by Andrej Karpathy
Here are the 6 most clip-worthy moments — auto-detected from the transcript. Tap a timestamp to jump straight to it on YouTube.
🎬 Turn these moments into shareable vertical clipsCaptions, 9:16, ready to post — early accessIt presents a surprising fact that the vastness of the internet, when filtered for text, is a manageable 44 terabytes.
Caption You think the internet is HUGE? 🤯 The *text* data for LLMs is only 44TB. Mind blown! #AI #LLM #DataScience
It reveals the crucial filtering process that prevents LLMs from ingesting harmful or undesirable content.
Caption Think AI learns *everything*? Think again! 🚫 Here's what's filtered out to keep LLMs safe. #AIethics #DataFiltering #LLMSafety
It highlights how language filtering decisions directly impact an LLM's multilingual capabilities and potential biases.
Caption Why is your AI better at English than Spanish? 🗣️ It's all in the language filtering! #AIbias #MultilingualAI #LLMdevelopment
It uses a powerful metaphor to describe the vast, raw text data that neural networks learn to model.
Caption Imagine the entire internet as a giant tapestry of text. 🤯 This is what neural networks learn to mimic! #AI #NeuralNetworks #BigData
It simplifies the abstract concept of bytes as unique symbols by likening them to emojis, making it highly relatable.
Caption What if every byte was an emoji? 🤯 This is how computers see data! #DataScience #Emojis #TechExplained
It explains the clever algorithm (Byte Pair Encoding) used to efficiently represent text for neural networks.
Caption How do LLMs handle massive amounts of text efficiently? 🤔 It's all about Byte Pair Encoding! #LLM #Tokenization #AIhacks
Generated from the full transcript of this video · 2outube — change youtube to 2outube on any video.