t_mann

15 ms·

151.
by t_mann 2y ago
Curious thought: could those large price movements have something to do with the fact that DeepSeek is financed by a hedge fund (rather than the more typical VC)? It is unclear how DS will make money from its current strategy of sharing muc
152.
by t_mann 2y ago
Paul Romer has received the Nobel Memorial Prize in Economics. While that's just an appeal to authority, it still has more substance than dismissing a long post without even having read the majority of it. I encourage you to read the w
153.
by t_mann 2y ago
Ok, but then I'd still prefer a method that sends users to a unique URL. OP's method may help with obfuscating the changing of the code, but I'm sure there are ways to better achieve that without having to introduce this quas
154.
by t_mann 2y ago
You don't even need to change the QR code to treat users differently, that's the point. You just send users to some fixed URL, which is by far the most common use for QR codes (hence completely unsuspicious), and you decide who ge
155.
by t_mann 2y ago
All of those could be done much more stealthily server-side, though, I don't get what the QR code modification would add here? Also, neither use case makes use of the hack described in the OP. Where I could see an attack based on that
156.
by t_mann 2y ago
> Because memory consumption grows linearly with sequence length, the maximum context window is limited by practical hardware constraints I thought the number of parameters grows quadratically with context window length - what do they me
157.
by t_mann 2y ago
> Cheaters, by definition, avoid real work, so they won't be using custom models unless the stakes are sufficiently high. I think it's a non-sequitur, some might get more enjoyment out of tinkering with LLMs than out of whateve
158.
by t_mann 2y ago
Are you in the "it's just like with calculators" camp? I think there's a fundamental difference between a tool that just simplifies some very specific solution steps vs a tool that takes the literal assignment question,
159.
by t_mann 2y ago
The post clearly described a non-convex shape, and "flat-surfaced shape" should be a pretty clear instruction as well. The shape described may be visualized as a cylinder with a cone cut out, where the base of the cone aligns with
160.
by t_mann 2y ago
A sphere is only locally approximable by flat surfaces, but it's nowhere actually flat, which was a requirement in the previous post.
161.
by t_mann 2y ago
The thinking in the other post, that the mass increases as you move away from the center, in a manner that the two effects cancel out, intuitively seems like it should be feasible. Remember that the center of mass is just an abstraction, yo
162.
by t_mann 2y ago
That way you should be able design a disc-shaped earth with constant strength of the gravitational force on the whole surface. But it would still have a center of mass (likely lying outside the shape you're describing, in the void bene
163.
by t_mann 2y ago
That's comparable to the revenue of a single (top-level) soccer club: https://www.sportico.com/leagues/soccer/2024/manchester-unit... https://www.espn.com/soccer/story/_/id
164.
by t_mann 2y ago
Have you tried Samsung DeX? I find it pretty amazing, you connect your Android smartphone to a docking station, and you have a usable Desktop OS. The only thing holding it back from becoming a full workhorse for eg document editing imho is
165.
The mindset that brings unlimited willpower(bbc.com)
4 points by t_mann 2y ago|2 comments
166.
by t_mann 2y ago
Didn't MS once say something to the effect that Windows 10 would be the 'last' major Windows version and that they'd switch to a more continuous update model? In reality, it was probably one of the versions with the shor
167.
by t_mann 2y ago
Which comments are you looking at? By a brief scan, the vast majority of comments, including practically all the top-voted ones, are calling out his "opsec troll" as a deflection strategy, which appears to have been confirmed now.
168.
by t_mann 2y ago
> I was told that I had the second biggest crowd, second only to a Pokémon bean bag game (which did look pretty cool). Some adults were curious, but most importantly, a handful of questions from kids who wanted to know how I built it. It
169.
by t_mann 2y ago
Tbf, it's not like this approach can't work at all. Airbus was born out of a similar dynamic, and it's giving Boeing a run for its money now. Afaik France has a couple of other giants in technology-heavy industries such as sh
170.
by t_mann 2y ago
Another good candidate: the Privilegium Maius [0], which contained several provisions to elevate the status of the House of Habsburg and their territories in central Europe and helped them rise to the first tier of European aristocrat famil
171.
by t_mann 2y ago
Here's 8 densely printed pages of contributors to a discovery that lead to a Nobel prize: https://arxiv.org/pdf/1207.7214#page=26 Guess how many of them were included in the prize. It's a shame that the Nobel
172.
by t_mann 2y ago
I suppose the logic could be: if you're going to consume AI generated content anyway, why not use a setup where you have control over the system prompt and other parameters? Not sure if ChatGPT qualifies there, though.
173.
by t_mann 2y ago
How so?
174.
by t_mann 2y ago
> Adobe will simply produce a crap Figma, bundle it in Creative Cloud, and Figma will die a slow death > My priority is the widespread availability of high quality products and services Seriously?
175.
by t_mann 2y ago
I don't care about Figma or Adobe, but you simply haven't made a good case why antitrust enforcement is good in general but bad in this case. Your only argument is that it makes it harder for other companies to grow to that size,
176.
by t_mann 2y ago
So anti-trust action is good against FAANG and bad against Adobe? What's the general principle here?
177.
by t_mann 2y ago
So you start by saying anti-trust action is negative for the economy, then quote someone who says we should have more anti-trust action?
178.
by t_mann 2y ago
The argument rides on the well-known Goodhart's law ( when a measure becomes a target, it ceases to be a good measure ). However, it only puts it down to measurement problems, as in, we can't measure the things we really care abou
179.
by t_mann 2y ago
> I use this method often for prediction applications. First, it’s a sort of hyper parameter selection, so you should obviously use a holdout and test set to help you make a good choice. What the article is talking about is inference, n
180.
by t_mann 2y ago
Can you point to the parts of the document, or other resources about Mr. Beast, that warrant a comparison with tobacco companies?

More ›