529 karma · joined February 20, 2022
I think this results in quite different cultures.
Does it sound surprising to you that it was perfectly normal to rent a perfect acceptable two bedroom apartment in a safe town on the interstate for $300 a month and still easily find dignified, decent paying jobs without 1000 applications?
I've lived in many cities and work in tech now, and I can confidently say that, as it concerns the professions and jobs that unambiguously sustain and improve life, no community on the planet was more productive than my home state. There is more to the story than some shale.
I loved the winters. I loved the people. I loved how its natural beauty was subtle and rewarded the patient, unlike El Capitan or the Black Hills. The economy was fine before oil appeared.
> Block 1: We couldn't calculate fast enough. Solution: The GPU.
> Block 2: We couldn't train deep enough. Solution: Transformer architecture.
> Block 3: We can't "think" fast enough. Solution: Groq’s LPU.
#2 is outright wrong. Deep networks were made viable from residual layers and their refinement. #3 is also incorrect; "think" = compute so this is the same statement as #1.
Also, the "limestone" analogy is pretty weak.
Any feedback is super appreciated.
I'm aiming it at new college grads who haven't had to maintain several services at once in production.
I think that one’s opinion of this article’s premise is really an index into what your program was like.
I totally forgot that it has a readable (I.e. guessable domain name) because AWS’ equivalent service doesn’t. I also had a company subdomain pointing to it so someone got to put up a malicious page on our domain for a day :(
I sort of take this as a compliment, because I've been writing like this my whole life and if it reads like LLM slop, then there's the implication that the result of all of OpenAI's A/B testing and post-training leads to something like my style which at least means it gets people to engage with it!
We had this project (all public research) to classify buildings and identify their different subsystems (e.g. load-bearing structure, roof type, ventilation type) to figure out the expected casualties if there was a WMD event of some type. We could get decent data for much of the world, but for some places we had literally nothing beyond a tiny picture of it from satellite imagery.
I had been playing with using GPT-3 to try to have it autocomplete forms like the following. This was 2021 before we had good APIs for instruct models, so this was just straight up letting the LLM regurgitate after pretraining. Here was the type of prompt we used:
""" Engineering building report for building located at 123, X Street, Knoxville TN Prepared by Benjamin Lee, FE --- Building footprint area: 1200 m2 Roof type: built-up roofing Facade material: brick HVAC present: """
Surprisingly (at the time), this was a decent prior. You could also add all sorts of one-off points of interest and amenities like swimming pools and other trivia to help guide the conditional probabilities.