"Amazon Ditches Just Walk Out Checkouts at Its Grocery Stores
Just Walk Out relied on more than 1,000 people in India watching and labeling videos to ensure accurate checkouts.
700 out of 1,000 Just Walk Out sales required human reviewers as of 2022.
'The primary role of our Machine Learning data associates is to annotate video images, which is necessary for continuously improving the underlying machine learning model powering'"
I'm a bit early, but I'd like to wish all of you a very HAAPPPY NEWYERAR!
I spell real gud because Adode Photoshop generative fill helps me!
So, I know Adobe Photoshop's generative fill feature is primarily for generating missing background elements and not for creating entire images, but this is still pretty bad.
The prompt I used was "Light colored Happy New Year sign with sparkling confetti streaming behind it."
I then tried a simpler prompt and it gave me a simpler image, but still misspelled Happy New Year badly.
Earlier today, I was looking for some #GameDev inspiration and found this page. Who else remembers classic games like "Tetris", "Space Invaders" and "Client-server game"? Vivid memories! ❤️ #ChatGPT#AI#AIFail#WebPollution
Then I went to have a look around Pixabay, and found a darkly hilarious #AI-infused rabbithole of fuckery.
Pixabay, if you didn't know, is a stock image site. Which now features a below-image bar featuring ready-made AI #Canva graphics for any image. ANY image, including staged CGI #NuclearWar ones.
Presenting what will be my holiday card: WARMEST GREETINGS! Very, very warmest greetings.
That the "merry nukemas" card is signed by "Gray" just...
@theconversationau "just a really, really good next-word predictor."
It should be made more explicit. A tool that predicts the next word is doing so based on lots of training data made by people, and scraped by search engines. In that training data there are lots of examples of what word comes next. So it's feasible to train a model using it. But there is no way to know whether the sequence of words relates to something factual, or even whether they were created by a person with a good grasp of the language they are using. There is also no way to tell whether the training data might have been poisoned in some way, by another 'AI' for example.
Such predictions are going to be based on what a model has already seen. Given this there is unlikely to be anything new of substance in synthetic text. Just endless grammatical re-hashing of existing ideas and concepts, including false ones.
This is so far from what we would call intelligence, that we should really stop referring to it as such. #AIHype#AIFail
According to the #iNaturalist automated classifier, this is a “common hoopoe”, a bird we Catalans call a “puput”. My laugh was heard from the other side of the mountains. Opening up to suggestions from beyond nearby then yielded a springtail genus—what I initially expected.
Seira sp. springtail (tentative) http://www.inaturalist.org/observations/178251249 Ridiculously small. Only saw it because I was staring at a crevice cricket and these critters were scurrying around under the strong phone light
Night photography isn’t my thing—no equipment, no training, no practice, and barely ever the occasion.
"AI" companies think that we should have to opt-out of data-scraping bots that take our work to train their products. There isn't even a required no-scraping period between the announcement and when they start. Too late? Tough.
Not acceptable. #RequireOptIn
Here is a confident Hinton predicting in 2016 that "within 5 years, all radiologists will have been replaced by deep learning technologies": https://www.youtube.com/watch?v=2HMPRXstSvQ
These days he's warning people of an "AI" apocalypse coming soon (maybe within 5 years?). And, no, he doesn't mean flooding the internet with fakes.