Nice.

Yea I don't trust any AI models for facts, period. They all just lie. Confidently. The smol model there at least tried and got it right at first... Before confusing the sentence context.

Qwen is a good model too. But if you wanted something to run home automation or do text summaroes, smol is solid enough. I'm using CPU so it's good enough.

Here you go: Review of SmolVLM https://www.marktechpost.com/2024/11/26/hugging-face-releases-smolvlm-a-2b-parameter-vision-language-model-for-on-device-inference/

Model itself: https://huggingface.co/spaces/HuggingFaceTB/SmolVLM

And you can use Ollama to run it locally, and Open WebUI to access it in browser.

Try again. Simplified models take the large ones and pare them down in terms of memory requirements, and can be run off the CPU even. The "smol" model I mentioned is real, and hyperfast.

Llama 3.2 is pretty solid as well.

Building on this, and not being too hyperbolic about "realism": he's wearing a full body set of reinforced armor, that is almost certainly going to assist in compressing the wound and his injury buying a massive amount of of time to start with. Assuming for 5 seconds he slaps some quick clot into the hole one he get in The Bat, or before, then bleeding out wouldn't be a main concern, notright away. Organ damage is his biggest risk, and if he avoided a direct stab into a kidney or something (the armor has gaps but still covers vitals), he could live if he's lucky with some back alley sutures to his intenstine, etc.

So, him living isn't the most insane thing to consider given his known resources and what he could likely have done in a few moments off screen. And over-explaining it in the moment would've killed the pacing of the film.

Thing is, for your average user with no GPU and whp never thinks about RAM, running a local LLM is intimidating. But it shouldn't be. Any system with an integrated GPU, and the more RAM the better, can run simple models locally.

The not so dirty secret is that ChatGPT 3 vs 4 isn't that big a difference, and neither are leaps and bounds ahead of the publically available models for about 99% of tasks. For that 1% people will ooh and aah over it, but 99% of use cases are only seeing marginal gains on 4o.

And the simplified models that run "only" 95% as well? They can use 90% fewer resources give pretty much identical answers outside of hyperspecific use cases.

Running a a "smol" model as some are called, gets you all the bang for none of the buck, and your data stays on your system and never leaves.

I've been yelling from the rooftops to some stupid corporate types that once the model is trained, it's trained. Unless you are training models yourself, there is no need for the massive AI clusters, just for the model. Run it local on your hardware at a fraction of the cost.

I'm sticking with Floorp for now. Faster development doesn't mean it will be sustainable, but we'll see. Or necessarily that all feayures may be good. More iant better. Brave and Opera GX have a bunch of features too.

But I'm using Firefox and Floorp as a variant because I want something that gets fundamentals right. And if Zen becomes huge, great.

Yea, fuck this. I like Nvidia and they objectively have better software on the cutting edge. But they stopped focusing on thos market now and are all in on AI while screwing over the high end on gaming. No DLSS is worth this bullshit.

I'd rather pay less for an AMD card with 90% as good performance, and just slightly jankier software, than pay the Nvidia tax this time. I was tempted to get in on the low end of the 5000 series, but I'll grab a mid-tier AMD and be happy.

"Listen, making the entire market dependent on one corporate benefactor just sothey aren't a 100% monopoly and only a 99% one is important"

Jesus Christ Mozilla, do you hear yourself?

Flipside of this coin: this results in repair monopolies because users cannot repair their own vehicles and equipment and manufactures use this exact excuse to claim they HAVE to run a monopoly cause the EPA. Literally John Deere has said this.

In truth, illegal vehicle mods have been and will always be a thing. Manufacturers should still provide all the tools to users to repair, and emissions checks on trucks will have to be smarter to catch cheaters. Make the penalty for a deliberate violation (willful not accidental) so egregious that no one would consider it, even if it saved 50k+ per truck.

Easily. Read my other comment, but this would pay for itself in a single afternoon if you didn't maintain your vehicle and would otherwise get shutdown due to an exhaust failure.

Or if you drive a truck in stop and go traffic, and the filters clogged up early, causing you to delay and let the exhaist system do a cleaning cycle (take 1hr, requires engine running and high throttle.) Etc etc.

When you run out of DEF or the DPF is clogged, you can't run your truck for more than a brief while. You get half output in a limp-mode to go refill your DEF or have the DPF serviced. DEF is the reactant for the exhaust that makes diesel burn cleaner, but means modern trucks have 2 tanks now. Users hate it, but it cuts emissions massively. Also adds a few grand to the vehicle exhaust system in hardware and sensors and control units. Anyways:

Time = money.

For a commercial or even semi private vehicle if you bypass even one indicent of downtime by doing this is paid for itself.

That said, the DPF is a filter, and can physically clog and cause an exhaust fire if there is no monitoring software. I hope at least this guy had it wait till it was almost critical and then stop, not entirely disable the stop signal. Otherwise there is a serious risk to the vehicle and passengers.

For low res, no.

Hi res, sure. Make it optional, or let players download the region they like. Or just the airports with much lower res landscapes, etc etc.

Or just, let them have it all and make these choices. Memory is CHEAP nowadays. If you're a flight sim enthusiast, a few terabytes for the map data is the least expensive part of your setup by far.

They need to make sharing controller presets easier. Like MASSIVELY easier.

I cannot emphasize this enough. Let users send controller profiles and attach them to forum posts, etc.

This is needed desperately because navigating presets is straight up busted. Currently it just promotes whatever has the most users, which is whoever published first.

Doesn't matter if there is an objectively brilliant control scheme for Resident Evil 5 or something, nope, you're using the first layout published by GoonerMaster69 in 2014 or whatever.

Both these factors make using the steam controller a chore. It's incredibly powerful, but setup is a fucking CHORE. I love it, but making it work well is always me tweaking for way too long, or using g the suboptimal poorly named presets at the top of the list.

The RE5 example is real. RE5 is amazing with the steam controller if you have a trigger full pull mode shift to mouse input so you can have perfect accuracy with no acceleration. But try finding this in the preset list.

God I love having a future where my ability to play a fucking flight simulator depends on both internet access and server reliability.

Completely unnecessary to boot. Store a low res copy locally, offer the high res as regional packs. 0 reason to stream this data in.

Kadokawa is a massive publisher of books and media in Japan.

For that reason alone I hope this is blocked. It would make Sony too huge in Japan to reasonably compete with outside of Shueisha and directly incentivize brand exclusivity deals.

The iPod. Most I've seen on resale online have been crazy marked up.

Any advice where to find one of these cheap?

Not gonna argue this point because both candidates for President were staunchly pro Israel.

So they vote for the non-genocide down-ballot, and different, even more vocal, genocide support on the top?

That's what doesn't compute. It 100% would make sense if they said to hell with it and switched party affiliation across the board, that would be consistent even if you argue their logic. But this is internally inconsistent.

I'm only wondering about the statistically improbable break between the state where they voted all Dems downballot but then swapped to Trump just for President.

This is so statistically unlikely, having one or two states this way would be improbable as hell, but (I am not claiming this is true) there are apparently many instances where this happened.

I'm not sharing any of this stuff I see but so far this is the one that got my eyebrow raised.

"AI" cannot make rational choices.

It is a giant word association machine.

For the love of god this should never be involved in military applications.