Hacker Newsnew | past | comments | ask | show | jobs | submit | mrbungie's commentslogin

Well, depends on the software project itself and where you are in its development lifecycle but:

- (1) A lot of developing can be just chores around managing scaffolds and repeatable work, and due to this macros, autogenerated code and other tools have been a thing at many layers for a long time; and

- (2) I remember copy-pasting from Google/StackOverflow (i.e. mostly search + pattern matching with some minimal reasoning) being criticized as a low-effort mode of development during the 2010s, before ChatGPT and AI assisted coding tools took over that part.

So yes, I'd argue a huge amount of software development problems can be solved without ever actually reasoning from first principles, AI tools just made that more visible.


> go through much more than just predict-next-token training (RLHF, presumably now reasoning training, who knows what else).

Yep, but...

> To say they LLMs are 'predictive text models trained to match patterns in their data, statistical algorithms, not brains, not systems with “psychology” in any human sense.' is not entirely accurate.

That's a logical leap, and you'd need to bridge the gap between "more than next-token prediction" to similarity to wetware brains and "systems with psychology".


Sad as it defeats the purpose and spirit of libgen.

Being more concrete: I don't think they made libgen hoping the content would be used to create an even worse walled garden for knowledge.


It's fun when then you read last Nvidia tweet [1] suggesting that still their tech is better, based on pure vibes as anything in the (Gen)AI-era.

[1] https://x.com/nvidianewsroom/status/1993364210948936055


Not vibes. TPUs have fallen behind or had to be redesigned from scratch many times as neural architectures and workloads evolved, whereas the more general purpose GPUs kept on trucking and building on their prior investments. There's a good reason so much research is done on Nvidia clusters and not TPU clusters. TPU has often turned out to be over-specialized and Nvidia are pointing that out.

You say that like I d a bad thing. Nvidia architectures keep changing and getting more advanced as well, with specialized tensor operations, different accumulators and caches, etc. I see no issue with progress.

That’s missing the point. Things like tensor cores were added in parallel with improvements to existing computer and CUDA kernels from 10 years ago generally run without modification. Hardware architecture may change, but Nvidia has largely avoided changing how you interact with it.

Modern CUDA programs that hit roofline look absolutely nothing like those from 10 or even 5 years ago. Or even 2 if you’re on Blackwell.

They don't have to, CUDA is a high-level API in this respect. The hardware will conform to the demands of the market and the software will support whatever the compute capability defines, Nvidia is clearer than most about this.

But for research you often don't have to max out the hardware right away.

And the question is what do programs that max out Ironwood look like vs TPU programs written 5 years ago?


Sure, but you do have to do it pretty quick. Let’s pick a H100. You’ve probably heard that just writing scalar code is leaving 90+% of the flops idle. But even past that, if you’re using the tensor core but using the wrong instructions you’re basically capped at 300-400 TFLOPS of the 1000 the hardware supports. If using the new instructions but poorly you’re probably not going to hit even 500 TFLOPS. That’s just barely better than the previous generation you paid a bunch of money to replace.

And yet current versions of Whisper GPU will not run on my not-quite-10-year old Pascal GPU anymore because the hardware CUDA version is too old.

Just because it's still called CUDA doesn't mean it's portable over a not-that-long of a timeframe.


Portable doesn't normally mean that it runs on arbitrarily old hardware. CUDA was never portable, it only runs on Nvidia hardware. The question is whether old versions of Whisper GPU run on newer hardware, that'd be backwards compatibility.

> There's a good reason so much research is done on Nvidia clusters and not TPU clusters.

You are aware that Gemini was trained on TPU, and that most research at Deepmind is done on TPU?


> based on pure vibes

The tweet gives their justification; CUDA isn't ASIC. Nvidia GPUs were popular for crypto mining, protein folding, and now AI inference too. TPUs are tensor ASICs.

FWIW I'm inclined to agree with Nvidia here. Scaling up a systolic array is impressive but nothing new.


Sure, but their company's 4.3 trillion valuation isn't based on how good their GPUs are for general purpose computing, it's based on how good they are at AI.

> NVIDIA is a generation ahead of the industry

a generation is 6 months


For GPUs a generation is 1-2 years.


What in that article makes you think a generation is shorter?

* Turing: September 2018

* Ampere: May 2020

* Hopper: March 2022

* Lovelace (designed to work with Hopper): October 2022

* Blackwell: November 2024

* Next: December 2025 or later

With a single exception for Lovelace (arguably not a generation), there are multiple years between generations.


The most fun fact about all the developments post-ChatGPT is that people apparently forgot that Google was doing actual AI before AI meant (only) ML and GenAI/LLMs, and they were top players at it.

Arguably main OpenAI raison d'être was to be a counterweight to that pre-2023 Google AI dominance. But I'd also argue that OpenAI lost its way.


And they forgot to pay those people so most of them left.

To be fair, they weren't increasing Ads revenue.

They literally gave away their secret sauce to OpenAI and pretended like it wasn’t a big opportunity.

Just as expected from a big firm with slower organizational speed. They can afford to make those mistakes.

> It surprises me how hyper focused people are on AI risk when we’ve grown numb to the millions of preventable deaths that happen every year.

Companies are bombarding us with AI in every piece of media they can, obviously with a bias on the positive. This focus is an expected counterresponse to said pressure, and it is actually good that we're not just focusing on what they want us to hear (i.e. just the pros and not the cons).

> If anything, AI may help us reduce preventable deaths.

Maybe, but as long as it development is coupled to short-term metrics like DAUs it won't.


Not just focusing only on what they want us to hear is a good thing, but using more noise we knowingly consider low value may actually be worse IMO. Both in terms of the overall discourse but also in terms of how much people end up buying into the positive bias.

I.e. "yeah, I heard many counters to all of the AI positivity but it just seemed to be people screaming back with whatever they could rather than any impactful counterarguments" is a much worse situation because you've lost the wonder "is it really so positive" by not taking the time to bring up the most meaningful negatives when responding.


Fair point. I don't know how to actually respond to this one without an objective measure or at least proxy of a measure on the sentiment of the discourse and it's public perception.

Anecdotically I would say we're just in a reversal/pushback of the narrative and that's why it feels more negative/noisy right now. But I'd also add that (1) it hasn't been a prolongued situation, as it started getting more popular in late 2024 and 2025; and (2) probably won't be permanent.


Fair point. I actually wish Altman/Amodei/Hassabis would stop overhyping the technology and also focus on the broader humanitarian mission.

Development coupled to DAUs… I’m not sure I agree that’s the problem. I would argue AI adoption is more due to utility than addictiveness. Unlike social media companies, they provide direct value to many consumers and professionals across many domains. Just today it helped me write 2k lines of code, think through how my family can negotiate a lawsuit, and plan for Christmas shopping. That’s not doom scrolling, that’s getting sh*t done.


> focus on the broader humanitarian mission

There is no humanitarian mission, there is only stock prices.


You can say "shit" on the internet, as in "I bet those two thousand lines of code are shit quality",or "I hope ChatGPT will still think for you when your brain has rotted away to shit".

Nobody likes people like you, so I hope that temporary high of snarky superiority gets you through the day, buddy :)

I like people like me, so I like your comment too. This will keep me going all week!

> obviously with a bias on the positive

Wait, really? I'd say 80-90% of AI news I see is negative and can be perceived as present or looming threats. And I'm very optimistic about AI.

I think AI bashing is what currently best sells ads. And that's the bias.


I know what you mean, but LLMs are just a tool. Probably the joy is actually taken out by some form of pressure to use them even when it doesn't make sense, like commercial/leadership pressure.

It is, but biased evidence, as he's both directing and checking that frontier LLM output and not everyone is Terrence Tao.


Obviously, not the best plot to use according to Data Visualization theory and common practice, but I think it candidly conveys the point anyway.

As someone else points, the data is the worrying aspect, as it points towards state-of-the-art models not being able of making more than 0 consecutive steps without errors.


You don't, but oh boy, the experience is worth it. Bazzite[1] has it quirks but it mostly works fine in desktops.

[1] https://bazzite.gg/


Imo if you just have a regular desktop PC, use Ubuntu/Fedora, not a dedicated 'gaming' distro. Bazzite's good as a stand in for steam os on non Valve handhelds, but Steam and Proton work just fine on a regular boring Linux distro.


Bazzite is a lot less messing around though. Stock standard fedora doesn't have the drivers needed for modern xbox controllers. Doesn't have a controller usable interface, etc.

If your PC is connected to a TV than Bazzite is a much better experience.


I mostly agree, with the caveat the Bazzite is also a good option for PCs that spend their life permanently connected to a TV as a gaming box. It makes for a great big screen sofa experience too vs using typical Linux distro desktop UIs or Windows. Roll your own Steam Machine, essentially.


Bazzite is just Fedora Kinoite with some tweaks for gaming, like automatically including Nvidia drivers.

I've joined the Kinoite kult since it's much easier to deal with an atomic system.


Debian / Fedora are riddled with features gamers will never need.


So is windows. The point being that you can have your cake and eat it too with a stable distribution, proper drivers, proton and Steam.


I'm making an assumption that most HN commenters aren't using their PC only for games.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: