They've blew through the low hanging fruit, there was never that much of it, and are chasing white people now. Wild times.

If you can't be bothered to read it, ICE detained a brit grandma for 2 months even though she was 100% legal. ICE agents allegedly get a bonus per detainment, which motivated grabbing her along with her husband who was on an expired tourist visa.

Oh that's terrible we should really get some bases that are a lot closer. God that could take us decades though.

They will want credits on the tariff surcharge.

The end user, no.

It's a Fight Milk commercial.

The govt owes me so much ------- money now.

Weapon systems don't know what gender presses the button.

------- ammo on the line? Dudes are better.

The major take away from the recent electoral blowouts was, IIRC, to run a candidate who matches the district

#98 | Posted by jpw at 2026-02-17 11:41 PM | Reply | Flag:

Also this is Austin County. It's majority Hispanic. Neither candidate is one. It has a high % of new voter registrations. They register and then do not vote. It's a tech driven area with a high median household income. That means college educations. In Texas having a higher voter efficacy doesn't mean they vote, often the opposite, as they know their vote doesn't matter. Single party state ----.

The major take away from the recent electoral blowouts was, IIRC, to run a candidate who matches the district

#98 | Posted by jpw at 2026-02-17 11:41 PM | Reply | Flag:

That's poli sci 101 stuff.

Minimum Differentiation, Hotelling (1929)
Median Voter Theorm, Black (1948)

Those are the 33%- > 50% of Americans who are eligible to vote, but fail to show up when it counts.
Breate them?
Or welcome/encourage them?
And what are the magic words necessary to win not only their votes, but their commitment to show up at the polls when it counts?

#76 | Posted by A_Friend at 2026-02-17 07:56 PM | Reply

Plot twist. There are no magic words. It's a magic act, and that magic act is getting the party people that talk to much to shut the ---- up already.

The Roman Republic and Roman Empire are considered proto-nation states and the Holy Roman Empire is a direct precursor to the modern concept of a nation-state that began with France.

That fcnking guy?!?

#3 | Posted by hamburglar at 2026-02-12 06:10 PM | Reply | Flag:

You when talk ---- for a living, somebody in the crowd of people you are talking ---- about will eventually answer you with violence.

Children being vaporized by thermobaric weaponry

#6 | Posted by pumpkinhead at 2026-02-11 07:50 PM | Reply | Flag:

They need to go back to vaporizing children with traditional high explosive weaponry. Then it'll be fine.

Yet here in Houston I can get illegals from most of the Home Depots. ICE doesn't go to those anymore.

The FDA routinely approves drugs where the full biochemical pathway is not understood.

Do FDA standards apply to the device? If yes, it's SaMD. So it'd be the app/device developer, individiual clinicians, the 3rd party AI vendor if it exists, and the user themselves.

A fitness tips app has a low risk. A treatment recommendation model is very high risk.

Which lawyer wins?

#22 | Posted by snoofy at 2026-02-10 11:47 PM | Reply | Flag:

The one who's client has deeper pockets.

Though in the culture, grandpas in their 40s are definitely a thing...

#20 | Posted by LegallyYourDead at 2026-02-10 09:09 PM | Reply

Off by 30 years bruh.

Backpropagation explicitly computes the contribution of every hidden unit via gradients. If those contributions were unknowable, training wouldn't work.

Rumelhart etc, Nature, 1986. A foundational paper for ML.

Data science was my capstone dude.

Anthropics base salary for a phd in interpretive research is something like $500k/annual + bennies.

If the values had no meaning, downstream layers couldn't use them to make consistent decisions. They're not meaningless, it's just an encoded representation of the model is optimized to interpret. Math framework and task context is how you can interpret them.

If you were correct then backpropagation, debugging, and pruning would not work. Model updates would be blind luck. The entire field of Interpretive Research wouldn't exist.

AI tells me things I know are outrageously wrong and takes no responsibility for it.

#10 | Posted by Zed at 2026-02-10 05:27 PM | Reply

That's an LLM. This device isn't using an LLM.

The hidden layers are just that -- hidden.

#11 | Posted by snoofy at 2026-02-10 05:34 PM | Reply

They're not hidden, that's a euphemism that means it's not part of training data or prediction target.
It's the intermediate layer. h = sigma(Wx + b). Where x is your input vector, W is your weight matrix, b is your bias, and sigma is your activation function.

Training a model for cancer detection using annotated MRI data is a data science 1 topic, college Jr. level.

The hard part is teaching systems to older researchers and building a stack for their domain, and keeping someone on staff to maintain it. Students come and go from teams very quickly.

Stanford Will pay you 200k annual as a glorified research assistant if you can help their physicians develop ML models for their various topics.