Artificial intelligence still has some way to go

loooooool

that one hurt

This is impressive: https://www.independent.co.uk/life-style/gadgets-and-tech/protein-folding-ai-deepmind-google-cancer-covid-b1764008.html

https://www.youtube.com/watch?v=8HZ4DnVfWYQ

closed captions recommended there

fractal cucumber!

https://www.youtube.com/watch?v=gg7WjuFs8F4

The Biden-Harris administration is committed to rebuilding an economy that welcomes everyone as full participants.

— The Democrats (@TheDemocrats) December 1, 2020

It's great that journals are getting authors to do plain language summaries along with the more technical abstracts.

This one - from an article titled 'Potential and limitations of machine learning for modelling warm-rain cloud microphysical processes' - is plainer than most. 🙂 https://t.co/UyKoDvQPtT pic.twitter.com/W3a1OCrBxs

— Tim Baxter #athome (@timinmitcham) December 5, 2020

why haven't scientists always been doing that, omg

terrific

At first I thought an AI wrote that and I was like “I guess they don’t have that far left to go!”

This seems to be the thread previous Google web audio experiments were on, and this involves machine learning though I don't know exactly how, so, bloody hell, this is impressive, and also cute:

https://artsandculture.google.com/experiment/blob-opera/AAHWrq360NcGbw

Control your own blob choir! 4-part harmony! Sings Christmas carols (click the 🎄)!

Introducing Blob Opera, an experiment I created for Google Arts & Culture: https://t.co/dfLnCXSR0R pic.twitter.com/UYbZH8jXE2

— David Li (@daviddotli) December 15, 2020

That's pretty neat!

yeah, i love it

Ohh, apparently it also takes midi input. Which is also an excuse to post this Satie gymblobédie.

Discovered that @daviddotli's awesome Blob Opera project works with my MIDI keyboard! 🎹 Enjoy these blobs singing Satie. 🎵 @googlearts pic.twitter.com/6ZKHaK2o4X

— Alexander Chen (@alexanderchen) December 17, 2020

A big reason why these blobs are so wonderful is they are not the source of the intelligence which is making the music. That's being supplied by real musicians.

I kinda want to use that to make some fake vocals for Bicep-sounding tracks.

neat!
https://openai.com/blog/dall-e/

pika pika pic.twitter.com/ErIF703OBb

— Max Woolf (@minimaxir) January 5, 2021

gotta AI fast pic.twitter.com/myb8oPzVuZ

— Max Woolf (@minimaxir) January 5, 2021

OpenAI’s new DALL-E is GPT-3 but trained on language + images. It can generate pictures from text that blend concepts in interesting ways https://t.co/CtFwPcJ5L6 pic.twitter.com/wxDzIExqMX

— Mark O. Riedl (@mark_riedl) January 5, 2021

a human worker who is apathetic and slow, replaced with a machine that is never tired

a honking eggplant begging for licks

very much want to test this out

saaaaameeeeee let me in let me in let me in

your mom

https://venturebeat.com/2021/01/11/outlandish-stanford-facial-recognition-study-claims-there-are-links-between-facial-features-and-political-orientation/

Kosinski and coauthors, preemptively responding to criticism, take pains to distance their research from phrenology and physiognomy. But they don’t dismiss them altogether. “Physiognomy was based on unscientific studies, superstition, anecdotal evidence, and racist pseudo-theories. The fact that its claims were unsupported, however, does not automatically mean that they are all wrong,” they wrote in notes published alongside the paper. “Some of physiognomists’ claims may have been correct, perhaps by a mere accident.”

You may recall this is the same discredited charlatan Stanford prof who claimed AI could detect sexual orientation.

very much want to test this out

In my experience it is simultaneously true that

a) systems like this show very impressive performance;
b) what you see in the demo is the cream of the crop and if you mess around with it yourself you'll quickly come to see what kinds of tasks it's good at doing and a much larger universe of superficially similar tasks it's not good at all at doing

lol I was p confused by that post for a sec

i'm just saying a lot of these things that look like magic when you see the curated demo don't FEEL like magic when you play with them yourself, and that's because they are not, in fact, magic, just well-designed systems that do certain things and not other things

no I get it -- I just briefly thought you wanted to try out the neo-phrenology classifier

https://www.openphilanthropy.org/blog/new-report-brain-computation

and yes to eephus but the upside is that once you start playing with them that means EVERYONE'S playing with them and they tend to iterate quickly.

Lately every time I read a blog post or essay like this: https://marksaroufim.substack.com/p/machine-learning-the-great-stagnation

that completely loses track of its thesis--if it even has one--and just jumps around from topic to topic ranting a bit about each one, I wonder if it's going to be followed up in a couple of weeks with another "Ha! That last blog post was entirely written by GPT-3 and you fools couldn't even tell!!!"

i thought this was really interesting and well executed.

Here's a sacrilege - "Ballet Mécanique" by Fernand Léger and Dudley Murphy edited to the sound of "The Promo that Got Away (Insert Story Here)" by ghosts4hire

As usual with AI results there are parts where I want to manually polish things, but I here's the raw output. pic.twitter.com/2Nyvr55sTX

— Mario Klingemann (@quasimondo) January 24, 2021

er, sorry: explanation earlier in thread

For an upcoming collaboration I created an algorithm that edits music videos autonomously by selecting the best matching scenes and cuts for a given song. Here's a test with some found glitch material.

Song: Time Traveler (eroded by time mix) by ghosts4hire
CC BY-NC 3.0 pic.twitter.com/LqsKSoHuvg

— Mario Klingemann (@quasimondo) January 24, 2021

brilliant

This seems like an article Sarah Conner would read three months before Skynet took over:

https://www.forbes.com/sites/davidhambling/2021/01/27/drone-swarms-are-getting-too-fast-for-humans-too-fight-us-general-warns/

that’s p fucked up

thisisfinedog.jpeg

sarah connor needs to move to zion in the matrix - makes sense to hole-up deep underground and let the machines fight on the surface

https://thisanimedoesnotexist.ai/

uh that is very close to nsfw

And yet

wow, didn't know this existed and was a free download.

https://www.youtube.com/watch?v=He6MCalvDf4

https://www.theverge.com/2020/2/13/21136248/ar-pianist-massive-technologies-virtual-musicians-pianos

Lmao

The latest generation of adversarial image attacks is, uh, somewhat simpler to carry out https://t.co/CBnA53lEcy pic.twitter.com/h4e0bShq9i

— Brendan Dolan-Gavitt (@moyix) March 4, 2021

Good mugshot too

A Pennsylvania woman created "deepfake" videos of her daughter's cheerleading rivals, depicting them naked drinking and smoking

And then she sent them to her daughter's coaches to try and get them kicked off the squadhttps://t.co/WG49eJtODj

— philip lewis (@Phil_Lewis_) March 13, 2021

"An empty room"
Looking at daylight, reflections and shadows in GAN rooms
Oddly specific in all vagueness pic.twitter.com/WWqOGg0XnK

— Erik Swahn (@erikswahn) March 8, 2021

Very rad

Automatic "photoshopping" to order based on plain text commands. For example, give it a photo and tell it "mohawk", and it puts a mohawk on the person in the photo. Just to be clear, this is very different from eg an instagram "mohawk" filter would be, because afaik it isn't pre-programmed to do this stuff. Like, I think you could give it a novel prompt like "fish hat on head" or something and it would try to make that image for you.

https://github.com/orpatashnik/StyleCLIP
https://arxiv.org/abs/2103.17249

Pretty cool

update from a friend in Indianapolis:

"There’s a pilot program in Indy to test driverless cars. The manager said the goal is for people to use them to and from IUPUI, and take them to the hospital, and so forth. They appear to have test drivers who do not control the steering. How do I know? Today, one somehow ended up straddling two lanes and managed to rear end me and the car to my right."

Parkour is the perfect sandbox for the Atlas team to experiment with new behaviors. Through jumps, balance beams, and vaults, we push Atlas to its limits to discover the next generation of mobility, perception, and athletic intelligence. https://t.co/xZRNVnhrkc pic.twitter.com/E0ssh45ZCi

— Boston Dynamics (@BostonDynamics) August 17, 2021