@pbloem@sigmoid.social
@pbloem@sigmoid.social avatar

pbloem

@pbloem@sigmoid.social

Assistant prof. at the Learning and Reasoning group, Vrije Universiteit Amsterdam (searchable).

This profile is from a federated server and may be incomplete. Browse more on the original instance.

pbloem, to random
@pbloem@sigmoid.social avatar

If you post to , you license that content to them under CC-BY-SA. You can't revoke that license, so you can't remove that content.

It shouldn't surprise you that they stop you from trying. Imagine a Wikipedia editor trying to remove their edits.

However, what it does give you is the right to host all of SE elsewhere. To fork it, if SE violates its contract with the community. This is how SE was always designed to work. From an interview with Spolsky:

baldur, to random
@baldur@toot.cafe avatar

“Why Would I Buy This Useless, Evil Thing? - Aftermath”

This x1000. Just… why? https://aftermath.site/why-would-i-buy-this-useless-evil-thing

pbloem,
@pbloem@sigmoid.social avatar

@futurebird @dr2chase @baldur It's low as in the number of bits you use to represent the number, so it's indeed about fewer trustworthy significant digits.

The key is to use it only for parts of the network that are linear, like matrix multiplications, since the errors caused by low-precision don't blow up there.

It definitely matters after training too (during "inference"). After a network is trained, you can get down to an even lower number of bits, like 4. This helps with speed and battery.

futurebird, to random
@futurebird@sauropods.win avatar

More seriously, can someone steel-man the case for executive power being less effective if subject to future evaluations and constraints?

I don't think I've ever looked at any singular leader of any organization and thought "the problem is they can't do what they want easily enough"

When I was younger I might have said that the director of a film or play can do better work if they are an unquestioned dictator... but I've grown to even reject that idea.

pbloem,
@pbloem@sigmoid.social avatar

@Alon @mattmcirvin @futurebird

It's especially odd in Europe. I would love a functioning high-speed rail network spanning the continent like China has, but it's the lack of unified rules and standards that makes this difficult.

Perhaps a strong European leadership could cut through the red tape, but that's exactly what all the authoritarian parties over here abhor.

taylorlorenz, to random
@taylorlorenz@mastodon.social avatar

“These emails tell a dramatic story about how Google’s finance and advertising teams, led by Raghavan with the blessing of CEO Sundar Pichai, actively worked to make Google worse to make the company more money.” https://www.wheresyoured.at/the-men-who-killed-google/

pbloem,
@pbloem@sigmoid.social avatar

@futurebird @taylorlorenz

Apparently, the index is the real bottleneck. A high quality index is hard to create.

I wonder if a decentralized, (SETI@Home style) volunteer effort could work. Everybody runs a little crawler, and the results are unified over DHT.

Then we build loads of open source search engines on top of it, filtering however we like.

futurebird, to random
@futurebird@sauropods.win avatar

Have you ever tried to measure a cat with a tape measure? It will make you question the solidity of dimensional space.

Pica, small cat
Height: 22 to 47cm
Girth: 33 to ??cm (ouch cm)
Length: 37cm
Length w/ tail: 66cm
Weight: 5.9 pounds/ 2.7 Kg
Max Length (Dangle): 74cm! oh no

she's getting longer.

Rate of Enlonganmentation
dML/dt: 1.2cm / second

No signs of increase stopping. Cat keeps increasing.
All clear! ALL CLEAR!
Soon she will be in all locations.

This was a bad project idea.

pbloem,
@pbloem@sigmoid.social avatar

@futurebird The horizon is looking a little... furry today.

carnage4life, to random
@carnage4life@mas.to avatar

Twitter is the worst thing to have happened to perception of billionaires. It has normalized the richest people in the world saying the most ridiculous things you have ever heard.

pbloem,
@pbloem@sigmoid.social avatar

@carnage4life https://youtu.be/e-c-yiTE7zM?feature=shared

TedUnderwood, to ai
@TedUnderwood@sigmoid.social avatar

Computer-assisted studies of fiction tend to count words and trace themes. It's harder to measure the things that really keep readers turning pages: mystery, suspense, and surprise.

But large language models may change that. In this post, I experiment with a new way to measure uncertainty at different points in a story. https://tedunderwood.com/2024/01/05/can-language-models-predict-the-next-twist-in-a-story/

pbloem,
@pbloem@sigmoid.social avatar

Very cool. I think there's lot of promise in having LLMs "read" texts like this and using it to analyze the text itself.

Can I ask why you went with the distance in the embedding space? The model itself already produces probabilities, so you could check the probability of the truth under the model distribution (if this is available in the GPT API), or just the entropy of the predicted part (alhtough NNs are notoriously bad an producing well-calibrated uncertainty).

pbloem, to random
@pbloem@sigmoid.social avatar

https://www.theguardian.com/commentisfree/2023/dec/28/meaningful-debate-racism-we-asked-20000-people

pbloem, to random
@pbloem@sigmoid.social avatar

Rules for review requests:

  • If you don't let me decline without logging in, you are not getting a reply.
  • If you don't let me decline without giving written reason, or suggesting alternative reviewers, you are not getting a reply.
  • If you make an account for me on your system without my involvement, you are not getting a reply.

I'm happy to do all these things, and I probably will if you ask me, but not if you make me.

john, to ai
@john@sauropods.win avatar

Can somebody please get an AI image generator to generate a fox with a human face or head for me?

The other way around is really easy and does not count.

pbloem,
@pbloem@sigmoid.social avatar

@john DALL-E 3 seems to manage, and it's suitably majestic.

pbloem,
@pbloem@sigmoid.social avatar

@john First try, I promise...

pbloem, (edited )
@pbloem@sigmoid.social avatar

@john @AggroBoy The success rate is 3/5. Not counting this one, which is technically correct, I guess.

pbloem,
@pbloem@sigmoid.social avatar

@john @AggroBoy ChatGPT does do some stuff behind the scenes, including writing its own prompt. You can see part of that in the filename when you download the image. The prompt it wrote for the first image I posted was

"A surreal and imaginative depiction of a fox with a human head, blending the natural orange and white fur of the fox with the distinct features of a h..." (the filename cuts off there).

pbloem,
@pbloem@sigmoid.social avatar

@john @AggroBoy Ah, I can just ask it what prompt it used. Here's the full thing.

MrLovenstein, to random
@MrLovenstein@mastodon.social avatar

Secret Panel HERE 💀 https://tapas.io/episode/3016236

pbloem,
@pbloem@sigmoid.social avatar

@MrLovenstein

"Based"

I've looked it up too, multiple times, but it doesn't stick. I'm out.

pbloem, to random
@pbloem@sigmoid.social avatar

altman: other companies are catching up with our research. :( how do we keep our edge?

GPT-5: You need to distract their researchers. Keep them glued to Twitter despite themselves. Here's the plan...

pbloem, to random
@pbloem@sigmoid.social avatar

A round little number from my holiday cottage on Vlieland.

ErikJonker, to ai Dutch
@ErikJonker@mastodon.social avatar

Interesting,🤔 Google Bard doesn't want to answer this question "What are the differences between Hamas and Hezbollah?" , ChatGPT gives a good answer , https://chat.openai.com/share/306e9096-a0ca-4e30-bac6-29a4e034ac5b
It only forgets to mention Hezbollah is Shiite and Hamas mainly Suny, it does when I ask it as a follow-up question. But Google Bard doesn't want to talk about terrorist organisations it seems (=censorship)

pbloem,
@pbloem@sigmoid.social avatar

@schotanus @ErikJonker

They fuck you up, your mum and dad.
They may not mean to, but they do.
They fill you with the faults they had
And add some extra, just for you.

astro_jcm, to Cat
@astro_jcm@mastodon.online avatar

Alley , one of my favourite as a kid 🐈 Oh how many hours I spent chasing mice and running away from that flying broom!

pbloem,
@pbloem@sigmoid.social avatar

@astro_jcm

CGA graphics sure were an aesthetic.

pbloem, to random
@pbloem@sigmoid.social avatar

My hotel room from last week. In an old converted church.

maxkennerly, to random
@maxkennerly@mstdn.social avatar

Happy to be corrected by an expert, but seems clear to me that every "AI detector" for text / image / video / audio is a complete scam and they should never be used in any context.

https://www.404media.co/ai-images-detectors-are-being-used-to-discredit-the-real-horrors-of-war/

pbloem,
@pbloem@sigmoid.social avatar

@maxkennerly For images, there are clear indications that detecting fakes in at least some settings is possible. Many models leave clear artifacts that are imperceptible to the naked eye.

For natural language, there doesn't seem to be any sound basis to build a detector on (unless the text is explicitly watermarked).

ZachWeinersmith, (edited ) to random
@ZachWeinersmith@mastodon.social avatar

Does the word "cyborg" imply in particular a human-machine combination?

pbloem,
@pbloem@sigmoid.social avatar

@ZachWeinersmith Could be animal-machine. But there has to be a partly organic brain.

rolle, (edited ) to bluesky
@rolle@mementomori.social avatar

Just noticed Bluesky doesn't ACTUALLY add the alt text in the img itself. This is a big mistake and makes the service very inaccessible.

The text is actually in the very last part of the DOM (Document Object Model) tree... and is triggered via click. So, if the screen reader doesn't support JavaScript same way than for the seeing people, descriptive text is nowhere to be found. Even when it is, it can't be accessed easily.

#Bluesky #Accessibility #WebDev

pbloem,
@pbloem@sigmoid.social avatar

@rolle This is astonishing. Rolling your own alt text display, so that only sighted people benefit from it...

Also, that HTML, good god.

pbloem,
@pbloem@sigmoid.social avatar

@rolle

Designer: Our UI is basically a bunch of rectangles below each other.
Dev: Well, this is going to take at least 15 nested DIVs.

malwaretech, (edited ) to random

Twitter made a new ad for the platform that features a phone scrolling through Twitter. They ended up pulling it and remaking it after people noticed it contained a tweet mocking Elon for trying to blame Twitter's advertiser exodus on the jews. The second version of the ad, which is still online, contains a tweet about creampieing a rotisserie chicken.

pbloem,
@pbloem@sigmoid.social avatar

@malwaretech The real question is, do they consider the chicken tweet news, sport or culture?

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • JUstTest
  • kavyap
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • tacticalgear
  • cubers
  • Youngstown
  • mdbf
  • slotface
  • rosin
  • osvaldo12
  • ngwrru68w68
  • GTA5RPClips
  • provamag3
  • InstantRegret
  • everett
  • Durango
  • cisconetworking
  • khanakhh
  • ethstaker
  • tester
  • anitta
  • Leos
  • normalnudes
  • modclub
  • megavids
  • lostlight
  • All magazines
    •