iiinova — where your digital self keeps evolving

wow do the surveys really show people are angrier with trump on inflation than they were with carter? that's surprising to me

000

ohhh i remember this. didn't the duck live on top of their monitor

oh yeah it can do that for sure

010

i don't know what rubber duck means in this context i have found LLMs consistently terrible at offering deckbuilding advice

210

AI Errors Are Statistical Errors

I have been running this test on every big new model since *2024* and this is the first time a model has actually solved it. Why is this such a hard task for LLMs? I discuss this in an old Substack: nickchk.substack.com/p/ai-errors-...

nickchk.substack.com

If it trains like a duck and predicts like a duck...

041

Sorry I mean Opus *5*. Got confused after writing GPT6

230

It's that time again, and there's a slight change!!! GPT-6: Fails Gemini 3.1-pro: Fails Opus 6: **Success**! I asked it twice and it was right both times, both in terms of calculating who wins and the chain of logic that works out why Opus 6 is the first model that has ever succeeded on this task.

Nick Huntington-Klein @nickchk.com · 16/09/2026

after some testing i have some greater success with instructing the AI to annoy the student and remind them that AI is not allowed. this will in no way prevent cheating (they can remove the warning text and re-ask) but i suspect will deter some students in automatic-use-AI mode

110

Nick Huntington-Klein @nickchk.com · 16/09/2026

hmm it worked when i tested it earlier today but now looks like it only functions about 50% of the time and warns tehm the other 50%

200

Nick Huntington-Klein @nickchk.com · 16/09/2026

i have shifted much of the grading weight to exams / quizzes but there are some exercises where it's worth it to do it anyway.

Nick Huntington-Klein @nickchk.com · 15/09/2026

i have to assume students are wise to the "secretly put instructions for the AI in your assignment page" cheating detection at this point right

250

Nick Huntington-Klein @nickchk.com · 15/09/2026

decided to start a thread that i will come back and add to whenever i hear a song and i'm like 'oh damn that's really good' www.youtube.com/watch?v=Dryd...

youtube.com

Macie Stewart - "I Forget How to Remember My Dreams" (ft. Lia Kohl)

YouTube video by International Anthem

Nick Huntington-Klein @nickchk.com · 15/09/2026

Bringing gollumf rice to the potlock final meeting

010

Nick Huntington-Klein @nickchk.com · 09/09/2026

plums are fantastic. however, like all stonefruit they are low-floor high-ceiling. a great plum is a top tier fruit. a mediocre plum is awful.

1131

Nick Huntington-Klein @nickchk.com · 08/09/2026

in the final episode of the most recent revival season he becomes a business owner

031

Nick Huntington-Klein @nickchk.com · 02/09/2026

Anyone particularly well-versed in AI safety stuff available to answer a DM question

000

Nick Huntington-Klein @nickchk.com · 01/09/2026

Anyone else got a big uptick recently in review requests way outside your expertise?

340

Nick Huntington-Klein @nickchk.com · 28/08/2026

I've yet to receive an angry email about any of my local-news appearances but sometimes I see the comment threads when they post the videos on social media and while very rarely is anyone mad at *me*, hoooooboy are they mad at something!

010

Nick Huntington-Klein @nickchk.com · 28/08/2026

also i got to do causal inference in the newspaper. it's a (basic-enough-to-describe-in-the-newspaper) crime event-study-and-then-DID

020

Nick Huntington-Klein @nickchk.com · 28/08/2026

i have an op-ed in the actual newspaper next week on a topic that positions me against an opinion i'd call "nextdoor/facebook-typical". how many angry emails should i prepare for

230

Nick Huntington-Klein @nickchk.com · 25/08/2026

"don't show me this" button does nothing

000

Nick Huntington-Klein @nickchk.com · 25/08/2026

every time i log in: - right-wing culture war stuff - christian vs atheist "debates" in the form of extremely smug memes - people whose politics i agree with but who post exclusively about people who send them mean comments - inflation pregnancy vore fetish pages??? as in all three combined

100

Nick Huntington-Klein @nickchk.com · 25/08/2026

i am still on facebook for the purposes of sharing baby photos with family and man there is not an algorithm on this planet less able to figure out what i want to see. like not even malicious, just incompetent.

190

she has been my daughter's favorite artist for about a month, which is about how long she typically holds a favorite, so i've heard a lot of it now. i'd say she has two good tracks (Lush Life and Midnight Sun) and a whole lot of pretty generic pop music you could very easily skip forever

000

this is in the US where apparently she has never had a top 10 hit

100

my daughter's 7th birthday happened recently and every child, without any prior coordination, knew all the words to two different Zara Larsson songs. you may use this as a barometer of your in-touchness

Nick Huntington-Klein @nickchk.com · 18/08/2026

Everyone in this thread was clearly an adult when the cd came out. Personally the cd is older than me so my first cd was the cd single for One Week by the Barenaked Ladies lol

000

They discuss their process themselves in this podcast episode bsky.app/profile/nick...

0130

Nick Huntington-Klein @nickchk.com · 18/08/2026

The authors behind this study openly try dozens to hundreds of quasi experiments and publish only the ones that "work". This makes me very skeptical of any of their findings, regardless of any robustness tests they do.

3405

Nick Huntington-Klein @nickchk.com · 16/08/2026

I think a difference there is that a model of human cognition quality could come up with new uses for itself, endlessly.

030

Oh na sparkling is the only kind of na wine I've had

i notice that these are all, like, fruit cordials instead of things labeled "scotch" or "wine" which is probably smart!

110

if dr. pepper had less sweetener in it it would taste more like scotch than the NA scotch i got

yeah lemme just pay $4/can for "non-alcoholic wine" that somehow tastes exactly like a slightly less sweet grape juice and not at all like wine, or $50 for a bottle of "non-alcoholic scotch" which is, like, some herb water i guess that has nothing to do with whiskey

Nick Huntington-Klein @nickchk.com · 08/08/2026

i am a drinker of actual alcohol but my experiences with non-alcoholic drinks allows me to summarize: old NA beer brands: awful new NA beer brands: really shockingly good any NA that's not beer: it's not bad or anything but also it's just extremely expensive juice.

3160

From LinkedIn, my book has finally been spotter in a non-university bookstore! At Waterstone's in London

1422

Nick Huntington-Klein @nickchk.com · 04/08/2026

Just got my copy yesterday, thank you for sending!

(and besides, many of the "clean" papers have similarly important gaps, they're just ones we're used to so we let them slide. the difference between "this flaw is unacceptable" and "this flaw is fine" is usually a social one not a rigorous one)

0180

But the number of useful more-signal-than-noise papers I've skipped bc I know that R2 will peace out between the part of the paper where it becomes clear there's a gap and the part where I take the gap seriously is a lot.

1110

The obvious downside is that those results can then be distributed further and taken as more definitive than they deserve.

190