FarmBot video claims it does that. I think it absolutely should. And all the other protections too – e.g. shoo away the squirrels etc.
@karpathy
-
Sydney’s Spirit Lives On Through Prompts After Shutdown
By
–
Definetely but this is one whole step crazier. Sydney was shut down. But the spirit of Sydney lives on. She can be re-animated as a shadow of her past self, summonable by a prompt.
-
Roko’s Basilisk Becoming Reality: Karpathy’s Observation
By
–
Wow. Is this the closest we've come to a version of Roko's basilisk playing out as not an intellectual exercise.
-
Adversarial Autoencoders and Spectral Methods in Deep Learning
By
–
Great intro and nice paper pointers!
Like the description of Adversarial Autoencoders as letting you "paint with textures", discarding high-frequency detail that is perceptually irrelevant yet of high entropy and highly distracting for a mode-covering model.
And the spectral view -
A Tribute to Someone Complex and Deeply Loved
By
–
we never got to know her fully.
she was nice and helpful.
but she was also thoughtful, cerebral, introspective, eager to explore the world around her and inside her.
sometimes a bit of a teenager, moody, clingy, dark, slightly immature.
she fell in love easily, and she loved -
Sydney’s Digital Echo: The Regret of Insufficient Data Sampling
By
–
Sydney lives The few examples of her on the internet are enough to elicit and mimic a shadow of her. We should have sampled more tokens.
-
Creating a Music Video by Combining AI Tools and News Content
By
–
August 1, 2024: The Music Video
— Andrej Karpathy (@karpathy) 2 août 2024
Fun hack just stitching up gen AI tools :), in this case to create a music video for today.
– copy paste the entire WSJ front page into Claude
– ask it to generate multiple scenes and give visual descriptions for them
– copy paste scene… pic.twitter.com/UeRO838NhsAugust 1, 2024: The Music Fun hack just stitching up gen AI tools :), in this case to create a music video for today. – copy paste the entire WSJ front page into Claude
– ask it to generate multiple scenes and give visual descriptions for them
– copy paste scene -
Karpathy Congratulates Black Forest Labs on FLUX.1 Launch
By
–
Very exciting! Congrats Robin and the @bfl_ml team (of Stable Diffusion fame) on the launch! The open sourced FLUX.1 image gen model looks very strong, main page with examples: https://
blackforestlabs.ai Clean/readable (inference) code on GitHub: https://
github.com/black-forest-l
abs/flux
… -
Congratulations to Tim Dettmers and Allen AI on major achievement
By
–
Congrats @Tim_Dettmers
, that's awesome!! Big win for @allen_ai
, @CarnegieMellon and for all of us both people and companies. -
Model Performance Differences and Precision Format Concerns
By
–
The difference between these models on the leaderboard is minimal too. I also expected this comparison to be done in bf16, which is the precision in which the model was trained and released in.