Reminder that once AI tycoons are done training their models, inference is literally a money printing machine and profits will be insane
model training is never done, you need to constantly re-train with newer cutoff dates or your model will be stuck in the past
They literally cannot stop training before AGI, local models are improving twice as fast as frontier models are.
>>108733547models have internet search tools now, boomer
>>108733530Two more weeks
>>108733810i know. that works for a while, but eventually you can't fit a few years of history in a 1M context window.imagine using an LLM trained up until 1990, it would be useless in today's world even with web search, it wouldn't even understand any modern concepts
>>108733983it will take a long time before the world changes so much that a few web searches is not enough for the model to update its understanding of how things work anymore
>>108733530Its not gonna work, we are at least 60 year to a working real AI.And that is gonna become Skynet in probably 48 hours.
>>108734070i disagree
>>108733530>>108733810Imagine asking grok about current events and getting replies about week/months old events, or being told that everything that happened a couple of days ago is fake.Good luck removing inference from commercial LLMs.
CC more like SexSex
>>108733530>take a dollar>take a photocopierliterally a money printer.
