[artheclown]

they have news

[tomhollandnet]

TOM HOLLAND in The Devil all the Time (2020) dir. Antonio Campos

[adriftinhilbertspace]

This Hank Green situation is kind of sad. Aside from the outdated beliefs about AI, it's clear that most of the aggrieved population have never done research. Like research research.

Everyone seems to think you can just... Google stuff? And it comes up? bruh the moment you get anywhere near the frontier of knowledge, accessibility craters. Hard.

A real life example: during my PhD, my whole (subsubsub-)field existed basically entirely because a certain sensible conjecture turns out to be false, so we need a weird subtle painful approach rather than just using the obviously true fact.

How do we know it's not true? A counterexample was given in 1913 in a paper written in french, which no one actually cites, has no existing english translation, is nigh impossible to find a pdf of (no one could tell me its name - only its author's name), and is over 100 pages long, seems to be (I think) concerned largely with only tangentially related stuff, and is impenetrably dense due to lacking many modern concepts and notations. No one I'd ever met had ever read it, let alone understood it. And no other published proof exists - that I know of. It is not a simple proof. I wouldn't be able to prove it myself.

One day, my supervisor and I were emailing a colleague, and I offhandedly mentioned that it was the "only known counterexample". The colleague responded - oh! That's not true. Here's a handful more.

He did not supply any proofs. My supervisor went and figured out and wrote down a proof for one, and satisfied himself. He did not publish it, and it is basically illegible. He tried to explain it to me, but it was unintelligible. I could not make heads or tails of it. I still don't know, myself, for absolute certain, that a counter-example even exists. I had all the access in the world to journals and leading researchers in the area and I still couldn't get access to a remotely legible explanation of the foundational fact that spawned my field.

Research is THICK with folklore, and often the only way to learn something vital is to be wrong in the right way in front of the right person at the right time.

Y'know what tool is capable of cutting through a huge proportion of that reliance on luck and the hundreds of hours that go into just trying to figure out what's known and where to find it (even for established researchers familiar with their field)? yuh. LLMs. and nowadays they're basically reliable. They check online. They check themselves. And they've read the entire literature and picked up on much of the unwritten folklore whose presence in the literature is only the shadow it casts.

You'd have to be insane not to be using them at this point. Even just to ask "is it known whether or not blah?". Genuinely an invaluable tool. You don't have to offload your cognitive. You don't have to sell your soul or publish slop. You can just... benefit. And people are coming thick and fast at Hank Green - who has contributed so, so much to humanity - for using LLMs. For research. For his videos.

Sad!

[positivelyprime]

Perhaps this is slightly contained within math research and not as relevant in other sciences/humanities, but I'm slowly starting to slightly agree. At this point I think it is basically a given that LLMs will become very valuable in the near future for research, but it is perhaps easy to use it 'wrongly'. The reasons why a researcher would use an LLM (scouting the field for sources and folklore, concept testing for small but tedious calculations, perhaps proof checking?) are different from the interests of your average student (i.e. generating proofs for homework assignments), which I think muddles the discussion quite a bit.

However, I will add that in my experience ChatGPT is still quite prone to hallucinations. This makes it more of an unreliable tool to me with a high ceiling but also the potential to waste half an hour trying to convince you something works before admitting defeat.

[everoutoftouch]

If you have Spotify reblog this and tag what your number one song on your “on repeat” playlist is.

[oysters-aint-for-me]

always sunny s18 trailer!!!

[martyrstar]

Mom vs Bro Strife

Child abuser vs Alcohol Abuser, who's winnin?

[y1f4]

nice lady . her heel still stings though .

(also @shedletskyyy look , cool shit)

[bleedingspiral]

[tenderteeths]

the first time i’ve ever felt seen

[sin-ophelia]

Fangirls Through the Ages by Lid Thom

[heartpdf]

[sizzlingsandwichperfection-blog]

References Warhol? Check. Two narrators witnessing the same things through different eyes? Check. Hollywood sinking into the goddamned horizon like a literally fading star? Check.

[artheclown]

Barred owl I met in Iowa recently :)

[vavistus]

Half Man (2026)

[artheclown]

Jellycat party at the park.

[tomorrowillbeyou]

they are one-point compactifying me tomorrow

[importantpics]

[liquoricewizard]

@birds-in-posts

[birds-in-posts]

Grey heron (Ardea cinerea)

[yellowmagicalgirl]

Do you write your "z" with a line through the middle, and do you work/study in a field where you have to use a "z" coordinate?

Yes, yes

Yes, no

No, yes

No, no

I write my "z" (and possibly other letters) exclusively in cursive, not print

I have never written anything by hand in the Latin alphabet

Voting ended onApr 29

In my experience, people who put a line through their "z" are usually working with equations that require both a "z" and a 2.