But wasn’t ChatGPT trained on huge amount of data from social media, Reddit etc.? With distributed I mean that if the server fails the LLM still stays alive, like this. So I thought about fail-safe. Maybe I didn’t thought it through enough …
But wasn’t ChatGPT trained on huge amount of data from social media, Reddit etc.? With distributed I mean that if the server fails the LLM still stays alive, like this. So I thought about fail-safe. Maybe I didn’t thought it through enough …
sparse_neuron@lemmy.world 1 year ago
What do you mean by “stay alive”? LLMs are statistical models that require a lot of number crunching in order to output a response. That’s not something you can just host without some major costs associated (Far more than any single Fediverse instance for sure, GPUs are not cheap to rent).