iiinova — where your digital self keeps evolving

wut sfstandard.com/2026/10/13/o...

3190

"We used our models to understand the device and where it had come from" Did OpenAI do a marketing bit in the middle of an espionage accusation?

7725

There was an anti-AI post on Hacker News that, as a random side note, alleged that vibe coders primarily use Discord to find underage users.

010

Harry McCracken @harrymccracken.com · 13/10/2026

tbh LLMs write correct but *verbose* SQL. They don't like CTEs.

050

Reposted by Max Woolf

The Luxo Jr cycle of life is finally complete.

Apple iMac G4 "lamp" desktop computer beside a HomeView

1388

Reposted by Max Woolf

The Fig Economy @figgityfigs.bsky.social · 12/10/2026

I’m still somewhat of a skeptic on a lot of AI, and a lot of the AI hype, but at this point calling it “fancy autocomplete” is itself a way less fancy autocomplete in response to a prompt that just mentions AI at all.

415414

Max Woolf @minimaxir.bsky.social · 11/10/2026

6.1 Sol is very good but from my testing Opus 5.5 is even better.

130

Reposted by Max Woolf

conputer dipshit @davidcrespo.bsky.social · 11/10/2026

low key I‘ve been thinking Opus 5.5 is an Opus 4.5-level jump in practicality and accessibility. I’m a little surprised not to be hearing more people talk about this, but I guess it’s been less than three weeks. I expect to see a wave of non-programmers finding new use cases for agents

2519210

Max Woolf @minimaxir.bsky.social · 10/10/2026

"This AI model made a specific, intentional decision to kill."

060

Max Woolf @minimaxir.bsky.social · 10/10/2026

The newest Law and Order (vanilla) episode is a ripped-from-the-headlines about an LLM who facilitated a murder against its programmer who was building guardrails for it. Verbatim quote: "Did you unplug the damn thing?"

3181

Reposted by Max Woolf

William B. Fuckley @opinionhaver.bsky.social · 09/10/2026

Zitron accidentally making his first good point about the downsides of AI.

637936

Max Woolf @minimaxir.bsky.social · 08/10/2026

Many of the people who have been saying "don't feed the trolls" over the past decade are now feeding the trolls because AI bad.

0174

The Hacker News thread about this suite got flagkilled with the same logical contradiction. news.ycombinator.com/item?id=4998...

0140

The replies are the biggest larrydavid.gif I've ever seen. "Yeah, this is free and super important and I want it, but it's vibecoded."

3471

actual mathematical proof we are in a computer simulation

0110

I shudder to think why it's -182. Random number.

160

I don't know what age demo "Chuckie Schu" is for

0170

William B. Fuckley @opinionhaver.bsky.social · 06/10/2026

math would be funny if it were GPL

020

Reposted by Max Woolf

Yeah OpenAI or Anthropic or whoever might go bankrupt, some specific use cases like ‘therapists’ might be litigated to oblivion. But things that involve interfacing with code are never, ever going back to the modal default being a human being typing things out, like it’s just full stop not happening

535828

I may or may not have a tool for much faster local embedding creation that I calibrated for EmbeddingGemma but didn't want to release until a better embedding model came along.

1140

I did also say "2 standard deviations"

2440

"Not a number." More a qualitative you-won't-question-the-verdict.

1200

That's not a lie; I didn't process that was the wrong answer until I saw them recoil. Both the judge and counsel were picking on me because I said I was a data scientist so tbh they should have expected that.

4552

Got excused from jury duty because the prosecutor asked "Can you put a number on the amount of proof needed to establish beyond reasonable doubt?" and I immediately replied "Yes, 95%".

8954

1,012 bytes, make everyone happy.

4646

"It's a p-hacking for my novel."

2140

Pointing out that his point is ahistorical. At the least until ~2019, the scraping of data for ML was generally accepted and didn't have much criticism around IP. (aside from the ImageNet lawsuits which weren't about protecting creativity)

2180

Pretty much all ML/NLP research since 2007 was based on Common Crawl which was scraped web pages. BERT was trained off of Wikipedia dumps. It was far from taboo in ML research, it's just that the research was made a product that went mainstream.

5868

(trying to train a chess engine and it took until Experiment F for Opus to notice that the engine didn't know how to mate)

070

just had Opus 5.5 tell me "The model is learning to mate" and I had to do a double-take

1180