Ajen @ Ajen @sh.itjust.works

Posts

1

Comments

619

Joined

2 yr. ago

Huh? A priest did a nazi solute and was exiled for it. Refusing to give them credit for doing the right thing just makes you look bad. This is Lemmy, everyone here knows how bad the church can be.

That doesn't explain why the new bikes have older technology than the bikes they're urging people to trade in.

You're drawing a line between progressives and catholics, but the catholics just defrocked a nazi.

I could agree that it's an abbreviation, but I wouldn't go so far as to call it "diminutive."

If they send 2 emails per subdomain per year, that could easily be 10s of millions which would make the cost per email measured in thousandths of a cent. And I could see the number of subdomains being larger by a factor of 10, maybe more.

Another angle: someone with IT experience needs to manage the system that seems emails, and other engineers need to integrate other systems with the email reminder system. The time spent on engineering could easily add up to thousands per year, if not tens of thousands.

I'm guessing their figure is based on both running costs and engineering costs.

If your server goes down you'll miss incoming emails, and IMO residential ISP and power service isn't reliable enough for your main email address. If mail can't get delivered people get the impression you gave them a fake email address, which can be more than embarrassing.

It's the only service I don't self-hosted, although I do backup my email on my home server. Email protocols don't tolerate downtime, if your server is temporarily unreachable you won't get messages and people will probably assume the email address/domain isn't even valid.

It's a Linux joke though

I don't know what, if any, CS background you have, but that is way off. The training dataset is used to generate the weights, or the trained model. In the context of building a trained LLM model, the input is the dataset and the output is the trained model, or weights.

It's more appropriate to call deepseek "open-weight" rather than open-source.

I used the word "source" a couple times in that post.. The first time was in a general sense, as an input to generate an output. The training data is the source, the model is the "function" (using the mathematics definition here, NOT the computer science definition!), and the weights are the output. The second use was "source code."

Weights can be changed just like a compiled binary can be changed. Closed source software can be modified without having access to the source code.

The weights aren't the source, they're the output. Modifying the weights is analogous to editing a compiled binary, and the training dataset is analogous to source code.

Opioid manufactures also manufacture other drugs, like pharmaceutical grade cocaine. That's the only connection I can see - I don't think the cocaine molecule is similar enough to opioids for it so be a useful precursor.

This doesn't answer OP's question, but since other people might be interested, MIT also has free graduate level courses. If you choose to pay for a certificate of completion for the courses then they can also count as credits towards a degree at MIT, Harvard, etc.

https://micromasters.mit.edu/

That seems like a much lower salary than they would get running a for-profit tech company, though. Moving out of SF would also make it harder to hire people with a lot of experience in tech.

I'm happy to question my own biases, but only when the other party is willing to act in good faith. You clearly aren't. I checked, and none of your other posts in this thread cite sources that disprove what I've said.

Oops, I didn't realize I was posting on ML. That's my bad, I should have expected this bias.

For the record, I don't see any evidence or sources in your responses to me.

Have nice day.

I understand the subject well enough to know you can't back up your claims with evidence. You clearly have an agenda here...

"It's possible, you just have to train your own model."

Which is almost as much work as you would have to do if you were to start from scratch.

That doesn't mean it's straightforward, or even possible, to entirely remove the censorship that's baked into the model.

Your slicer should have generic profiles for each material which should give acceptable results with most brands. Density and flow tend to by the same across brands for any given material, unless it's a "rapid/high flow" or foaming filament, in which case the manufacturer prints the recommended settings on the spool. Basically if it's not printed on the spool you can use just the defaults in your slicer.