5,051 karma · joined November 4, 2011
I think if you're past pgvector performance you won't be listening to a random guy talking about pgvector but have a good understanding of the space.
If you're new (like I was a few months ago) save yourself the time I wasted on the noobtraps I mentioned. It scales way better than the OpenAI API for my use cases.
Lucene (or rather elastic/open search) is way overkill for my needs
I'm coming at this from a perspective of a competent dev trying to build a tool with OpenAI APIs - which, from what I can tell, is a growing topic.
If you're experienced with building APIs and new to the LLM stuff - skip the langchain and chromadb nonsense. Just use the OpenAI APIs and pgvector.
Chromadb and langchain are usefull when writing notebook prototypes to get an idea of how this stuff works - but discard immediately after that phase and save yourself the trouble of porting later.
pgvector is even supported out of the box on Azure and AWS RDS.
Just spin up a docker container [1], add a vector column to your table and you're ready for embedding search.
[1] https://hub.docker.com/r/ankane/pgvector
If you're starting out with a prototype - do yourself a favour, steer clear of the chromadb examples with langchain. In fact steer clear of langchain in general :) Just go for OpenAI API and PostgreSQL+PGVector - you'll have to do some boilerplate - but the stuff in langchain is just terrible, you'll have to rewrite it and do the boilerplate at some point anyway and this stack is super simple to deploy.
It's boring, problems I'm solving are technically trivial. Most of the real problem is tied to a domain I give 0 shits about and am so bored by that I have to force myself to stay focused. Things I build have practically no value to me, and often no value to anyone except that someone decided to allocate a budget that way.
I would never work on these projects if it didn't pay well. I would prefer craft work if it paid the same.
I'm not complaining though ! I live very comfortably, I came from extreme poverty and I don't see any realistic alternative path that would end me in a better situation.
But I think expecting people to do most programming jobs out of passion is not realistic. If you're really good maybe you'll land in a role that gets you doing something technically innovating that also pays well but I'd say that's top few % of devs, most of us are in the dredges where enthusiastic people are a detriment (worst projects I've seen were a result of someone trying to be too clever for their own good, not lacking skill)
Anyway this seems like it would be killer for blender to integrate camera tracking.
Python was all over the place when RoR came out, meanwhile I only remember hearing about Ruby on random forum posts or by some enthusiast before RoR.
Python was popular as a teaching language, how it gained a foothold in academia.
I remember 20 years ago when I was starting out it was popular for game scripting (Lua was more embeddable but python was also used). plenty of first class interfacing methods to native code (still remember c++ boost shipping python binding generator).
Many tools shipped python as an embedded scripting language.
Python was used for creating build systems.
Not to mention any distro out there ships python, it replaced perl in that regard. Batteries included and being available helped it become the norm for random scripting.
Python was big before numpy/ML.
Otherwise seems like snake oil thing.
Otherwise you're just rationalising freeloader behaviour.
Spotify suggestions and playlists weren't a seller feature for me for years. Maybe I'm missing out on a few podcasts.
The rate of progression seems to be logarithmic - so we got "something looks plausible" but to get that last 10% it's probably going to cost more in HW than just using humans, unless there's some breakthroughs. Just like self driving cars.
My impressions at least looking at the developments from a sort of technical perspective - they are hitting all kinds of scaling problems - both in terms of data available, runtime complexity, hardware available, etc.
nVidia raking it in is a perfect example of how inefficient the whole thing is. Models seem to be doing fairly simple math computation (nowhere near the complexity of a general purpose GPU core required), things are limited by memory bandwidth and memory available. I'm sure a better hardware design specifically for transformer inference could make it cheaper and faster. But it seems like anything like that is years out for general market and nVidia is raking in on selling repurposed GPU architectures, and nobody else can even compete with that based on software stack alone.
People selling us on AI replacing software developers are getting fleeced for billions because they can't port their software stack to similar hardware from a different vendor...
IMO being responsible for other people is tiring and unrewarding, I just want a structure to tackle interesting projects without a middleman.
If you're a couch potato dev who sits all day and does no physical activity - just doing random shit with your bodyweight or empty bar will get you sore for the first month.
Overloading yourself with pointless movements that are meant to target specific muscle groups is completely irrelevant for the first year. Go to gym - find heavy stuff you can pick up - try not to do anything stupid.
SS is great because it has a low number of movements, volume is irrelevant when you get sore from a barbell squat, learning how to lift is deceptively complicated (breathing/bracing/muscle activation).
The article keeps straw-manning "longterm" this and that - SS is your first year in the gym (or less). By the time you're trough with it, unless you're totally oblivious, you'll pick up on other stuff when you have the time. And you'll have a decent intuition of how to lift, understand recovery implications, consistency in training, and a decent base strength. It's a great base to get you started and have an achievable progression framework that doesn't require much tutoring or time.
I've seen plenty of people get from 0 to nice lifting strength in ~1 year by doing SS. The only thing that's relevant is consistency - if you come in 2-3x a week and do something difficult for an hour for a year - you will see results - it can be total fucking around - you will still get results. It's good to start with SS type exercises because those well defined movements target your entire body, and if you learn to execute them correctly you reduce your chance of injury.
TL;DR - for any newbie out there that's in bad shape and looking to start gym - save yourself time and avoid "optimal muscle development" crowd. None of those arguments apply to you and are completely irrelevant. Don't trust me ? Go to the gym and do 5x5 with an empty bar and see if you're sore after the workout.
And who do I trust more - company that bundles third party crap ware and openly talks about OS ads as monetization avenue. Or a company selling me xxxx$ machine and wanting to sell me a next one with the OS in the future. As much as I dislike Apple walled garden - their incentive structure is way more aligned with me than Microsoft desktop.
> Then Windows would be better than MacOS in this regard because WSL2 is exactly Linux, not just "fairly similar" to Linux.
Haven't seen that be better than a VM.
Doing anything WSL on Windows FS is dog slow and vice versa, split toolsets conflicts (different git config between host and guest, different SSH). Just SSH into a VM - it's a way better experience - the boundaries are clear and the editors know the implications of working on a remote machine.
Not saying you can't use Windows, but you can also eat from a trough - I just prefer not to.
Random things on top of my head :
MacOS doesn't come with candy crush, Instagram, TikTok, Spotify etc. prepopulating your start menu. Phone home telemetry and ads in OS ? Yummy
Dealing with Windows dev environment is always a PITA eventually - unless you're doing stuff where Windows is first class citizen (like games). For backend stuff it's almost implicit that you're running on Linux in prod and macos is well supported because it's fairly similar. On Windows it's always some path issues, stuff randomly breaking between updates, missing/incompatible CLI, etc.
Brew is pretty good. Chocolatey is garbage.
MacOS is fairly visually consistent. Windows regularly has me in Windows XP era screens, reached through 3 inconsistent UX steps developed along the way. Even Linux is better in this regard.
I like Linux when it works. Mac works more often. Windows is just a dumpster fire at this point.
You'll probably have similar overlap between C# and F# implementations as you would with say TypeScript - they are just that different in practice IMO.
TypeScript - powerful type system but shit underlying stdlib and language (no pattern matching/switch expressions)
Dart - worse than TS because the object model is closed - so no dynamic freedom, but the type system and expressions are weaker then the rest. Also 0 meta programming facilities - Java level of boilerplate and code generators
C# - closest to ML featureset out of the mentioned, but unlike TS doesn't have sum types which will make a lot of things more tedious.
Most apps are budget constrained - doing 2x native over flutter budget doesn't get you 15% better UX. It probably doesn't even give you two working apps in the budget. And getting to them you're still likely 2x bugs and worse UX because of the time spent redeveloping common shit.
If you're not budget constrained, have a team that can execute better UX without flutter, and this matters to your app - go for native. But there's so many scenarios where it doesn't.
But I feel like COVID situation triggered stuff that either accelerated existing trends or introduced new problems, and I don't think there's going back.
People quoting productivity drops from new hires during COVID omit insane hiring in tech - implying that remote work is to blame and not their structure inability to scale to utilize those unprecedented numbers of new hires productively. Especially since most of the hiring was aimless "we need to grow headcount to capture the market growth". People getting paid serious money to literally sit around. That's going to cascade in other ways - I feel like entire industry standard of acceptable delivery sinked
Getting people back to the office sounds like a nice narrative because it's relatively easy to fix and corelate.
It's also, conveniently, entirely not the fault of upper management for insane over hiring and underutilization.
If you just need those two things there's plenty of options - why are you forcing yourself into vscode ? VSCode is great because of the plugin ecosystem - if that's not your thing use something else - there's plenty of alternatives.
If you don't even have rudimentary stuff like message broker/database you're not an AWS/Azure alternative.
I don't think this will work because cost of improvement with current method is exponential and we're already at capacity with HW.