Don't the models typically train on their input too? I.e. submitting the question also carries a risk/chance of it getting picked up?
I guess they get such a large input of queries that they can only realistically check and therefore use a small fraction? Though maybe they've come up with some clever trick to make use of it anyway?
Yeah, probably asking on LMArena makes this an invalid benchmark going forward, especially since I think Google is particular active in testing models on LMArena (as evidenced by the fact that I got their preview for this question).
I'll need to find a new one, or actually put together a set of questions to use instead of just a single benchmark.
If there’s an issue with the core then the salt tank can act as a heat sink in a way a battery can’t?
The boiling / pressure water reactors all have requirements on active cooling being maintained in emergencies - I’m not familiar with this design nor to what extent the salt is intended to fulfill such a function, but it’s plausible that it could buffer things for idk 1h-3d maybe?
The holy grail is the “walk away safe” reactor, I would hope / presume all the novel / modern ones fulfill that?
A lot of employees at successful startups & FAANG make most of their money from the stock, no? And they need to buy houses and send their kids to fancy schools too, no? So sure, we can reduce it to stock holders, but I’d bet dollars to donuts the 90% of employees who aren’t posting on hn are at least passively ok with “improving metrics”, and some ambitious ones are driving the enshittification initiatives hard.
IMO the reason devs started being paid in stock in the first place is VC-style grow at all costs mentality. The fundraising economy didn’t work without fabricating compensation and only paying out on hits.
No other industry operates with such a blurred distinction between employees and owners. Well, save for the gig economy, itself a tumor on American-style big tech.
Personally I'd be much happier with a stable income with not much upward mobility but also not much risk of falling downwards. Which is what Europe is geared more towards. I don't constantly want to be in a race. Just to live my life.
If they employees want it, fine but don't be surprised if we customers start finding alternatives. And/or pirating their content (e.g. when it comes to streaming services).
But yeah American companies aren't there to support the employees. The only one they answer to are the owners or large shareholders (whichevery applies), and their only goal is to make those richer. Customers and employees alike are nothing but consumables, a raw resource you only treat right if you can't avoid it.
The animation on the page looks an awful lot like autoregressive inference in that virtually all of the tokens are predicted in order? But I guess it doesn't have to do that in the general case?
The example in the linked demo[0] seems less left-to-right.
Anyway, I think we'd expect it to usually be more-or-less left-to-right -- We usually decide what to write or speak left-to-right, too, and we don't seem to suffer much for it.
(Unrelated: it's funny that the example generated code has a variable "my array" with a space in it.)
So, in practice there are some limitations here. Chat interfaces force you to feed the entire context to the model everytime you ping it. Even multi step tool calls have a similar thing going.
So, yeah we may effectively turn all of this effectively into autoregressive models too.
This, plus the A/B testing of headlines to maximize clicks has lead inexorably to the current information environment.
Our intuitions, outrage, and knee-jerk reactions are being weaponized to gain clicks, votes, donations, and "action".
Many a dictatorship has fallen in the wake of social media revolutions. I wonder how long democracy can last?
In a would-be-funny-if-it-weren't-tragic ironic twist both of the two main US parties see themselves as the last guardians of democracy and frame their opponents as Evil, against which "any means necessary" is the only reasonable course of action.
(Yes, the party you disagree with is way worse and it's all their fault, this whataboutism indeed has to end, absolutely)
> After this article was published, a USDA spokesperson said Fong left the office Monday on her own accord.
> "She was accompanied by two friends who she paused to take selfies with on her way out. Security officials did not play any role in her departure,” the spokesperson said.
Given the preceding quote that she did not believe the terminations were lawful, that would be a complete reversal of her decision from days prior.
Phyllis Fong, a 22-year veteran of the department, had earlier told colleagues that she intended to stay after the White House terminated her Friday, saying that she didn’t believe the administration had followed proper protocols, the sources said.
In an email to colleagues on Saturday, reviewed by Reuters, she said the independent Council of the Inspectors General on Integrity and Efficiency “has taken the position that these termination notices do not comply with the requirements set out in law and therefore are not effective at this time.”
Fong declined to comment and the Office of the Inspector General did not respond to multiple requests for comment.
Ideally the "% Limit" column would:
1. Be right-aligned
2. Have consistent formatting (i.e. same number of digits after the dot)
3. A little bar underneath each number showing relative scale (i.e. top entry is full width, last entry is 216.7 / 32571.4 = 0.00665307601, though maybe on a log scale for confusion? ;)
I guess they get such a large input of queries that they can only realistically check and therefore use a small fraction? Though maybe they've come up with some clever trick to make use of it anyway?