Remote

PhDog

@dogfox@kpop.social

mastodon 4.7.3

Semi-lapsed analog photographer.
Physicist by training, machine learning researcher by trade.
Current research is in finding a job. #FediHired

Pas de guerre entre les peuples,
Pas de paix entre le classes.
De même:
L'amour aux stans
Le feu aux entreprises

438 Followers

429 Following

50 Posts

Joined October 12, 2024

Pixelfed:

@SkipperMarch@pixelfed.social/

Blog:

https://dogfox.codeberg.page/

PhDog @dogfox@kpop.social

· 2mo ago

Got a bunch of CD from my aunt. While ripping them I re-found this which was a darkroom classic for me. I remember making my own developer from scratch to this. Emmylou Harris --- My Baby Needs a Shepherd https://www.youtube.com/watch?v=QHjRYF5SszI #Music

My Baby Needs a Shepherd

2

1

0

0

PhDog @dogfox@kpop.social

· 5mo ago

I drank a little bit. Have this meme I found on my laptop:

9

0

8

0

PhDog @dogfox@kpop.social

· 3mo ago

Replying to

https://www.youtube.com/watch?v=vLUTDOLpQCo @jerry@infosec.exchange

Ren is angry

2

0

0

0

PhDog @dogfox@kpop.social

· 3mo ago

Replying to

This is incredible. @agturcz@circumstances.run @jerry@infosec.exchange

2

2

0

0

PhDog @dogfox@kpop.social

· 5mo ago

#TWICE #Momo #MomoHirai

kpop.social

4

0

1

0

PhDog @dogfox@kpop.social

· 5mo ago

RE: @GottaLaff@mstdn.social

This is both totally nuts and very plausible.

#Epstein_files #Epstein #USpol

Laffy: "Popping in with a really remarkable, intriguing, …" - Mastodon 🐘

3

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Challenge level: Impossible @lyonsinbeta @ana

3

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

I see he still doesn't quite understand the thing I explained to him last time.

3

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Yo, where tf is Sander anyway?

3

0

1

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Holy Christmas! Keep in mind, Mistral large 2 is about 1/8 the size of GPT-4 if you believe Wikipedia's guess that GPT-4 is a trillion parameters. Mistral large 2 is only 123 billion. https://mistral.ai/news/mistral-large-2407 https://en.wikipedia.org/wiki/GPT-4 @mttaggart

3

1

2

0

PhDog @dogfox@kpop.social

· 6mo ago

I can only read so much of this interview-prep crap before I start giggling too hard and need to stop.

4

0

0

0

PhDog @dogfox@kpop.social

· 7mo ago

Replying to

Well, you don't want a bad back AND an apocalypse, do you? @Xeniax

4

1

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Hey, @momo@kpop.social ! That stupid ass Void Within nonsense is back on in Neopia! It's got very silly angst fanfiction-quality writing!!

🍑 :gay: :yerinvomfrog: (@momo@kpop.social) - kpop.social

2

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Here is a nice paper spelling out what should be obvious, that LLMs are not thinking: https://arxiv.org/html/2512.19466v1 @creepy_owlet

Epistemological Fault Lines Between Human and Artificial Intelligence

2

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

That sounds like... work! 😱 @jocafa @minego @nyquildotorg

2

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Good meeting with my friend today about the secret project. We will see what happens!

3

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Well, have you? Sounds like a good idea tbh. @grimalkina@mastodon.social

2

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Replying to

I get what you're saying a lot better now. Thank you. Strength of agreement isn't the same as purity. Purity also insists on completeness of agreement with predefined doctrine, if i am reading right. In that case, I think I agree with you that that is always pathological. @mttaggart @glyph @matt

2

0

0

0

PhDog @dogfox@kpop.social

· 4mo ago

Replying to

Wh-Oh, shit! I mean... /jk I am looking forward to this article. @mahryekuh@hachyderm.io

1

1

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Thinking about how the things I really value and care about havr so little cultural support in the myths I carry around with me, at least the ones that connect me to current time and place.

But then I say this to myself: you have the right to be an immigrant from your strange past.

2

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

This snuck up on me this evening. Good album, really.

IU --- Looking at You

https://www.youtube.com/watch?v=MmqVRM3u6ZY&list=PLk34QlH6hnPMMyJulVo4cLEITg_v55jN3

Looking at you (바라보기)

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

I really like beingfriends with my friends. :)

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Sure. NP. And I should say, the original plan didn't have in it what to do instead of "assume the distribution is normal." It was very much "This might be normal. Just pretend it is and if these numbers don't match, figure out what it is instead." Which is all in the "building/working" bucket rather than the "planning/conceptualize" one. @impactology@mastodon.social

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

This is my first cryptography project, but not my first time using statistics for something. @impactology@mastodon.social

1

2

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Sure. Right now I am working on some statistical problems for a cryptographic protocol I am setting up. I built most of it under the assumption that certain random variables are normally distributed. I did not know if they were but assuming that let me use a lot of pre-made libraries and build more important, structural stuff. Now, certain things aren't working but I know why and how to fix them because the weakness of that assumption (and how to correct it) was noted in my plan. @impactology@mastodon.social

1

4

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

That's not true. Good conceptualization and planning help you recover from/avoid binds you can get into while building. @impactology@mastodon.social

1

6

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

I read that paper. It was a good one! It has to do with *training* a model (of finite capacity) on it's own output. The simplest part of the argument is that it will loose the rarest data: It's less likely to come out, so it won't survive repeated rounds of training. Repeat ad infinitum until the data is just the dumbest shit. So an agent left on it's own on code won't necessarily do that since it is not re-training itself as it goes. @creepy_owlet

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

quick question but a long answer. @nateb@mastodon.thenewoil.org

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Worth a try! @bifouba@kolektiva.social

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

It's very weird to bundle them like that. Looks like they want t hide the worst. The only way to sus it out I can see is to take their 1.14g of Co2 per 400 tokens and try to estimate how many tokens what fraction of the 20 kilotons buys you. I guess the hing to do would be to dig up user/revenue numbers to try to estimate that. Revenue ~30-60 million USD/yr: https://electroiq.com/stats/mistral-ai-statistics/ Users pay ~$20/mo. so, ~4-5 million users? How many tokens you think they ran through in 18 mo.s? @mttaggart

Mistral AI Statistics By Revenue And Facts (2025)

Mistral AI Statistics By Revenue And Facts (2025)

Mistral AI Statistics - Mistral's valuation jumped from US$260 million in June 2023 to US$6.2 billion in June 2024.

1

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

That's pretty cool and makes a lot of sense. @screwlisp

1

3

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

Damn... @toxi

1

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Replying to

I CW K-drama reviews/opinions/reactions, just b/c not everyone gaf. I do food too because other ppl do, but would never think to if I hadn't seen it. @pockets

1

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Replying to

I also had questions about this section. While I read, I wondered how you thought of maintaining a social unit of any kind without censure and expulsion in some cases. And that's not a trick question btw; maybe you havr some ideas. Or to pose the question to this toot, what makes you think purity itself is the problem, as opposed to a fixation on it? Compare with money, which isn't itself evil, but an over-veneration of it can ruin a person. @mttaggart @glyph @matt

1

2

0

0

PhDog @dogfox@kpop.social

· 9mo ago

Replying to

I used pine the same year as you! I felt like a l33t haxx0r, lemme tell ya! @astro_jcm

2

1

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Alright, enough computer. These bits have been booped enough.

1

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Hey! Dahyun wasn't even in this episode! WTF are we doing here?

1

0

0

0

PhDog @dogfox@kpop.social

· 7mo ago

Did some interview stuff, trimmed the cat's nails... not a bad start to the day!

1

1

0

0

PhDog @dogfox@kpop.social

· 7mo ago

Replying to

Cartesian vectors don't form a group under multiplication* so division isn't defined. *either kind. They are not closed under the dot product and there is no cross product identity. @efi @ScriptFanix @StefanEJones

1

1

0

0

PhDog @dogfox@kpop.social

· 9mo ago

Replying to

I CAN HEAR YA! :yaya: @me@social.sandervankasteel.nl

1

0

0

0

PhDog @dogfox@kpop.social

· 4mo ago

Replying to

Erin go bragh!! @aral@mastodon.ar.al

0

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

Replying to

This is the only correction I could offer to OP: some people who got in early on crypto wound up loaded, which happens to some people who get in early on Ponzi schemes. @the5thColumnist@mstdn.ca @blog

0

0

0

0

PhDog @dogfox@kpop.social

· 5mo ago

Replying to

I feel like I should read more (i.e. more than the few pages I have already) Mencius. @0xabad1dea

0

0

0

0

PhDog @dogfox@kpop.social

· 7mo ago

Replying to

LOL! @0xabad1dea@infosec.exchange @ratsnakegames@mastodon.social

0

0

0

0

PhDog @dogfox@kpop.social

· 6mo ago

I just saw someone on the other site say that some professor's AI writing assistant was trained on middle school diaries and that is a pretty sick burn.

0

0

0

0

PhDog @dogfox@kpop.social

· 9mo ago

Replying to

If this is that gender ideology I've heard so much about, I might just have been converted. @Natasha_Jay

0

0

0

0

PhDog @dogfox@kpop.social

· 11mo ago

Replying to

Conversations ight go different at other times of year. @me@social.sandervankasteel.nl

0

0

0

0

PhDog @dogfox@kpop.social

· 3d ago

Replying to

This sounds right. @mattsheffield@mastodon.social @__d@mastodon.social

0

0

0

0

PhDog @dogfox@kpop.social

· 4mo ago

Replying to

Where's the Linux reference? @xssfox@cloudisland.nz

0

2

0

0