HTTPSConnectionPool(host='translate.google.com', port=443): Max retries exceeded with url: /m?sl=en&tl=zh-cn&hl=zh-cn&q=Hi,%20everyone.%20So%20I've%20wanted%20to%20make%20this%20video%20for%20a%20while.%20It%20is%20a%20comprehensive%20but%20general%20audience%20introduction%20to%20large%20language%20models%20like%20ChatGPT.%0AAnd%20what%20I'm%20hoping%20to%20achieve%20in%20this%20video%20is%20to%20give%20you%20kind%20of%20mental%20models%20for%20thinking%20through%20what%20it%20is%20that%20this%20tool%20is.%0AIt%20is%20obviously%20magical%20and%20amazing%20in%20some%20respects.%20It's%20really%20good%20at%20some%20things,%20not%20very%20good%20at%20other%20things,%20and%20there's%20also%20a%20lot%20of%20sharp%20edges%20to%20be%20aware%20of.%0ASo%20what%20is%20behind%20this%20text%20box?%20You%20can%20put%20anything%20in%20there%20and%20press%20enter.%0ABut%20what%20should%20we%20be%20putting%20there?%20And%20what%20are%20these%20words%20generated%20back?%20How%20does%20this%20work?%20And%20what%20are%20you%20talking%20to%20exactly?%0ASo%20I'm%20hoping%20to%20get%20at%20all%20those%20topics%20in%20this%20video.%20We're%20going%20to%20go%20through%20the%20entire%20pipeline%20of%20how%20this%20stuff%20is%20built,%0Abut%20I'm%20going%20to%20keep%20everything%20sort%20of%20accessible%20to%20a%20general%20audience.%0ASo%20let's%20take%20a%20look%20at%20first%20how%20you%20build%20something%20like%20ChatGPT.%0AAnd%20along%20the%20way,%20I'm%20going%20to%20talk%20about,%20you%20know,%20some%20of%20the%20sort%20of%20cognitive,%20psychological%20implications%20of%20these%20tools.%0AOkay,%20so%20let's%20build%20ChatGPT.%0ASo%20there%20are%20going%20to%20be%20multiple%20stages%20arranged%20sequentially.%0AThe%20first%20stage%20is%20going%20to%20be%20the%20pre-training%20stage.%0AAnd%20the%20first%20step%20of%20the%20pre-training%20stage%20is%20to%20download%20and%20process%20the%20internet.%0ANow,%20to%20get%20a%20sense%20of%20what%20this%20roughly%20looks%20like,%20I%20recommend%20looking%20at%20this%20URL%20here.%0ASo%20this%20company%20called%20Hugging%20Face%20collected%20and%20created%20and%20curated%20this%20data%20set%20called%20FineWeb.%0AAnd%20they%20go%20into%20a%20lot%20of%20detail%20in%20this%20blog%20post%20on%20how%20they%20constructed%20the%20FineWeb%20data%20set.%0AAnd%20all%20of%20the%20major%20LLM%20providers%20like%20OpenAI,%20Anthropic,%20and%20Google,%20and%20so%20on,%0Awill%20have%20some%20equivalent%20internally%20of%20something%20like%20the%20FineWeb%20data%20set.%0ASo%20roughly%20what%20are%20we%20trying%20to%20achieve%20here?%0AWe're%20trying%20to%20get%20a%20ton%20of%20text%20from%20the%20internet,%20from%20publicly%20available%20sources. (Caused by ConnectTimeoutError(, 'Connection to translate.google.com timed out. (connect timeout=300)')):Google
=====
Windows-10-10.0.26100-SP0
version:v3.78
frozen:True
language:zh