I am hearing about a specific group at Amazon focussed on next gen LLMs and sounds kinda cool and cutting edge work. Anybody working there hows the work pressure culture and quality of work? Do you like it and recommend? Looking to switch to good ML teams internally. #datascience #machinelearning #tech
Avoid they are f of sht!
^^ Can you elaborate? Specific feedback can be helpful
Hey! Yes you heard right. Can’t divulge too much here though. I am an SDM in the AGI org. Been here since last 4 years. DM me if you would like to know more about the space.
I'm going to bet that this is a "in office, in Seattle" type of organization?
You are correct
There might be some cool teams but most of it is teams with no vision and clarity on what to build. They were figuring out their existence last year about 6 ‘on the ago and still figuring out. It sounds cool because they seem to be the one team that is handed the ML baton but stay out of it if you are in a good place now or figure out a team with clear goals and has ML opportunity.
We work closely with them. They're from the old Alexa team. Not sure about WLB but tech wise I'd say they're solid.
F
Most of it is under the Intelligent Decisions org under AGI now. I was in a sister org, but worked with them frequently. A friend of mine has told me some horror stories that are frankly unbelievable - building a tier-1 service in a week, L10 exceptions to release without security reviews because "it'll slow us down", and managers shouting "we are red, you'll be allowed a weekend when we're green". It's scarily bad.
If anyone from AGI team is reading this ..... what % of your ML workflow (just model training + inference and not serving) is dependent on packages from huggingface ecosystem ? and how much are we building inhouse?
Not a lot. HF models are definitely one of the the inputs into the ML workflows but the underlying compute+packages are internal and proprietary (especially inference). We keep forking existing open source tools and libs for internal consumption whenever appropriate. There also exist 1 or 2 teams specifically for contributing back to open source.
The content of the work is definitely interesting, but job security would be another matter. If you don’t care about RIF/organization closure threat, you may give a try.
Run! Incompetent and in a clusterfuck!