> They’ll train on your uploaded docs and photos, to be sure, but it won’t be correlated to your chats and projects, unless Anthropic is doing things that would make their counsel have heart attacks.
And if they do, they'll apologize with a blog post.
built this to play with some distant friends - has hangman, skribbl-like drawing and guessing game, word bomb party (personal fav) and a cool card game. would love to get suggestions and add more so try it out with friends, partners and family. ps: no ads ever
I somewhat suspect that's a temporary state of affairs. Nothing prevents a couple tons of batteries being put on a truck with a CCS charger to achieve the same thing.
It won't be efficient but neither is driving a liters of fuel to someone from a gas station.
This was Providence in Oregon. I thought that's what it meant, too. But in this case there's a contracted "allowed amount" that is all the insurance will pay the doctor. If the doctor chooses to charge more, you get to pay that.
Which really freaks me out. If you're insured at a contract rate of $5000 and they bill $30,000...?
Huge disagree, as a manager. It depends on the thing, of course. If you're rushing into a giant re-architecture by Tuesday, that's dumb. If you have some change you want to make, go for it.
My default is to trust engineers based on my experience with and expectations for them. If they want input—anything from a deep review to a gut check—I'm happy to help. If you're looking for a gut check, this is a fine way to do it. It communicates your level of confidence, which is an important data point for me.
If someone is adding a GH action, do we need a prototype? Maybe! But also maybe not. Bias towards action. Not YOLOing, not hacked together crap, not vibe code merged without review. But I've found that great engineers are often more hamstrung by permission checks than the issues they're meant to prevent.
Nope , you haven't been in this community. The most fun i had been in gaming . They teach each other and help each other , share unit , share economy. Its not like COD or LOL communities.
You seem to mix up 'applicable' and 'funny'. Jokes are supposed to be funny. And given that there is already a OS part of the project, the joke doesn't even make fucking sense anymore.
You'd probably get much further along by fine tuning a small BERT style encoder model based classifier for it. IMO, even something as simple as training a linear classifier on the CLS token embeddings from a frozen encoder might work.
We had a number of ways of detecting sockpuppets, and I was part of the mod team at the time, so I could verify they were real users. The change in standards was a combination of things; the old guard, many of whom had left at that point, leaned towards computer literate STEM students who had grown up reading actual novels. In contrast, the new generation of posters was full of ESLs, nontechnical users, and people who grew up reading stories on FFN and RR; the new users were also unfamiliar with forum etiquette and acted like they were on discord. Suddenly, I was having to remind users that English was the official language of the board.
After 2023 or so, a new problem popped up: posters would read a chapter, hallucinate a series of events that never happened, get mad about what they thought they had read, and then attack the author! Arguments born from a complete lack of reading comprehension were breaking out in all the busiest story threads, with waves of reports every time a new chapter was published. It would not be an exaggeration to say that the average IQ of the board had dropped 20 points.
There was also the monetization aspect to consider; fanfiction used to be by fans, for fans - the work of amateurs, in every sense of the word (amātōrem, 'lover'). While that certainly didn't mean everything was /good/, it at least meant it was a work born of genuine passion. The new authors treated it like a business, churning out updates every week and advertising to readers that they could read several chapters ahead on their patreon. Without even touching the ethics of the matter, you can imagine how the incentives changed the output.
Using even double the total tokens and taking, what, 2-3x the time?, still seems worth it if prices are 5x+ cheaper (which OpenRouter [1] claims is the case).
On NeuralWatt for my personal projects at home (not affiliated, just a happy customer), I get so much more mileage out of GLM than I get out of Claude at work, specifically because it's priced as a hammer I can pound any nail-shaped-object with, not a delicacy I need to carefully budget-analyze to try to figure out if it's worth burning my monthly spend limits on this task.
I suspect that the GP's numbers are total, not per ton-mile of cargo (which, to my mind, makes them useless for efficiency discussions). They might make sense for discussion of the actual article, though.
The ideas of GA aren't new. As the article explains, the ideas of Clifford algebras are common place throughout research and deserve to be introduced earlier. Understanding bivectors and wedge products is much more important than, for example, Euler angles in my opinion.
The geometric product on the other hand obscures much of the structure, and serves no pedagogical or fundamental purpose. Not everything that's new is better, just by virtue of being new.
You might have misunderstood my point about what's niche, or I misunderstood which formulation of Maxwell the post I was replying to was referring to. Either way, it feels this discussion went off the rails rather immediately...
The government mandate forcing them to restrict access of new models to American has really cut their legs under them.
This identity verification is a best effort to kinda stay afloat: they can now offer bleeding edge models to US nationals, but not to the other 95% of the world. Their influence is gonna tank quite seriously if the previous mandate is not reversed.
Realistically, their choices are to either implement this, or restrict access to new models entirely, which is a sure way to fall into complete irrelevance.
Each test is responsible for a part of the UMAT subroutine, checks it's correctness, compares the subroutine (FEA) with the analytical. Plot their results and show whether the tests passes or fails. AI is built into the process too, but in a manner where they are deterministic and with strong grounding on the physics.
You can also write your own tests in Fortran and run it. You can also share it to the community for others to use.
