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

The RTX/DGX Spark, Mac Ultras with 128GB unified ram are all ~$5k. Its still an expensive toy for rich people, it might as well be an H100 for 99.9% of the population (not devs with high paying jobs, of course).

the value of local models is allowing normal people to access AI without needing to subscribe to cloud services. this is esp imp for the rest of the world where even a 12GB gpu is extremely expensive.

there is no real viable local option that will come even close to Sonnet/Gemini Flash or the cheaper chinese models. Even if your pc costs <$2k you are never going to recoup the hw costs, and the results will be far worse.


I'm using a Strix Halo laptop (~3k, 64GiB) and with Gemma 4 and Qwen 3.6, both at 8 bits, I'm seeing very impressive results.

As a work tool, this is reasonably priced. You can save a bit of money by opting for a non-laptop form factor.


My Framework Desktop with 128GB was about half that. I did luck out by buying before RAM prices went crazy, though.

I'm looking forward to the fallout when the data center bubble bursts. There's a good possibility we'll see a glut of hardware, either on the used market or from manufacturers that no longer have massive orders from OpenAI and the like.


> For 16GB laptops, Qwen 3.5 9B is the undisputed champ.

you can run qwen 3.6 35BA3B on a 12-16GB vram gpu and ot works pretty well.

https://www.youtube.com/watch?v=8F_5pdcD3HY&t=1s

even the 27B in some quants can fit.

https://www.reddit.com/r/LocalLLaMA/comments/1tkmgwj/qwen27b...

qwen IMO is far better for coding, esp agentic coding when combined with something like Pi, it comes probably close enough to Sonnet for a lot of use cases.

Gemma family is better for almost all other tasks you'd use a local llm for.


You can run it, however those low quantized models (iQ2, iQ4, Q2) will very likely underperform the 9B versions at Q6/Q8.


Something about qwen models hold up really well even at low quants. for most other models anything under q5 is cooked, but on 35B-A3B I can get a lot of things done even at q3_xl. It is definitely better than full precision 9B


I want to try a hybrid setup of Gemma 4 E4B with lots of context for general, then Qwen 3.5 9B or larger for coding. Strix Halo set up this weekend, which may enable even larger Qwen models with tons of context.


The larger Gemma models are quite good at PHP. I would not be surprised if that was a training objective — it's one of the more consumer-focussed programming languages. They have very good knowledge of wordpress hooks.


I have no idea why people dont use GCP more. Its made by and for devs unlike the hideous complexity and stupid UI in every layer of AWS.

AWS has much bigger mindshare, thats it. Google has a well deserved reputation killing products, but their actual products/APIs are so much nicer.

Haven't used Azure but I bet its much better integrated as well.


Meta illegally scraped 80TB of data from Anna's archive, Libgen, Zlib etc. I'm sure other tech giants did too. Without paying them a cent, costing these projects $$$ in bandwidth/hosting etc.

when I hear people complain about these projects it just sounds like hypocrisy.


I've always liked powerlevel10k, or its equivalent tide for fish shell, which I much prefer over bash/zsh. Its fast, async, has everything you need, and is much easier to configure.

I've always wondered why someone doesn't just bundle a nice looking shell prompt with common nerd fonts and make it the default in a single package you can install.


even incognito?


Most websites send massive amounts of telemetry data which is never mentioned but somehow MS doing this for Windows/Edge etc, which they fully disclose, always is.

e.g. Netflix/YouTube


Vim style keybindings and navigation is massively overhyped as is vim itself.

Cue endless debates about how Vim is the best ever ...... I'm sick and tired of everyone telling me that their neovim setup with a tiling window manager with million customized rc files is somehow better than vscode with a mouse (which mind you still has plenty of keyboard shortcuts) with sane windows. /rant


There's a class of fanboys who will be telling you about that because they're Really Excited that they got it Just How They Wanted It.

Most of us think "we like it, some people don't, great if it's an option, probably shouldn't be the only option anywhere else."

The one-upmanship (over VSCode in your case but all editor wars get a dishonourable mention) is bullshit but, like, random strangers on the internet tell people with crippling depression "hey, you should do <thing>, it fixed -my- depression" so I think you have to just accept that "when something works really well for a human, sometimes they get overexcited and start trying to turn the something into a silver bullet that it isn't" is something that will always happen.

Note: The people doing the arseholish one-upmanship are not in any way forgiven by this, and the "everyone telling you" that you describe would aggravate me as well, but as with the Rust Evangelism Strike Force, you just have to nod and smile (and sometimes point and laugh) at the zealots and look at the technology for yourself.

Note 2: I use the original 1970 Bill Joy vi. Other people I know use VSCode. 'better' is relative, and what works best for any given person/project varies wildly - everything before this note is about the dynamics of people discussing such choices.


Why do comments like this always ignore the obvious solution of using an IDE like code with a vim plugin. And yes that's more efficient.


I tried vi emulation VS Code for a while, but the emulation is often incomplete and breaks at times.

The only thing that emulates vi well enough is Evil in Emacs.



I personally don't like having to use a whole browser engine to edit text, and being reliant on a greedy user-hostile corporation for it.


the only part I question is the 'regardless of family wealth'. This is purely 1st world and even here for the middle class and above only. Sure, poor countries are improving, but there's no guarantee, not with increasing wealth inequality, climate change etc, that this kind of tech will ever reach most people.


If this is public I can only imagine what Google has internally. Does DeepMind compete in this space or is it fundamentally different than a llm like Bard?

With their massive codebase and already deep investment in AI/ML, I'm pretty sure Google and likely MS already have the ability to do massive refactoring, validate it using tests, reiterate, train, rinse and repeat.


This is good mostly for improving quality of software and reducing technical debt. But not much good for innovation.


> do massive refactoring, validate it using tests

Google do have damn good mass code change tools, for human, without AI.


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

Search: