In this atmosphere of excited tweets about simulated conversations with an 'AI', let's not lose sight that they are mostly *language* models.
They model language as a sophisticated, human-produced data layer of extreme complexity, but *not* of consciousness or sentience.
For one, I wouldn’t be able to do any of this if I didn’t know how to code. Right now it enables me to do stuff I already know how to do, much faster.
But there’s a lot of built up expertise that goes into knowing what to ask it to do, and understanding its output.
@julien_c I doubt it. And it was never "engineering." It was closer to a copywriting craft based on hearsay. That said, prompt "chaining" and model combination flows is where we're already going.
the best ideas i ever shipped came from engineers on my team. an engineer's job is NOT only to code. they are incredible problem solvers and are ridiculously creative.
they understand the tech, know what is possible, and will be building it.
involve them early and often!
@schmatzarella@robinysmith I also experienced the “great vector awakening.” Mine is built with a Go back end, but the results are just as impressive, using OpenAI’s embedding and Pinecone. It’s worth diving into the math of computing vectors and their similarity formulas, and which makes sense for you.
BREAKING: Herschel Walker claimed on financial documents that he’s a resident of Texas, not Georgia. So he’s apparently commiting tax fraud, election fraud - or maybe both. 😂🤣 https://t.co/8t3ydSIJcw
It's a pattern that occurs at many startup orgs, usually during a product crisis. Followed by farming off engineering work, which often fails and wastes a ton of money. Then if you're lucky, there will be a shakeout at the top, and you get the chance to hire the right talent.
The most ill-conceived idea that seems to have ossified among certain group of people is that Twitter was failing because the engineering culture wasn’t high-performing enough
The problem with Twitter was product and leadership, and not engineering not executing well enough.
@realGeorgeHotz Agreed with @BorisMPower. I've been working with vector embeddings for semantic search, and my results are way better than when I used full text search on inverted indexes. Much easier too, and OpenAI has an embedding API to help with the calc, but you can embed tweets yourself.
The toxicity flows from above that manager, usually by one cloaked by a political veneer. And in that situation, if you’re not a toxic manager, you will be pushed out yourself.
I don't have data to support this (and I'd like to see responses with it), but in the tech industry, layoffs during recessions beget the next generation of huge startups.
@sonyatweetybird@sequoia Perhaps I missed an in-depth examination of how this was derived, but why would OpenAI's offerings not included? I can see reasons why they would, or would not be, but I assume you have a reason you might share.