At the expense of 18% more watts! Comparing the x090s in the article.

Ah perhaps it is a LinkedIn lunatic afteral. I wouldn’t want to work with that toxicity for millions lol

Ah I saw it mentioned but paywall blocked the rest lol

Distillation is basically reverse engineering for AI cool.

I thought training a model from AI data reduced its effectiveness? Wouldn’t this mean they still did something crazy since they got the opposite results?

If an employer doesn’t think I am worth it and won’t pay me more without asking I’d rather find one that apreciates what I do and leave. Last place I worked had every VP call me and ask if they could do anything or pay me more. Welp should’ve asked me that a year ago mate. They also said misogynistic things and made it a lot easier to leave lol.

Fines would have to be something crazy like Tik Tok ban $5000 per user type of deal

Sounds similar to OpenId connect for authentication, service requests scopes which pulls varying info and user can be shown a consent screen with what data is being requested for approval.

I’d like a similar model for data sharing, though you will need privacy laws since you can revoke access in this case, but currently there would be nothing preventing storing your data at the time elsewhere or sharing it.

Nzbgeek and nzbplanet are good. From my experience these 2 have very similar quality.

Drunkenslug is a lot harder to get into but it does find items the other two don’t have on occasion. On the downside, drunkenslug lacks a one time subscription and the other two have a one time subscription. It does however have a free tier so you can probably be fine still using it as a last resort to try finding harder to get content.

Docker Hub is the bane of my existence lol. I updated every image I use that has github as alternative yesterday now giving time for my rate limiting to go down. Unfortunately still a few that are doxker hub only ironically including lemmy!

Ah yeah I haven’t seen anything on that. That’ll be next weeks headlines probably lol

They aren’t necessarily smaller from my understanding. Say it has 600B parameters, it more efficiently uses them. You ask it a programming question, it pulls 37B parameters most related to it and responds using those instead of processing all 600B.

Think of it like a model with specialized submodels that a specific one may provide the best answer, and uses it.

But the robot has to run windows vista

I believe it would have lowerr operational costs assuming the models the only thing different and they target the same size. Deepseek does the “mixture of experts” approach which makes it use a subset of parameters thus making it faster / less computational.

That’s said I have a basic understanding of AI so maybe my understanding is flawed.

I added renovate to my project over the weekend. I got 26 PRs for updating things I have missed, so it is working well for most part!

The only issue I have with it are a few docker images come from docker hub and I am getting 429 response codes for pinning digests. Do you have any tips for renovate on this? Ideally I’d like it to just update and pin digests on the next update to avoid api hits.

I am doing a regex datasource for most of them since my k8s resources are in yaml files and found right now it strips - alpine and such from the version tags… Haven’t looked into this issue too much yet though.

If its at the end forever it might be because you are putting too much soap in.

In the US you’d have a 4 to 6 hour wait, $3000 to hospital for using the ER, $2000 for the doctor, and if there were scans and such a $$$$$ for using them! Oh also they will take months to bill you but also send it to debt collectors if you don’t pay it for a month so then all of your personal data is sold and you get harassed to pay your debt!

My primary use would probably be a model for home assistant to reason what was said, and based reason what is desired like “Turn off all outlets” being said in the living room would turn them off. Or saying “its cold” to adjust the thermostat. Also hoping it would be able to figure out speech to text inaccuracies and reason what is wanted still. I think a smaller model would be an appropriate use for this also since response times should be on the quicker side.

I wonder of my Subaru is exposed since the connectivity services use Toyotas. I imagine it has its own employee portal for them to utilize. That also raises the question of what privacy rules are followed, Subarus or Toyotas?

With these new models is it worth trying it without a Nvidia GPU / running on CPU?

Of course it’s Ohio!