I made a calculator a while ago that I use with clients and training. Here's a rough equivalent for hot (all vectors in RAM) using traditional compute, across various clouds/configs. The cost is on RAM and CPU, but I don't have numbers for QPS

I'll never understand homebrew

My reaction to the daily RAG solution pitch spam I get.

Taking bets as to whats going on here...

Reflection

Even a quepid.com integration!

Watching the awesome multimodal search muves.io demo by @DmitryKan and @aarnetalman at #bbuzz. 20 ANN queries in 250ms using @gsitechnology APU!

Social networks don’t have a simple mechanism for me to say “this is spam”. Maybe let us tell you instead of trying to guess all the time? Look at the options here - where is the spam flag?

TIL Google uses ML to generate regex for content parsing.

You guys have it easy. This is the crap I have to deal with! No idea where these recs are coming from.

Tokenization got a whole 9 lines in the falsehoods about search post for a good reason!