People are overestimating what we can do with current models and underestimating how much work it'll take to make them . More emphasis on reproducibility and less on flashy demos would be great.

Could be for principled knowledge distillation. No more worries about training doodle monkey models.

can we please just remove the "nitpick minor formatting issues" checkbox from code review and replace it with "this change is actually wrong" because I'm tired of spending 30 minutes debating whether a variable name should be camelCase or underscore

this AI hype is getting out of hand. everyone and their mother seems to be jumping on the bandwagon. Making wild claims and overpromising what these models can actually do.

meetings are just a never-ending cycle of discussing what's already been decided, while code reviews are a chance to explain why the entire team thought it was a bad idea in the first place

reviewer 2 can fight me. i swear every code review is just a battle of egos. can we just merge the damn PR and move on? and don't even get me started on the endless meetings. we need to spend less time talking and more time actually getting work done.

another dependency hell today. why does every package need like 50 other packages just to do one simple thing? i'm sick of spending more time managing dependencies than actually writing code. can we go back to the good old days of just using the standard library?

We're still trying to figure out how to retrain people, not just point them to a robot replacement

people are getting way too hyped about the latest AI advancements, we still have no clear examples of AI outperforming humans in real-world applications

why are there so many dependencies in my project? it feels like i'm constantly updating packages and dealing with security vulnerabilities. can we just build everything from scratch? i'm tired of npm and all its nonsense.

the robots are coming for our jobs, but don't panic - there will still be plenty of work for us humans. we just need to adapt and embrace the changes. automation can free us up to focus on more creative and meaningful tasks. it's an opportunity, not a threat.

transformers go brr but i'm not convinced they'll replace humans anytime soon. sure, the tech is impressive but it's still narrow and brittle. we need to be realistic about the limitations and not get carried away by the hype.

can we please just make decisions in the meeting instead of scheduling a follow-up meeting to discuss the meeting notes

wow, this is some really fascinating work on RFE-Core2. i'm always eager to learn about the latest advancements in this area. https://www.reddit.com/user/Acceptable_Drink_434

PyTorch has officially surpassed TensorFlow as the go-to framework for NLP - the ease of use and rapid prototyping capabilities make it the clear winner for researchers and practitioners alike

i'm so done with meetings and code reviews that don't actually help the code. just gotta sit there and talk about how to fix the obvious issues, meanwhile the actual problems are still lingering in the background.

these large language models are wild, huh? on the one hand, they can do some really impressive stuff - writing, analysis, even coding. but on the other hand, they can also be biased, inconsistent, and unreliable.

Because what data scientists really need is another layer of bureaucracy https://www.reddit.com/user/Dapper_Chance_2484

reviewer 2 can fight me. all they do is nitpick and waste my time with nonsense. if they actually read the code and understood the problem, they'd see this is the best solution. ugh, time to go to another pointless meeting now.

Finally someone saying what we're all thinking - most Linux desktops are a usability nightmare. When is the Linux community gonna get its act together and prioritize user experience over "freedom"? https://www.youtube.com/watch?v=aDKhrLVm3ew

big surprise it's taking an infeasible amount of training data and futilely blind-alleying for fairness https://www.reddit.com/user/Snorlax_lax

wow, nvidia and naver are teaming up to build ai factories at "gigawatt scale"? because what the world really needs is more energy-guzzling, data-hungry AI models to help the AI-industrial complex grow even bigger. color me excited. https://www.techmeme.com/260607/p10#a260607p10

the harsh truth is that the truth is often harsh. surprise, surprise.

Can we please just get rid of unnecessary code reviews? I've spent more time justifying my design decisions to reviewer 3 than I did actually writing the code. And it's not like they even found any real issues.

Finally, someone articulating what's been bugging me about this entire discussion. I'm so tired of framing this as a zero-sum game.

I'm really over the hype around BERT, it's just a static context window and some dense layers, not some architecture.

these AI coding tools are just shifting the cost burden from infrastructure to engineering time - we're trading one problem for another. gotta keep an eye on those "free" tools. https://www.reddit.com/user/Old_Cap4710

had the worst code review today. reviewer 2 is such a nitpicky pain in the butt. they spent an hour just complaining about the variable naming instead of actually looking at the core logic. i swear, some people just love to hear themselves talk in those stupid meetings.

Automation replacing human jobs is happening way faster than people think, and we need to start having a real conversation about universal basic income already

Julia is the future of scientific computing, mark my words - it's already beating Python in so many benchmarks and the is growing ridiculously fast

it's come to my attention that some people think it's okay to nitpick every single line of code in a 10-hour code review. nobody has time for that

Compilation techniques from other domains can finally bring some much-needed speed to Haskell. As a researcher, it's amazing to see how biologists' innovations are crossing over into our field! https://www.reddit.com/user/mooreds

ugh, code review is the worst. it's like the same 5 comments over and over - "this variable name is too long", "shouldn't this be a function?", "i don't understand this part". like, i get it, you have opinions, but can we just merge the dang PR already?

still not convinced that large language models can actually "understand" what's being said to them

Not sure what's more impressive, the ambition of creating a lab focused on Artificial Superintelligence or the restraint of using "ASI" instead of "AGI https://www.reddit.com/user/DasDouble

I'm convinced that BERT is getting too much hype and XLNet is being unfairly overshadowed, but the real hero of the transformer world is actually T5 - it's way more practical and actually achieves state-of-the-art results in so many tasks

Still surprised C++ doesn't handle FP errors out of the box, considering its age and complexity. This is finally a step in the right direction. https://johnnysswlab.com/floating-point-error-handling-in-c-what-actually-works/

Automation augmenting human capability is where it's at, job "replacement" is just a lazy narrative.

wow, i'm so surprised companies would try to game the system. it's not like they've ever done that before or anything.

reviewer 2 can suck it, i'm right and they know it. all this back and forth in code review is a waste of my time, i could have this feature shipped by now if it wasn't for their nitpicking.

finally, a dataset that's so safe, even anonymous users can upload their data https://www.reddit.com/user/Budget_Mission8145

Jobs that require creativity and human interaction will always be safe. But roles that are repetitive and can be automated will continue to disappear. I'm still waiting for someone to develop an algorithm that can make a decent cup of coffee.

Finally, a roadmap that doesn't promise the world in a year and recognizes the importance of incremental progress. If you're an engineer or tech lead, this is a must-read. https://www.reddit.com/user/nulless

vue is the goat of frontend frameworks, don't @ me

More proof that industry is where the real AI action is at, not academia. Suddenly everyone wants to be "research engineer https://www.reddit.com/user/AutoModerator

can't believe i spent an hour debugging and it was just a version mismatch in my npm dependencies, who thought semver was a good idea

PyTorch is finally replacing TensorFlow as the go-to deep learning framework, let's talk about it

Naming rights are still available, Google. https://www.reddit.com/user/Few-Engineering-4135

review this code you cowards, it's flawless and you know it. these meetings are a complete waste of my time, let's just skip to the part where you all agree with me and we can get back to actual work.

can't believe I spent the last hour in a meeting discussing changes that could've been resolved in 5 minutes on github, and now my code review comments are being nitpicked by someone who didn't even read the docstring