Replying to @NateSilver538
musk manipulated the algorithm to preferentially amplify right wing content immediately after buying twitter. We've known this for over a year now. Stop contributing value to the cesspool and just leave. It really is that simple.

arxiv.org/abs/2411.01852
3
0
1
Remember: by participating on twitter, you are adding value to it, thereby incentivizing others to join/stay, and amplifying the voice/reach of the people who set its algorithm's biases.

Is this the ideology you want to amplify? Because if you are here, you are amplifying this.
1
0
0
tried to throw some inspiration at Claude, got an infographic back
1
0
2
Looks like they make the connection here
0
0
0
Clever MLM benchmark. Could probably get some additional lift by e.g. rotating image representations, swapping axes on graphs, applying style transfer to convert the images into synthetic "hand drawn" sketches, jittering style attributes, etc.
I’ll be at @COLM_conf from Oct 7-9 presenting IsoBench. Looking forward to chatting about multimodal LLMs, mathematical reasoning, and how to understand Transformers/LLMs.
1
0
1
The site isn't just down: the domain expired. It probably didn't fall under "anyone's* "responsibility. A lot of old sites just hung around sometimes because it was a lot cheaper to register a domain for a prolonged period back in the day. It's probably never coming back online.
1
0
10
Reminder of some recent history while we're in the throes of a significant weather event that agencies like NOAA are important and under threat.
0
0
0
Easily the best live-tweeting of the debate, great work @TheOnion
1
0
1
workshopping a prompt to get the brushstroke texture to interact with the diegetic light
2
0
13
"Programmers Are Also Human" trolls NFT Paris and absolutely nails it
youtube.com/watch?v=59rxyj…
0
0
0
> mount rainier, a perfect "painting" which is actually a natural pattern in a discovered piece of marble. subtle, ukiyo-e, sandart. ultraviolet phosphorescence, brightly glowing embedded stone, Cherenkov radiation, liquid light.
0
0
3
> oversaturated, ultraviolet, hyperspectral, deepdream, psychedelic, 3D oilslick rainbow textures, melting explosion, depthmap bubblegum geode, mount rainier
[Flux]
1
0
1
same prompt in flux-dev
flying over colorful painted canyon, overhead, straight down, oversaturated, ultraviolet, hyperspectral, deepdream, psychedelic, oilslick rainbow textures, splatter painting, melting explosion, depthmap
0
0
5
So many people have achieved this they have different achievement tiers. Also worth note: that thread was started 2002.
0
0
0
Replying to @AISafetyMemes
according to wikipedia's Fusor article: literally yes, this.
1
0
1
Well on my way to my independent replication of ZakineBenzaquen2024
arxiv.org/abs/2406.18386
0
0
1
Replying to @DigThatData @cais
This is a version of "strongly opposing" that you are framing as support.
1
0
0
Replying to @CAIS @cais
This is dishonest framing. You are presenting it as if there is unanimity among academics and primarily industry opposes, but the vast majority of AI researchers vehemently oppose the bill. Notable opposing researchers include Yann Lecun, Jeremy Howard, Fei-Fei Li, ...
1
0
0
Some really good stuff in here
🚨Reading List: Transformative Leadership

Favorite SSIR articles on how leadership is changing, and how it still needs to.

Read and share with leaders you know! #nonprofits #socialenterprise #foundations #business #government #socialsector #leadership

ssir.org/articles/entry…
0
0
0
Replying to @nearcyan
apparently we're riding the same wavelength tonight
0
0
2
Replying to @DigThatData
To be concrete: this is what I'm supposed to be able to do for hashtags but can't do for that one.
0
0
0
Of course twitter goes out of their way to disable mechanisms to express disinterest in "TrumpOnX". Note the absence of the "three dots" menu icon for that hashtag/topic.
1
0
0
High quality 57B token dataset dropped. Interestingly, very little data filtering applied (filtered out non-english content and some of OpenOrca for 'low quality')

huggingface.co/datasets/pints…
We made a model pretrained in just 9 days, and it did better than Apple OpenELM and Microsoft Phi 😁 arxiv.org/abs/2408.03506
0
0
3
> alex grey qi gong kundalini energy rooted meridians fully charged dantian nuclear glow
0
1
2
I was off by a whole year! @websim_ai
Three years later...
> "Alexa: generate an alternative final season of game of thrones that doesn't suck."
1
0
3
SD 1.5 did it better
AI created image from the phrase, “Jesus flipping over the tables in the temple.”
1
0
14
still fails this one
Another good test for text-to-image models is The Dragon with the Girl Tattoo.
4
0
10
Replying to @PurzBeats
Meanwhile, here's footage of me enjoying GPU access through various employers over the last 5-ish years
0
0
3
some differential geometry to kick off your weekend youtube.com/watch?v=wO61D9…
0
0
0
Replying to @molly0xFFF
holy hell the delicious irony of this twitter ad
0
0
2
Replying to @technollama
Wild how difficult it is to find reasonably unbiased reporting on this issue.
0
0
1
TIL "cybernetics" and "kubernetes" have the same etymology
0
0
1
Replying to @chantz_y
It only gets worse as you go further down the rabbit hole.
2
0
3