Advertisement · 728 × 90

Posts by atticus goldfinch

7 hours ago 839 130 9 1

how many platner backers could I sway by saying "idk he's kind of Arthas-coded"

6 hours ago 8 4 0 0

Hit dog

6 hours ago 2 0 0 0
Post image
2 years ago 257 98 0 0
Post image
2 years ago 8 1 0 0
Post image
1 year ago 47 9 1 2
Post image
1 year ago 84 10 0 1
Advertisement
Post image
7 months ago 303 56 3 7
Post image
5 months ago 264 59 0 6
Post image

my god.

7 hours ago 9 2 0 0

just recently saw a thread full of redditors celebrating that marvel rivals gives you 3 bucks of in-game currency when you spend like 30 hours on a character, and that it's the only way people can afford to buy things.

we're talking about 10 cents an hour here, folks.

7 hours ago 14 0 1 0

my favorite incarnation of this is on video game subreddits, where if you let it slip that you can afford 10 bucks every few months on a battle pass or something for a free game, you're the elite bourgeois and are downvoted to oblivion.

7 hours ago 32 1 2 0

I think at some point you know you're being a dick. When the person you've been stringing along for hours is pulling their hair out and asking you to state clearly why you're hyperfocused on a dog moaning.. you either keep playing the game or you actually care about the convo and explain things.

7 hours ago 1 0 0 0
Post image

my god.

7 hours ago 9 2 0 0

I'm sort of rambly, but broadly: you argue the pattern, not the specifics. calling it wrong rock from the start removes its power when it appears.

7 hours ago 0 0 0 0

really though, you call out the behavior clearly from the start, you set boundaries, you give them no quarter -- 99% of the time they are unable to argue in those situations and silently slink off to weaker targets in their mentions.

7 hours ago 0 0 1 0

which is a cornerstone of good faith discussion! it's not some performative slopcratic method where you play at being on a higher intellectual plane than your prey!

7 hours ago 3 0 1 0
Advertisement

this tactic is phenomenally common in crank communities (not saying she's one.. just that similarities exist). I have probably seen this exact personality damn near 100 times from my old argue-with-creationists-all-night days.

only thing you can do is call it by its true name: wrong rock.

7 hours ago 2 0 1 0

she is not really interested in what other people are saying or what the output is. to her, "good faith" is trying to win the argument because she's right and has a divine right to prove it. that's all it really means.

when you say something, or show output, it gets routed to the next argument.

7 hours ago 1 0 1 0

So then he made it so it would _automatically_ ask these clarifying questions, exactly like what she wanted with the argumentum ad canis murmur -- and then that was cheating too, didn't count, for reasons.

7 hours ago 2 0 1 0

I'm not sure if you got to experience her conversation with Crespo, but he literally was building a tool to do what she said was impossible. When he'd show her output, she'd say it was wrong, so he'd refine, and then she'd say that refinement was basically cheating because it should just know.

7 hours ago 2 0 2 0

I interpret it as fundamentally bad faith and not actually engaging with your "opponent".

But IMO the behavior is downstream of pre-emptively thinking about how to "win" the next six possible arguments. She's not listening or engaging really, just trying to win.

8 hours ago 4 0 1 0

This is just normal. If you had asked people in 2015 what "reasoning" means, everyone would have said "dog noises", not "unsolved problems in mathematics"

8 hours ago 7 1 0 0
jpicone: "This just seems like testing if the llm will play twenty questions without knowing it's interviewing someone deliberately giving minimal information?

Why is this test so much better than all the other extremely obvious reasoning behaviours, like solving research level maths problems?"

Kathryn Tewson: "It’s a much more unbounded test, for one thing."

jpicone: "Well that's obviously false, the set of reasons why your dog could have made a noise is finite and the set of valid mathematical proofs is undecidable."

jpicone: "This just seems like testing if the llm will play twenty questions without knowing it's interviewing someone deliberately giving minimal information? Why is this test so much better than all the other extremely obvious reasoning behaviours, like solving research level maths problems?" Kathryn Tewson: "It’s a much more unbounded test, for one thing." jpicone: "Well that's obviously false, the set of reasons why your dog could have made a noise is finite and the set of valid mathematical proofs is undecidable."

Dodging the question when asked about undeniable examples of reasoning is also not a good look. She continued this thread by just asking why her dog made the noise, again.

Such good faith, nobody's ever had gooder faith.

12 hours ago 12 1 4 0

I think this is really well said honestly. "Will it play twenty questions with someone it doesn't know is being intentionally minimal?" is a very easy test for an LLM to pass if you add a few system prompts!

8 hours ago 4 0 1 0
Advertisement

unless and until republicans agree to a ceasefire in the form of strict anti-gerrymandering provisions: fuck them to hell, squeeze every last drop of blood from the stone, lock them out of every seat you can

10 hours ago 2622 324 50 15

The thing is, if you’re Iran, endless ceasefires while economic pressure mounts seems pretty close to ideal

10 hours ago 331 31 5 1

Can you give me visual examples of this?

11 hours ago 1 0 2 0

Stephen Miller, Whiskey Pete, AND Brendan Carr will all be hooting up Trump as they parade their regime media.

11 hours ago 262 75 6 1

Perhaps a special sort of stupid instead of a special sort of smart. Debate kid brain.

11 hours ago 3 0 0 0