bsky.social · 04/10/2026

As a counterpoint, LLMs are also convincing people they are God. LLMs produce sequences of tokens that are probabilistic based on their training. That they *may* produce truthful information is incidental.

arstechnica.com

Man told ChatGPT he was feeling delusional. ChatGPT insisted he was Jesus.

Man with bipolar disorder sued OpenAI after surviving ChatGPT-linked suicide attempt.

110

brndn_m @brndnm.bsky.social · 03/10/2026

You are begging the question about what constitutes bullshit (in the formal sense of the term "begging the question"). I'd recommend reading the Frankfurt paper first to see why the term is accurate.

3390

brndn_m @brndnm.bsky.social · 03/10/2026

It's a pejorative term for LLMs, since LLMs produce bullshit (in the Harry Frankfurt "On Bullshit" sense of the term)

4281

brndn_m @brndnm.bsky.social · 03/10/2026

I think that it is good for people to get out of false or conspiratorial thinking, but I am worried about the psychological mechanism at work here that makes people willing to trust the bullshit engine and externalize judgement to a machine instead of trusting researchers and experts

9950

brndn_m @brndnm.bsky.social · 30/09/2026

I just want to join the chorus of voices asking "wtf how did they score KH2 a 7/10?"

010

brndn_m @brndnm.bsky.social · 16/09/2026

In computer science, I would wager that 1 and 3 are generally low, but with a rate of ten times higher for papers that don't open source any code they write or models they produce as part of their paper. Poor use of stats is everywhere. It's terrifying to me how few academics have stats training.

001

brndn_m @brndnm.bsky.social · 13/09/2026

The same type of people who smoked weed in America 30 years ago (before the widespread push for medical legalization)

010

brndn_m @brndnm.bsky.social · 13/09/2026

I love Che Guevara behind Che Guevara

291

brndn_m @brndnm.bsky.social · 10/09/2026

I made a similar observation in response to the claim that China is behind the anti-datacenter trend in the US. As far as I can tell, "wait for the other guy to run out of money" is working pretty well as a strategy

060

brndn_m @brndnm.bsky.social · 09/09/2026

On the other hand, my program made me take vector calculus (which I have never used outside of vector calculus class), and for that I will never forgive them

000

brndn_m @brndnm.bsky.social · 09/09/2026

My CS degree also required zero classes in ethics or philosophy. It obviously should be a requirement, but I'm also concerned that it wouldn't do a lot since every case study of unethical behavior could be pulled from a company that had raised or is raising billions of dollars in startup capital

100

brndn_m @brndnm.bsky.social · 07/09/2026

I think it is good television

011

brndn_m @brndnm.bsky.social · 07/09/2026

This has led software devs to wildly overestimate how useful these tools are outside of their field. They often don't seem to realize that there were a ton of structural and data-availability advantages for LLM use in coding that can't be generalized to any other field

7552

brndn_m @brndnm.bsky.social · 07/09/2026

That's a part of this conversation that I've only seen raised recently: there is a ton of domain-specific training information (every public GitHub repo, every language's public docs) for having LLMs generate code, and it's possible to automate the loop of generating and testing code.

1240

brndn_m @brndnm.bsky.social · 05/09/2026

Americans are about as likely to be killed in a motor vehicle accident as they are by a gun (it's about a 15% difference). In the US, the average gun kills fewer people each year than the average car.

090

brndn_m @brndnm.bsky.social · 05/09/2026

And I'm pretty sure that the second one doesn't even know what outside is. She's literally living the life of cat royalty

010

brndn_m @brndnm.bsky.social · 05/09/2026

My first cat is a rescue who had been living outside prior to getting picked up by the shelter. Since we adopted, he has gone outside once, and I'm pretty sure he considers it the worst experience of his life.