3 ms·
The author thinks that ChatGPT and similar systems can easily be scaled up because they only cost about 5 million dollars to train. However, the author seems to
by uejfiweun 4y ago
The author thinks that ChatGPT and similar systems can easily be scaled up because they only cost about 5 million dollars to train. However, the author seems to ignore the cost of inference in his analysis. I've seen figures saying it costs OpenAI $0.02 per query to run ChatGPT. Until they solve that bottleneck, nobody's gonna be massively scaling up LLMs by a factor of 100 anytime soon.