pretty much every quality model these days is too big for a startup or medium sized company to train. Instead you just fine-tune existing big models or worse, do prompt """"engineering"""", which is such a monkey job it's not even funny anymore.
there is no more variety in the field, every state-of-the-art is now transformers. A new model just dropped with better metrics? Guess what, it's another transformer, but it's bigger this time!
I miss the times when finding and training an appropriate model was an exciting task. You had to read papers, prepare the dataset, try out multiple models, compare them based on metrics you picked based on what was required.
now you get an API key and spend 2 hours prompt """engineering""" a template for your need. Anything more involved is not considered to be a worthy time investment by the boomer sales.
it's over, data science bros, we are now regular software developers doing some monkey business on the side. I don't even enjoy software engieering that much, this isnfucking depressing
Attention is all you need, bro
KILL AI SPAMMERS
DECAPITATE AI moronS
DEFECATE IN AI IDIOTS
PULVERIZE AI MORONS
INCINERATE AI gayS
DECIMATE AI IMBECILES
EXPLODE AI RETARDS' HEADS
SHOOT AI garden gnomes
CRIPPLE AI DICKHEADS
You mad cuz you bad
I love bunnies
Haha imagine squeezing a fluffy one, but not too hard. Haha that would be fun haha.
same
yes they are the best household pet
this, most kino pet ive ever had thats actually chill and not a psycho like cats
I had bunnies growing up. They’re really sweet, and so soft to hug.
Yeah, their pretty great.
>tfw I've got an abandoned project about rabbits in space.
>pretty much every quality model these days is too big for a startup or medium sized company to train. Instead you just fine-tune existing big models or worse, do prompt """"engineering"""", which is such a monkey job it's not even funny anymore.
>there is no more variety in the field, every state-of-the-art is now transformers. A new model just dropped with better metrics? Guess what, it's another transformer, but it's bigger this time!
>I miss the times when finding and training an appropriate model was an exciting task. You had to read papers, prepare the dataset, try out multiple models, compare them based on metrics you picked based on what was required.
>now you get an API key and spend 2 hours prompt """engineering""" a template for your need. Anything more involved is not considered to be a worthy time investment by the boomer sales.
>it's over, data science bros, we are now regular software developers doing some monkey business on the side. I don't even enjoy software engieering that much, this isnfucking depressing
Trans rights
>prompt engineering
What career path should I tell my children to go down? Doctor and Engineer?
Youtuber
Engineer it is then.
prompt engineer
prompt doctor
The big companies are going to lobby politicians to create "regulation" for AIs that grants them a de facto monopoly over them anyway
Is this true? I was worried I was missing out since I assumed a career in AI would be too hard for my little pea brain
Machine learning / data science / etc was a thing of the 2010s when advanced 1990s algos could finally be run in reasonable timescales with reasonable data volumes on reasonably priced hardware. Now "AI" has turned into a bunch of api calls as a bunch of nerds automated everything.
If you want to understand what the api is doing you need to go through some pretty advanced mathematics. But you don’t have to and are cruelly disadvantaged for doing so. How many cameraphone users understand what a DCT is, or have even heard of it?
But I mean if you end up studying machine learning and aren't in the top 1% will there be any use for you?
Literally every company even vaguely in tech wants machine learning. No you won’t work at google, but you could easily get a job at a place like Thermia or 3DConnexion.
Anyone that collects data wants to be able to aggregate it into a database that can predict their customers needs and wants, and how to widen that customer base. And everyone is collecting data. Your heat pump doesn’t have a WiFi connection and a phone app because it’s more convenient for you to adjust your temperature that way.
You’ll be fine. You’ll be something of a code monkey but being a data science mandrill is way better than being a webapp macaque.
Mass surveillance still needs lightweight algorithms, can't just throw the heavy shit at everything.
What? Training the current GPT takes about 300k USD in hardware and power which isn't much for most companies.
Incorrect. The market is still very fertile for micro-sized ML for botnetted iot devices, and small/mid-sized ML for user, buisness, or industry programs to run locally.
If you want text or image generation then yea you're out of luck.
Glad I've completely stayed out of AI stuff and kept doing my backend webdev moron work