Post

Michael Odden
@michaelodden.com

Ah... I'm old enough to remember the sweet time pre-GPT2 when the majority of the Internet was still not _mainly_ generated garbage, and thus was to a certain extent searchable. 🎶You don't know what you've got 'till it's gone🎶

AB@alexbunardzic.bsky.social

People are complaining how LLMs are burning so much energy. But compared to the energy I caused google to burn today while searching for hours for a specific quote while it was feeding me shit results,

Michael Odden
@michaelodden.com

To be fair: that was a quite good use case for LLMs. Yet we are building shitloads of new DCs to host them, as the amount of requests, and the amount of thrown away computations done until a successful outcome is reached increases. That's just for inference. Then there's training too.

Michael Odden
@michaelodden.com

Yet, personally I find the energy argument in isolation to be among the least interesting ones when discussing all the blameworthy aspects of LLMs (and related products).

0 reposts0 likes