fennecfoxy

Eh I think the small model thing is kind of a no-go.

Reason being is that many workloads for AI are dynamically mixed, where training from multiple subjects comes into play and you just can't know exactly what mix will be required for each task ahead of time.

I was hoping loras would do this for us as well but they don't really seem to have worked out for llms (compared to in the image/video diffusion space).

Perhaps some future model will have some sort of "core" that can load/unload portions of itself dynamically at runtime. Like go for a very horizontal architecture/hundreds of MoE and unload/load those paths/weights once a parent value meets or exceeds some minimum, hmmm.

Tbf I've always hated that about HN formatting as it's not very clear at all that that's how it works.

If there's a newline in my comment, why not retain it? Whyyyyy?!

Nope.

Because late stage capitalism demands endless growth in order to pay executives and shareholders (especially those late to the train) more and more YoY.

And those requirements for growth mean that cost cutting is needed. Over the past few decades cost _have_ been cut, building things more efficiently, components becoming cheaper, larger volumes in mass manufacturing.

But we have already reached a point where there are no other places to cut than the quality of the product itself. Look to shrinkflation in food and other places - look at how "live action" versions are being made of previously animated movies, how game franchises from 2 decades ago are being brought back from the dead, the huge influx of remasters etc.

Why? Because it's cheaper to revive/reuse an existing IP than it is to create a new one + it guarantees success with the drooling consumer masses. And cheaper = more Ferraris for the multi millionaire/billionaire execs.

See how much Mario movie made? Just wait...bet you there'll be a live action version. ;)

Personal attacks are against the rules, by the way.

>It's here, right now.

I mean I've been forcing my good old 1080ti to run local models since a short while after llama was first leaked.

But I wouldn't say "local models are here" in the same way as "year of the Linux desktop!111"

Until someone can just go out and buy some sort of "AI pod" that they can take home, plug in and hit one button on a mobile app to select a model (or even just hide models behind various personas) then I wouldn't say it's quite there yet.

It's important that the average consumer can do it, I think the limitations for that are: things are changing too quickly, ram+compute components are exceedingly expensive now, we're still waiting on better controls/harnesses for this stuff to stop consumers not just from shooting themselves in the foot, but blowing their foot clean off.

Would be interesting to see a Taalas-like chip in a product, albeit there's so many changes going on atm with diffusion based models, Google's Turboquant (which as someone who has had to almost always run quantized models, makes a lot of sense to me).

Lmao no, my doorbell is one (doesn't have a battery at) and has been going for 7 years strong, as long as I've lived in this place.

Sure you have to press it very slightly harder than a regular switch and the travel is a little more, but not by much.

I think the doorbell cost like £20 in 2020, actually lemme check Amazon...ah in fact it was 15.99£ (now it's £20.99 hmmm) "TECKNET self powered doorbell".

Yeah agree. Having to rewind the motors just adds yet another task to the list for someone trying to build this thing.

Also BLDC based motors in bots seem to be popular now, used to be steppers and harmonic drives from what I remember reading.

Idk why electrostatic actuators aren't getting more popular like HASEL and a few papers on ones using fiber pumps come out recently. I suppose the danger that the high voltages present hmmm.

But surely we can replicate muscle fiber actions magnetically, just need some sort of crazy micro manufacturing process to make the magnets.

There's the mech/EE equivalent of crud, too.

Plus the demand is higher for software than it is for hardware most of the time. Pretty hard to find jobs in robotics compared to all the various kinds of software.

I would rather just have temperature sensors over the place. More reliable.

Tbf I've thought a decent bit about how most current AI is essentially just being used to digest what exists on a website/etc. Honestly even just the vector search/RAG part is useful, but more-so with a model to help do some initial filtering of it.

It's an odd use case - we have used language for a millionish years or so and it makes sense that that's the easiest way for us to get at information/do things.

But at the same time it's faster for me to read than listen, but it's often slower to type than to speak. It's faster to hit one button in a familiar place to do some predetermined thing, but much slower when the location of that button changes/gets hidden under submenus/I'm not familiar with an app or website.

On Android I constantly use the search function of the settings menu and I feel like this will be the golden UX going forward - a side by side UX + NL interface. So I can ask "how do I add a photo" and from there I get taken to the right place and can continue to add multiple photos in one go following the same pattern.

Though I suppose the nicer alternative is just "add all the photos I took near the waterfall from today".

space.com huh...

2s of visiting the site, before I've even had a chance to read the article: - consent for trying to enable notifications - modal for trying to get me to subscribe to emails Then skim, try to leave - Hijacks back button for a "before you leave" thing

Absolute trash.

As a Kiwi (now in UK) NZ doesn't get that cold for that long...mostly just wet, unless you're pretty far south.

In UK/other parts of Europe winter gets colder, lasts much longer, humid the entire time (so heat just escapes all over the place). Plus, the buildings here are a lot older - I think upgrading insulation would make a huge difference this side of the world.

I couldn't even imagine Canada. Almost moved there...decided to stay here. No -20c winters for me ty very much.

Borehole or the pipe grid they stick under your backyard/garden (if you have a decent sized one) end up way more expensive.

But tbf, AI and robotics are rolling along pretty well. I'm surprised there's not a company that's just build the "this robot installs your borehole/underground pipes in 3 hours by itself" robot.

Nah I'm gonna generate the hecky out of "relevant for this one presentation" little cliparts to add to my powerpoints.

But then I'm still going to take photos on film and enjoy Sunday afternoons in the darkroom doing prints.

It's possible to compromise and/or use the right tool for the right job. Saving me time for something of little or fleeting importance? AI. Making me feel good/physical work with hands/emotion chemicals in my brain? Film/traditional media.

Lmao, and pre-AI the world wasn't trending towards cost-reduced mass-produced executive-driven content?

And the average member of our species lapping it up like being a mindless consumer is going out of style?

Yes, look at how many historical inventors (like the Blue LED, the guys struggling to convince Gates and Balmer to make the Xbox) etc get/got nothing for their efforts compared to the huge sums raked in by the very people actively trying to prevent them from building the idea that made all the money.

AI is hugely beneficial to our species. Our tribalism and "yeah well they earned it!" response to capitalism's rampant production of billionaires is the real problem, not technology.

Why are footballers and movie celebrities paid 50$m a year? There's the answer.

No, it won't be. I did indeed get the same problems when trying to generate my own image for it.

However as someone who's mucked about with local image generation as well - I'd say that this is a problem with their implementation, it doesn't resolve fine detail because majority of requests it won't matter/it drastically increases compute requirements.

With local image generation bad features/incorrect fingers/disfigurement etc has been solved for a long time.

I think their new process involves multiple steps including sketching/fleshing out the idea before adding detail. The step that would fix this would be outpainting or similar to tile based upscaling.

From what I understand of image generation models they also struggle with fine detail in general because they aren't really trained for that. However for each tiny chunk of a detailed image like that there's nothing to say they can't allocate a 500x500 chunk for it to work in as its "idea/reference space" and then transpose that into the main image being generated - i.e. generate image features separately rather than all together.

Because if it's multimodal, oops all transformers and they're pretty much best in class for ocr now, afaik?

Dogs don't wear suits! You must be AI too ;)

I mean asking these transformers to do maths has always been the wrong task. It's like we're now considering "it doesn't have x tools built with traditional code built in".

Though I suppose we're testing their model + agent harness here as well. It really _should_ have all of those tools/reasoning available to accomplish a task like the above without issue.

>They are just focused on privacy

Is that why they rushed out introducing AI summaries etc in order to play catch-up and then backpedaled when they exploded in customers' faces/individuals concerned in false headlines threatened to sue?

I use Gemini on purpose all the time. It can start timers for me, add calendar entries without having to type it out, convert email to calendar or reminders etc. I'd use it even more if it had more access to other bits of my phone.

Omg imagine the response if it was the other...

You know what, I really cannot be bothered anymore.

Because they've invested billions into making their own chips? And those chips have pretty damn good performance/power draw/cooling etc. I think the answer is pretty straightforward.

Manual focus I keep for film, I feel like it's a part of the process.

But I do wish my Sony 50 was a little less noisy/slow. Suppose I should pick up the GM version at some point.

I feel like this is a political move between Hyundai and Google (favour by Google).

BD sat back on traditional programming/light ML techniques for ages whilst transformers went wild and it's only now that they're like "oh shit".

Hence the partnership with Google; BD lacks the capabilities otherwise. I bet their internal marketing departments did a bit of hand shaking to spin this piece as a favour for Hyundai/BD. Because from Google's (and our) perspectives - reading a gauge etc isn't that impressive and multimodal transformers solved that years ago, OpenCV many years before that also. But to BD it's impressive/a desperate grasp of "we swear we're using modern ML now! Yes our robot dances were sequenced and took dozens of takes but now we'll start doing it for real, we swear!"

Waveshare has some pretty rad ones. I've been tempted to use one of their rovers for something.

Be careful, because you can easily overpay out the ass for "robot kits" online.

Honestly? Because capitalism.

I think the thing is: does it need to last 20-30 years between replacements if a robot can easily replace it + they're cheap enough to add redundant ones. Do we really need crazy accuracy even on an industrial level...like this pipe will burst at 200psi so the gauge needs to be accurate to 0.001 psi so we can sound the alarm when it hits 199.999 psi somehow I don't think so.

Dumb silicon is so super cheap now, just look to nfc etc, 1c microcontrollers. We can litter our world with sensors.

Which I would love to see - but I'm also not discounting the usefulness of any robot just being able to read something we can read and vice versa.

I would look into VLAs/VLMs if you haven't already.

"The chip" no, a whole rack/deployment they offer takes 2.5kW. Not just one chip. Squeezing 2.5kW thru 1 chip would be mental.