Dedekind Slut's Avatar

Dedekind Slut

@chasmat.bsky.social

they can’t do that, that’s illegal

642 Followers  |  538 Following  |  3,726 Posts  |  Joined: 10.10.2023  |  2.3969

Latest posts by chasmat.bsky.social on Bluesky

A Democratic was in office when Yick Wo was decided, so clearly distinguishable

08.08.2025 02:50 — 👍 0    🔁 0    💬 0    📌 0

Somebody posted a screenshot from Claude and it does this without further prompting

08.08.2025 02:19 — 👍 2    🔁 0    💬 1    📌 0

Hey, this answers a question I thought of earlier today: there’s probably a category of uses where creating the tokenizer is the hard problem, right? And lo

08.08.2025 02:17 — 👍 1    🔁 0    💬 0    📌 0

The fact that it broke it down like that makes me think Anthropic specifically included training data to address this problem lol

08.08.2025 02:01 — 👍 6    🔁 0    💬 0    📌 0

I asked an AI knower about it if you’re curious

08.08.2025 01:59 — 👍 3    🔁 0    💬 1    📌 0

Fact Check: FALSE, immigration is a civil matter

(Sorry, that Wapo guy is retiring and I’m trying my hand at being a pedantic dipshit. Time to get paid after a lifetime of doing it for free.)

08.08.2025 01:57 — 👍 1    🔁 0    💬 0    📌 0

The juice for PE is to really ride the line between “useless” and “defectively designed”. This use case presents closer calls than most I think.

08.08.2025 01:37 — 👍 1    🔁 0    💬 0    📌 0

Tripod? 👀

08.08.2025 01:25 — 👍 1    🔁 0    💬 0    📌 0
Post image

Ok literal lmao

08.08.2025 01:10 — 👍 0    🔁 0    💬 0    📌 0
Post image

This is actually kinda fun. I guess I would write in @bradlander.bsky.social ?

08.08.2025 01:09 — 👍 0    🔁 0    💬 1    📌 0

This poll is going hard on the scourge of rent-stabilized tenants who own second homes, a thing that doesn’t happen (it’s always somebody’s mom who’s on the deed for credit reasons).

So I guess look out for that in the coming weeks

08.08.2025 01:04 — 👍 1    🔁 0    💬 1    📌 0
Preview
a man in a suit and tie is smiling and making a funny face . ALT: a man in a suit and tie is smiling and making a funny face .

this poll, asking follow up questions that start including things like “are you SURE”

08.08.2025 00:58 — 👍 2    🔁 0    💬 1    📌 0

>“Do you support the following policies enacted in 2019?”
>lists policies that don’t exist under HSTPA of 2019

08.08.2025 00:55 — 👍 0    🔁 0    💬 1    📌 0

lol, just got an incredible push poll on NYC rent regulation. Very interesting to see how REBNY or whoever frames the policies to get the answers they want, versus how those policies actually work

08.08.2025 00:53 — 👍 2    🔁 1    💬 1    📌 0

Do tell

08.08.2025 00:39 — 👍 0    🔁 0    💬 0    📌 0

Thank you to whoever at @bsky.app finally made it so that the back-to-top button on desktop UI now lines up with the pfps. I see you and appreciate you

08.08.2025 00:33 — 👍 1    🔁 0    💬 0    📌 0

Homophobia

07.08.2025 23:29 — 👍 1    🔁 0    💬 0    📌 0

Just get into bodybuilding bro, it's not hard. (I mean it actually is hard, which informs the analysis here.)

07.08.2025 23:28 — 👍 1    🔁 0    💬 0    📌 0

Me, staring at your display name instead of your chest, trying to work out the pun: 😐 🤔

07.08.2025 23:07 — 👍 1    🔁 0    💬 1    📌 0

...posts about LLMs not knowing how many b's are in blueberry. This is how the singularity comes, I'm pretty sure

07.08.2025 23:03 — 👍 1    🔁 0    💬 0    📌 0
Preview
A Guide to Building an LLM from Scratch | Symbl.ai Up until recently, building a large language model (LLM) from scratch was a difficult and involved process – only reserved for larger organizations able to afford the considerable computational resour...

Is this summary (randomly pulled from Google) basically correct to, say, a second-degree approximation? symbl.ai/developers/b...

07.08.2025 23:00 — 👍 1    🔁 0    💬 1    📌 0

Thank you! This is very helpful. The embedding layers contain a LOT less information than I was thinking they do. Which also explains why attention is so important for the juice.

07.08.2025 22:53 — 👍 1    🔁 0    💬 1    📌 0

A token is a morpheme, which you embed into a VS. The ML pumps out a metric/dot product from corpus. I assume that can be iterated, as long as you have more vectors than tokens? # of parameters = dimension? I'm trying to understand training before runtime, but they seem similar in important ways.

07.08.2025 22:27 — 👍 1    🔁 0    💬 1    📌 0

@segyges.bsky.social Hey, why are LLM chatbots so bad at counting letters in words? Because it's a fact that's not implied by the lexical relationships in the underlying training corpus?

07.08.2025 22:04 — 👍 1    🔁 0    💬 1    📌 1

A complete theory of quantum gravity requires a non-interacting "b_3" term in the word "blueberry" in order to be self-consistent.

07.08.2025 22:02 — 👍 1    🔁 0    💬 0    📌 0

I waver on that, as a recovering Baggy Clothes Guy, because swimming in your clothes can feel protective if you dislike your shape. Tailoring would actually help, but I could see a bad mindset formed early that just sticks. I wonder what he was like in his 20s/30s, but not enough to investigate.

07.08.2025 21:57 — 👍 1    🔁 0    💬 0    📌 0

I don't know—I know a fair number of feminine guys and I would not classify Donald Trump among them.

07.08.2025 21:53 — 👍 2    🔁 0    💬 1    📌 0

Also I am incandescent with fury that I feel the need to psychologize this man, but elections have consequences (and they have been a disaster for the human race).

07.08.2025 21:52 — 👍 1    🔁 0    💬 0    📌 0

He is somehow obsessed with his physical appearance to an overweening degree (many such cases), but refuses anything that would demonstrate effort in maintaining it, with my sense being that he would find it unbecoming to care. Thus, spray tan and a wig in place of, like, skincare and exercise.

07.08.2025 21:51 — 👍 4    🔁 0    💬 2    📌 0

Hmmm. I wouldn't say "feminine," rather "effete"

07.08.2025 21:47 — 👍 6    🔁 0    💬 2    📌 0

@chasmat is following 20 prominent accounts