OliverGuy

Imo, we are going to see a lot more specialised models in the near future as pushing the frontier gets harder. Qwen 3.8 27B for example is a pretty rubbish chat and general world knowledge model, but is stellar at coding (for it's size). Its small enough anyone with a modest gaming PC (16GB vram and 32GB of DDR5) can run it at speeds good enough for overnight coding sessions with an PR to review at the end of it.

Eventually, I hope we'll see many more specialised small models that run nicely on consumer and prosumer hardware. Like LLMs specific to income tax law for UK residents, or LLMs specific to building power delivery circuits for PD USB C devices. There probably is a point where the niche-ness does become to niche to justify the post training, but again, small models means much less training required.

Kind of begs the question if this even needs new/better data? Could you take the foundation model that's trained on everything, and then run a final fine tuning training run on just the set of foundation training data that you want the model to specialise in? Eg, extract just the training data that is about UK income tax law and then run a few epochos on just that? Any ML people know if this is a thing or if not, why it doesnt work?

Nothing wrong with Ryzen for a server, you can get AM5 server boards from Asrock and they support ECC. If you need more they make AM5 Epyc CPUs that are just Ryzen parts with Epyc written on the box and IHS

Yea, I'd be very happy with these systems if I could vend and rotate tempary read only credentials to these agents.

I can't get a read only API key for my YouTube, Gmail, Outlook etc etc accounts. Only full admin access via a browser.

I can get r/o creds for things like GitLab and AWS and K8s, which means I'm very happy to let my agents run wild at work because I know for certain they actually can't do any damage, they can read all they want but can only go as far as opening and MR that I can review and merge.

Why?

Even for domestic robots people don't care, see just about every cloud connected smart vacuum robot and the numerous data leaks

And for companies, most of them are fine using SaaS and cloud based hosting platforms, so why would it be any different for robotics?

Sure there is the latency argument, so at least something doing very short term decision making needs to run on device or at the edge, but for most task the end to end time is in the single didit seconds to minuites, so the extra 100ms or whatever that you get going out to some model in the cloud isn't a deal breaker.

I like this, but can we not have any of this in the root of the repo cluttering it up with the harness specific instructions for a half dozen plus different harnesses. Something like a dot agents dir

If you look at bike product photography it's always drive side facing the camera, which means front wheel on the right. If I had to guess this is probably where this comes from

Don't know if that's ever possible to know though unless you train a model from scratch but remove all bike product photography and adjacent materials from the training data?

4XGB10 is in quite a different price bracket than a single M5 128GB Mac Studio...

Not like that doesn't already exist, try scanning and printing a bank note (US, not sure if other countries have similar)

Can you give some specific technical examples where 4.7+ are making the wrong architectural decisions?

If you are a bike nerd and like taking photos of your bike(s) you'll almost always try and have it drive train towards the camera, which would be front on the right side. Possibly this is where is this coming from? Generally also true for retail listings too

Got a source for that?

It's location tracking more than anything, so yes they would most likely be as the use all the same networks as a normal device

Probably a bit niche at the moment really. The only people interested in that are us nerds, and the product segment is very adhoc - especially for the local crowd where an epyc, with a bunch of pcie riders and some 3090s on a steel frame is considered optimal

Curl is one of the most audited code bases, so that's not surprising.

Put it in front of a common-ish Python or NPM package and see what it finds, likely going to be a lot more.