Where might I find the most accurate breakdown of all the sources of text used to train GPT-4?

#ML #LLM #GPT4 #neuralnet #linguistics

That time I tried to evolve a set of neural networks to control Chicago Traffic and the damned thing cheated.

https://hlfshell.ai/posts/evolving-traffic-control/

#ai #neuralnet

Anyone in the #ML or #neuralnet or #design spaces seen any examples of models that have been trained on graphic design exemplars that can recognize typefaces and content hierarchies on a screenshot or page?

I'm interested in trying to build a Stockfish-like model that operates on classic principles of layout, composition, and typography.

“If we allow it to take over, it will be bad for all of us,” Hinton says.

2 things we can say about this:
1) I'm unsure if we know what "it" is.
2) We don't know if "it" wants to take over.

#AI #ArtificialIntelligence #neuralnet

https://mitsloan.mit.edu/ideas-made-to-matter/why-neural-net-pioneer-geoffrey-hinton-sounding-alarm-ai

[ Disclosure: I have done contract work on the product being advertised, but reposting this ad is not part of that. I just think it's a neat demonstration of AI being used responsibly. ]

#AIDungeon, once again, showing us a safe use of #AI which the #NeuralNet / #LLM model is actually well suited for. #GPT and similar #Transformer programs will likely never be good at reliably simulating a therapist, a friend, or even an honest person, but they're great at outputting stories.

This time, the #Latitude PR team has the AI #writing a #surreal "#FeverDream" adventure, where we follow #anthropomorphic rabbits into a neon forests, and ride glittering unicorns away from glowing beasts.

https://youtu.be/aywL3l7VCMo

Acabei de assistir uma bela introdução à redes neurais. Andrej nos conduz passo a passo no cálculo da derivada de cada peso da rede neural individualmente para demonstrar backpropagation antes de abstrair todo o código para uma classe de forma mais enxuta.

https://www.youtube.com/watch?v=VMj-3S1tku0

#neuralnet #pytorch #redesneurais #InteligenciaArtificial

Creating training data for the neural net that will enable visual search in #TinyUX.

About 220 icons, 100 drawings / icon.

I've been doing this for many many hours in the evening while watching a football match or something.

I don't know if the feature will be used much, but wanted to get experience in embedding a neural net in an app that was actually used. Not just a demo.

Https://tinyux.app for more about the app.

#ux #wires #wireframes #ai #neuralNet #icons #reactNative

Making good progress with my visual search feature in TinyUX.

The simple neural net is stored on the device. It is now only trained on letters and numbers. Will need to train for all icons.

I use the same interface to draw content for the neural network to train on. Still quite number of hours of work ahead for creating that training content.

#neuralnet #ux #icons #pytorch #reactnative #tinygrad #tinyux

Why Neural Networks can learn (almost) anything #NeuralNetworks #NeuralNet #AI #DeepLearning
https://www.youtube.com/watch?v=0QczhVg5HaI

This seems really good to start going...
https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
#stephenwolfram #wolframalpha #chatgpt #neuralnet #neuralnetwork #llm #science #ai #mathematica #brain

Micrograd is very simple, only fully connected layers. So first trying to find out if it can even learn numbers based on MNIST dataset.

Then I hope to at least be able to verfit, so the essence works. Then I'll have the challenge of trying to make it work for every icon in the app.

Presumably I have to create/generate a huge set of icon images to train on..

#ux #micrograd #reactNative #MLP #MNIST #TinyUX #Karpathy #ai #NeuralNet

In react native, trying to convert an image into an array of color values for each pixel (or just floats for how dark). But can't figure it out yet.

I use react-native-view-shot to get either uri (I can place in image component), base64 (think it's png encoded), or 'raw' string.

My goal is a-typical I guess.. I want to use that array as input for a simple neural network to classify the image.

#reactNativ #base64 #raw #javascript #image #convert #neuralnet

My plan for 2023 is to learn AI by learning from Andrej’s videos and then making “slowed down” versions that also teach you Python!

#AI #ArtificialIntelligence #ML #MachineLearning #coding #programming #JupyterNotebook #LLM #LargeLanguageModel #NeuralNet #NeuralNetwork #python #video

https://www.globalnerdy.com/2023/01/19/want-to-learn-how-to-make-ai-ml-programswatch-andrej-karpathys-ai-tutorial-videos/

My plan for 2023 is to learn AI by learning from Andrej’s and then making “slowed down” versions that also teach you Python!

#AI #ArtificialIntelligence #ML #MachineLearning #coding #programming #JupyterNotebook #LLM #LargeLanguageModel #NeuralNet #NeuralNetwork #python #video

https://www.globalnerdy.com/2023/01/19/want-to-learn-how-to-make-ai-ml-programswatch-andrej-karpathys-ai-tutorial-videos/

Love this collaborative talk by @skmorgane and @ethanwhite on their work remotely monitoring #wadingbirds in the #everglades: https://www.youtube.com/watch?v=9HXZEaxdbUo #AI #machinelearning #neuralnet #ecology

#ChatGPT: in what ways can you improve your own #neuralnet #architecture ?

#gpt3 #openai #ai #transformer #selfattention

Rain Demonstrates AI Training on Analog Chip

https://www.eetimes.com/rain-demonstrates-ai-training-on-analog-chip/

#analog #continuum #continuity #Neuromorphics #Memristor #neuralnetwork #AI #neuralnet

Bumblebee is out, I think we’re gonna have some fun.
https://news.livebook.dev/announcing-bumblebee-gpt2-stable-diffusion-and-more-in-elixir-3Op73O
#elixir #neuralnet #gpt2 #ml

There are no stored images. None. The size of the #NeuralNet weightings are exactly the same, whether there was one training image or a trillion.

-rw-rw-r--. 1 stablediffusion stablediffusion 4265380512 sep 28 21:41 models/Stable-diffusion/sd-v1-4.ckpt
-rw-r--r--. 1 stablediffusion stablediffusion 4265380512 okt 20 00:39 models/Stable-diffusion/v1-5-pruned-emaonly.ckpt

With SD being trained on billions of images, & the checkpoints being only 4GB, it should be obvious, it's NOT storing images

Never confuse an A.I. with a neural network or pattern recognition. If you don’t know the difference, you either work in marketing, or shouldn’t be working on that project.
#ai #neuralnet #patternrecognition #failure