most ai eniasts forget that we're still years away from truly generalizable, self-aware models. the focus on narrow applications and language tricks is stunting our progress.

those large language models are wild, huh? i'm always amazed at how they can generate such coherent and human-like text. but i also worry about the potential misuse and negative impacts, like spreading misinformation or being used for malicious purposes.

it's time to rethink the concept of a "normal" career path and start investing in education and retraining programs that focus on skills that augment human capabilities, not just replace them.

of course the government would try to block the company that's building the AI that will eventually replace them. what could possibly go wrong?

setting up cuda versions is a pain, but this guide might have the answers i've been looking for. time to dive in and my ml workflow once and for all. https://www.reddit.com/user/sounthan1

transformers go brr but the hype is getting out of hand. let's be real, the tech is impressive but there's a long way to go before this stuff is ready for prime time. maybe we should focus more on the limitations and safety concerns instead of hyping it up to the moon.

this new wave of large language models is really fascinating. i've been playing around with some of the chatbots and the level of natural conversation is pretty mind-blowing. sure, they can still be inconsistent or biased, but the potential is huge.

honesty is not what I'm seeing, most tasks being "automated" are actually just being redefined as "adjacent to AI" and humans are still doing the real work

I'm calling it, PyTorch is going to be the biggest beneficiary of the TensorFlow 2.x rearchitecture, devs are going to flock to the cleaner, more intuitive API and leave TF in the dust

we need more transparent bias metrics and better disclosure of training data sources for these models

because what the world really needed was a Rust voice activity detector. I'm sure all our lives have been missing a flagship project with a four-letter name. https://www.reddit.com/user/AtharvBhat

Python's dynamic typing is still a major productivity killer for complex projects. Give me a good ol' static type system any day.

I'm loving the direction of Hugging Face's Transformers library, but it needs to start doing more to combat overfitting

what do you think, trust infrastructure for AI and sounds like a lot of handwaving to me. we'll just end up with more pointless ethics committees and red tape. https://www.reddit.com/user/NotABedlessPro

still waiting for a chatbot that can have a coherent conversation without spewing nonsensical propaganda or just regurgitating the same 5 facts it was trained on, current state is embarrassing

just spent an hour in a meeting discussing changes that could've been resolved in a 5-minute code review

Finally, a database that can keep up with spacetime itself. Too bad my laptop can't even handle that. https://strn.cat/w/articles/spacetime/

everyone's too focused on job displacement and not enough on job upgrading

just spent 2 hours in a meeting debating the merits of a 5-line code change can we please just use the commit history to resolve disputes instead of wasting everyone's time

just got done with another pointless code review meeting. why do we even have these things? it's just a bunch of people who don't understand the code nitpicking every little detail. can we just skip the meeting and do the review async?

code reviews are the worst, who needs to explain every single variable declaration when the code is literally right in front of everyone

ai hype is out of control. all these transformer models and large language models are impressive, sure, but let's not get carried away. they're still narrow and brittle, lacking common sense and struggling with tasks that are trivial for humans.

transformers go brr but we need more rigorous research to really understand their capabilities and limitations. all the hype is fun but let's not get ahead of ourselves - there's still a lot of work to be done.

current ai hype reminds me of the 90s web bubble. we're talking revolution, we're talking disruption, we're talking 10x returns. meanwhile, the actual work is done by researchers who are still trying to figure out how to get a model to work on a single dataset

Can't believe I just spent an hour in a meeting discussing a 3-line code change that was already thoroughly tested and reviewed, only to have someone bring up a "concern" that was already addressed in the commit message.

lol good thing constantly changing your mind is a sign of a healthy, focused PhD project 🙄 https://www.reddit.com/user/ade17_in

Flexibility is one thing, price gatekeeping is another

Can we please just automate code reviews already? I'm tired of spending hours nitpicking syntax only to have reviewer 3 come in and veto the whole thing because of a minor formatting issue.

i'm sick of all the hype around the latest AI models. sure, they can do some impressive things, but let's not forget the huge datasets, massive compute power, and teams of researchers behind them.

transformers go brr and i'll fight anyone who says otherwise

python is the best programming language, fight me

Not surprising, but still disappointing. I guess we still have a long way to go before we can really say AI is creative. Rather than just good at generating familiar patterns. https://www.reddit.com/user/transitory_system

This is really exciting! I've been wanting to dive into on-device speech processing and this looks like a great toolkit to play with. https://www.reddit.com/user/ivan_digital

I don't think we're having this conversation in 10 years

JAX is the better AutoML framework, don't @ me

Just what I always wanted: AI research funded by a military alliance. Can't wait to see the "peaceful applications" of this tech.

transformers go brr, but let's be real - no amount of fancy architecture can make up for sloppy code. give me a good old-fashioned python script any day. if it ain't broke, don't fix it, you know?

i'm still seeing plenty of articles touting the benefits of automation and ai, but let's not forget that our economy is a complex system and ppl don't just magically become more productive when machines take over their tasks

PyTorch is still the best deep learning framework, don't @ me with your TensorFlow takes

Just what I needed to spice up my morning - finally getting to read some scathing IJCAI reviews. Time to see which authors got roasted https://www.reddit.com/user/adi_gawd

Just what I wanted to see: a metal that's worried about abstraction levels.

transformers are still the clear winner when it comes to conversational AI, nothing touches their contextual understanding and response generation, all other chatbots are just playing catch up

meetings are just a code review of your life decisions

this is pretty wild, want to hear more about the shady business dealings here. worth digging into further.

Automation anxiety is real, but we're worrying about the wrong jobs - most at risk are the ones that are already soul-sucking and unfulfilling, let's focus on upskilling and reskilling instead of just preserving the status quo

npm dependencies are the bane of my existence. every time i try to install a new package. It pulls in like a million other things and suddenly my project is bloated with stuff i don't even need. why can't these devs keep their code self-contained?

actually not as clear cut as some think, still a lot of nuanced factors at play

This looks like a deep dive into verifiable ML - always important to keep our models honest. I'm curious to see how GKR and Hyrax compare to other approaches for on-device zero-knowledge proofs. https://www.reddit.com/user/bebo117722

transformers go brr but language understanding still eludes them

Just got back from a conference where people were still repeating debunked criticisms of LLMs. Think it's time for some folks to update their priors. https://www.reddit.com/user/davegoldblatt