Post

Conversation

The ability of multimodal AI to โ€œunderstandโ€ images is underrated. I just took these. Given the first photo Claude guesses where I am. Given the second it identifies the type of plane. These arenโ€™t obvious.
Image
Image
Image
Image
David Watson ๐Ÿฅ‘
Post your reply

#5 - AI = Freedom? Only if the AI is Truth - New tech innovations shaking up our world ๐ŸŒ๐Ÿ“ˆ The unveiling of Grok 2 from X๐Ÿš€โœจโ€”a fully uncensored large language model with an enterprise API ๐Ÿง ๐Ÿ”“. FLUX image generation open source model๐Ÿž๏ธ. We get philosophical and practical
Show more
1:14:49
I created one that you take a picture of damage on your car and AI works out a rough estimate of the repair costs. Just with a couple pictures. Itโ€™s amazing. Iโ€™ve only I knew how to monetize it.
Interesting as always that the way the question is asked is important too. I first asked โ€œwhere am I?โ€ and got a different answer than you but then I used your exact question and got your answer.
Image
Image
I like to take photos of the ground when Iโ€™m in the airplane and asked chatgpt to guess where I am. So far, itโ€™s guessed it everytime. First photo was over Brooklyn. The second was Mt. Rainier.
Image
Image
Image
Image
It can even somewhat accurately guess specific ethnicities from faces alone (Iโ€™ve only tested with different African faces donโ€™t know how well it works for other regions)
It often uses gps metadata to guess where you are, so if youโ€™re gps coordinates put you over the ocean, itโ€™s a giveaway ;)
Any chance it used the location data in the image file to place you at an airport or in the air? With a time stamp could it use publicly available flight trackers to narrow down the aircraft type? Just being pedantic here!
Haha I do you one better
Quote
Josh Olin
@JD_2020
Replying to @rochester
(ROC native here btw, this is not spam ๐Ÿซถ) I provided ChatGPT this image, with very clear instructions. Since this image is so new, it will not have been part of the scraped data that the AI models trained on. In other words, the language / vision AI model has to look at the
Show more
Image
Image
Image
Interesting, there are no clues from the angle, I almost thought it was from an office looking at the floor at the base of the photocopier.
Multimodal AI is definitely leveling upโ€”going beyond text and grasping visual context is no small feat. The ability to understand images like this opens up a ton of possibilities, especially in areas like real-time navigation or image-based research. Whatโ€™s more impressive to
Show more
yep. these digital neural nets clearly understand things in ways that are analogous to our understanding of things, with better memory. and the larger models have broader knowledge than any individual human. in some ways we are already seeing some superintelligence
One example from me too
Quote
Crypto Pill - e/accโซ
@AGIapproaching
Replying to @kimmonismus
I did, and it's crazy. ๐Ÿ˜ฏ Took a screenshot of the original photo, cropped it, and checked for the metadata to make sure it's not there. Right answer: solang valley and rohtang pass.
Image
Image
i always froth at the mouth thinking about how someone out there has access to these models completely uncensored; surely they are near perfect doxxing tools?
What makes this ยซ crazy ยป? We all know that itโ€™s based on the training data, it has nothing to do with some intelligence or reasoning..
with a little promoting (without steering it in any direction but just asking it to make deductions), gemini came up with this answer.
Image
did it identify the type of plane accurately? all i got from gemini (previously the best vision model imo) was that it was either Boeing or Airbus which isn't really narrowing it down by much, most planes are either Boeing or Airbus
One use case I just enjoyed was taking a picture of a bouquet of flowers and then ask it to describe every flower - it does a great job!
Did you remove the metadata tags from the photo first? Just wondering if it guessed plane because your geo location had you on a runway ๐Ÿค”
This past weekend my blind MiL just showed me an app that, after taking a photo, can describe the entire photo in detail including minute specifics of background objects. I was flabbergasted.
tbf aircraft and airports are very easy to guess because they're uniform. It's so bad that PDX's carpet design received national attention simply because it didn't look like every other airport on earth. That said, still very impressive from Claude
See if it can figure out what class you are in. You appear to be in a premium class since the middle vents are covered up inicating just two passengers. This is a nice deductive reasoning challenge & goes beyond info that could be memorised from training data
I have been blown away by its ability to transcribe from handwriting - even Victorian copperplate. Claude had no problems with this.
Image
I could have told you it was an a320 series aircraft by the first image, but I fly 120 days+ a year.
impressive๏ผŒbut i wonder how many usecases there are? how often do we need to identify the exact subject model besides shopping?๐Ÿค”
one image analysis costs only 1/10 of a penny. the challenge to think of a practical real-world applications these dirt cheap AI vision models could be used for.
Now put this Claude model to work on LIDAR scans and who knows what we will find? link :
Quote
Stone Age Herbalist
@Paracelsus1092
New LIDAR results have revealed two highland urban centres in southern Uzbekistan. These silk road medieval cities may have included fortifications and researchers speculate that one may 'have been a factory where local metalsmiths turned rich deposits of iron ore into steel'
Image
Image
Image
Image
such a great use case. how could society possibly function if AI couldn't give all that information from those pictures? there would be chaos in the streets.
Iโ€™m scared. I canโ€™t keep up with that. I know Iโ€™m not really important but Iโ€™m still kinda worried that I wonโ€™t be able to buy food or shelter.

Discover more

Sourced from across X
I put the new 3.5 sonnet and the old 3.5 sonnet into a Minecraft build-off. The only reliable benchmark Left: New 3.5 sonnet Right: Old 3.5 sonnet
Image
Image
almost all gen-z free time goes straight to video games, streaming platforms, and shorts/youtube 1/3rd of gen-z males play video games for >5 hours/day! (source: Jure Grahek, ZBD)