Artificial intelligence still has some way to go

v cool

can't do much worse than we are right now

Get ready for people to start moving the Turing test goalposts: https://aiweirdness.com/post/620645957819875328/this-is-the-openai-api-it-makes-spookily-good

fucking hell
https://twitter.com/dog_fakes

This is Soren. He is having an existential crisis, wondering if maybe he isn’t just a lamb after all. 14/10 pic.twitter.com/OHf7kFfQvj

— dog_fakes (@dog_fakes) June 10, 2020

We only rate dogs. There is a broken pipe in the basement. Please don’t send Cheetos. This is a fire hazard. Thank you... 13/10 pic.twitter.com/JqA2yRGcJj

— dog_fakes (@dog_fakes) June 11, 2020

i am deeply curious if the alt-text is human or machine generated

Check Dan's link. The text is AI generated based on the images.

Ah, sorry, alt-text. Ignore my not reading properly. Apologies.

The picture texts are human written. I was mostly just impressed with the chatbot snippets from the link I posted, though I’m not sure I really trust the author not to have cherry picked or “cleaned them up”. The generated dog pictures are not special and have been possible for years I think.

The generated dog pictures are amusingly grotesque. Humans like to be amused. Therefore the dog pictures serve their highest and best purpose, which is not to pinpoint the precise attainments of AI dog picture generation in June 2020 or educate people as to how long it might be before AI can generate wholly believable pictures of non-existent dogs. That sort of evaluation is better made via papers published in academic journals than on a Twitter feed.

are u sure?

Neither the text or the images are believably “human” to me, though the text falls into a sort of prose uncanny valley while the pictures are all just deeply fucked up

i'm pretty sure the text was AI-written (and many of you didn't notice, heh!)

in fact, the entire post dan l posted was all about the text of those tweets - the AI generated visuals barely warrant a mention

Dan I, sorry!

My bad, the dog ratings text is generated, but the alt-text (which I do not see and do not know how to view) is human written. She doesn't really do a good job of explaining all that, tbh. It'll be fun to play with the API first-hand.

and with that, i believe that AI has now reached 100% level

we are now in the age of 3fa23

hold your cursor motionless over the image for the alt text. I would've been more impressed if that was machine-written as it seems genuinely aware.

yikes
http://www.shardcore.org/shardpress2019/2020/06/17/algonuts/

face depixelizer

🤔🤔🤔 pic.twitter.com/LG2cimkCFm

— Chicken3gg (@Chicken3gg) June 20, 2020

One reason this thread title is correct is that so many AI researchers fervently believe in long-discredited pseudoscience: https://cacm.acm.org/careers/244713-facial-recognition-software-predicts-criminality-researchers-say/fulltext

That is so gross. Predicting "criminality" by looking at a photo is bias-reinforcing bullshit.

With 80 percent accuracy and with no racial bias, the software can predict if someone is a criminal based solely on a picture of their face. The software is intended to help law enforcement prevent crime.

Each of these sentences is horrifying. But the second is by far the worst. The amount of unselfconscious intellectual depravity exhibited in this press release is bleakly stunning.

The thing where these companies simultaneously say that their systems are accurate but also impossible to quantify (what exactly is the AI seeing that is a marker for "criminality?") is some scary shit.

Like if it's not racial bias, what is it? Phrenology? Eye separation? Nose size? Hairstyle?

Yes, even a hint of the police putting this into practice is deeply disturbing. I'm also horrified to be (again) reminded that you can advance quite far in CS without learning the absolute basics of the history of science

i hope you all don't pay attention to china

I know you're just interrogating their bullshit, but it is obviously racial bias. Like what could they have possibly trained this on other than mugshots?

yeah totally.

The word "criminal" should be outlawed as a noun. It is an adjective and only applies to actions, not people.

In less creepy AI news, researchers have made an AI that can watch people play Pac Man and then recreate a playable game on its own without any help. Pretty amazing:

https://www.youtube.com/watch?v=3UZzu4UQLcI

I love the two-minute papers YT channel. The guy is so enthusiastic!

"What a time to be alive!"

I have no idea if any of this is actually GPT-3 output or not, but some of it is pretty funny https://arr.am/2020/07/09/gpt-3-an-ai-thats-eerily-good-at-writing-almost-anything/

improved interactive D&D chat bot.

https://medium.com/@aidungeon/ai-dungeon-dragon-model-upgrade-7e8ea579abfe

Example:
https://play.aidungeon.io/adventure/4e00e6e2-1ced-4681-bea4-d5bda66dd4c6

https://miro.medium.com/max/678/1*bs9aJJLm7ZGngUZ81bp1Gw.png

yo
https://player.vimeo.com/video/426819809

i am not an NLP specialist and i don't work on AGI at all, but this thread is correct IMO

listen.... ... the text / code / etc generation from GPT-3 *is* super impressive but you have you remember its basically just a real good search engine

— everest (@everestpipkin) July 18, 2020

did talktotransformer use gpt-2? that was unquestionably a glorified search engine. if the examples are legit then gpt-3 seems to cover its tracks a hell of a lot better but hard to imagine there's been a sudden quantum leap in machine understanding. (to be clear: gpt-2 understands diddly-squat.)

yeah, t2t used GPT-2 models, of various sizes at different times.

there is not a ~qualitative~ difference in the structure of the GPT2 and GPT3 models iiuc.

it's not even _that_ great a search engine ("which is heavier, a banana or an airplane" "a banana" etc)

Not that this proves anything, but the guy in the video I posted specifically had GPT-3 write functions that (although very simple) were novel and couldn't have just been a search engine result.

Not that this is AGI, but I'm convinced that long after AGI exists, people are going to say that it's not AGI, and if it is, it's no big deal.

so what if the computer's smarter than me, a calculator can do much longer division than I can

could the functions have been a few different search engines results mashed together? how many times did he try it and were all the results valid functions? talktotransformer came up with new and unique sentences but they were 100% a product of the input.

right. he's not saying it's literally a search engine. this is the money tweet in the thread

i use AI in my own work, and i'm moderately positive about its capacity as a tool. but what I'm truly excited about is not the capacity to launder knowledge into new mush, but to organize knowledge in ways that are transparent, traversable, and human readable

— everest (@everestpipkin) July 18, 2020

oh god looking back at the brutal and stupid ai and they've found a way to make an ai that thinks just like me

What would be an example of organizing knowledge in ways that are transparent, traversable, and human readable, because it seems like writing a novel computer program to specific requirements does that. People are having it write front-end apps that "compile" (or whatever you call it with JS) to working websites.

just in case ppl haven't seen these examples which we're kind of discussing the responses to without having explicitly posted in the first place:

This is mind blowing.

With GPT-3, I built a layout generator where you just describe any layout you want, and it generates the JSX code for you.

W H A T pic.twitter.com/w8JkrZO4lk

— Sharif Shameem (@sharifshameem) July 13, 2020

I just built a *functioning* React app by describing what I wanted to GPT-3.

I'm still in awe. pic.twitter.com/UUKSYz2NJO

— Sharif Shameem (@sharifshameem) July 17, 2020