hiyou102

Even in that article with much smaller vectors than what GPT puts out (1536 dimensions) QPS drops below 100 if recall@1 is more than 0.4. That's to say nothing of cost of regenerating this index using incremental updates. I don't get why people on HN are so adamant on the idea that no one needs scale beyond 1 machine ever.

A lot of people use some form of managed services if they are in the cloud. Be it S3 or Dynamo DB. Generally cheaper than running things yourself and operationally much easier too.

What kind of QPS are you getting and how fast are incremental index updates? That's the hard part.

What kind of QPS are you looking at? How are you handling 1536 dimensions? How long does an incremental index update take? These are the problems you run into in building such a system.

It seems weird to do this over a feature that is still in technical preview, we don't even know if this product will ever ship publicly. I'm guessing a public release is still years off given the number issues they need to work through before release. My understanding is that they are working on an attribution system to catch cases with common code. Beyond that this person seems to use the MIT licensed code which already can be used internally by a company to host a proprietary service without attribution. It would make more sense to be outraged if you were using AGPL or something.

What specific issues are you having? What do you mean by "B-class players, who lead all sorts of C-, D- and all the way down to Z-class characters"? I assume in this ranking you are an A+ player yourself, which is very impressive.

Microsoft doesn't use a monorepo. Bloomberg, Dropbox, and Amazon don't either.

I'm a little surprised that there isn't a single "top" YC company in Seattle. Why is that? It seems like the city has talent and capital.