Oh yeah, 100% true, “amount of traffic does not even remotely map to amount of actual work being done especially on such a horrifically inefficient system” was a point I meant to bring up but I sprung a leak in a blood vessel from sheer rage before I could even get there. Thank you.
Some agent may do 100 queries for 1 user question, but a single scraper bot will do 100 queries every single second of every single day, and costs less to run than the LLM.
A lot of times their training is lacking in whatever I’m trying to get them to do, so I have to direct them to read particular web pages in order to have a clue what they’re supposed to be parroting.
I mean have you used an agent? They are fucking dumb and make a ton of queries to get anything done.
Oh yeah, 100% true, “amount of traffic does not even remotely map to amount of actual work being done especially on such a horrifically inefficient system” was a point I meant to bring up but I sprung a leak in a blood vessel from sheer rage before I could even get there. Thank you.
Some agent may do 100 queries for 1 user question, but a single scraper bot will do 100 queries every single second of every single day, and costs less to run than the LLM.
Fair point. Scrapping is definitely the low overhead high impact problem.
Just pay for more cloud compute that will fix it…
Would you like to know more? That’s gonna cost ya.
As an aside the recent trend of using compute as a noun instead of a verb really pisses me off, it’s so goddamn stupid.
I agree. Though since my comment was intended to be annoying (in the same way the current reality is) I shall stand by its use here.
A lot of times their training is lacking in whatever I’m trying to get them to do, so I have to direct them to read particular web pages in order to have a clue what they’re supposed to be parroting.