ulber

14 ms·

1.
by ulber 3y ago
>I think we need something that at least captures the general mechanism by which LLMs predict the next token. This bit varies a lot since the capabilities involved in prediction depend on the data. If the text is a math book and the prom
2.
by ulber 4y ago
I bet they forgot to include that it can generate that 770MWh every hour.
3.
by ulber 4y ago
Maybe we don't, but we should let market forces sort it out by doing our best to tax the externalities of power generation too.
4.
by ulber 4y ago
The argument the article makes is indeed a bit blunt. The idea here is that tipping enables better price discrimination, which in turn allows a higher "quantity" of coffee to be produced. With coffee shops that "quantity&quot
5.
by ulber 4y ago
As an aside, Universal Paperclips is the best clicker game I've played and has zero monetization. https://www.decisionproblem.com/paperclips/
6.
by ulber 4y ago
As a crude point of comparison, you'd be buying this service at about 917 euros/hour. With a pretty lax 1600 hours/year you'd have to make almost 1.5M per year to match the rate.
7.
by ulber 4y ago
This was a great read. I kept expecting it to reference the "Curse of Dimensionality" [1] as an explanation for why we see no close-to-average samples in high-dimensional spaces. I'm guessing the book this is an excerpt from
8.
by ulber 4y ago
I get the point but also fully disagree. At the point where computer cognition is good enough for humans to truly fall in love, build friendships or come to hate as an enemy, I personally think not recognizing it as AI will become a fringe
9.
by ulber 4y ago
I feel this description overstates the freedom. Academia is full of rather hard deadlines, as the conference you're targeting a paper for might be the only fitting one for many months. Missing these deadlines can delay the whole "
10.
by ulber 4y ago
You could also train for this kind of interconnectedness by designing tasks that are explicitly multi-modal. For example, you could: - Stack boxes collaboratively by controlling your own arm and communicating with another agent helping you.
11.
by ulber 4y ago
They definitely need a brushing after smoothies, but apart from that I'm not the most worried about bacterial growth inside them. They're not used for cooking, so as long as there isn't significant biomass inside there it doe
12.
by ulber 4y ago
Here's his response: https://cheapeth.org/whalegate.html It casts the situation in a rather different light - you should read it.
13.
by ulber 5y ago
If you're advocating for a single tax, let it at least be LVT [1]. But seriously, the point of this Finnish tax is to internalize externalities of vehicles that contribute excessively to global warming. Even if there's some loopho
14.
by ulber 5y ago
I don't think SSAO is really ever selected over more realistic methods for artistic preference. High budget animated films tend to go for ray traced methods even when they have a non-photorealistic style. SSAO is just cheap.
15.
by ulber 5y ago
I think you misunderstood the approach. MuZero is being used to optimize the choices made in the VP9 compression. In modern video encodings there's many ways to encode the same content. As a very simple example, you can vary how often
16.
by ulber 5y ago
>single player Lovecraftian text editor I hope that was not a typo and would love to see what a Lovecraftian text editor looks like.
17.
by ulber 5y ago
DPI is deep packet inspection in this context. HyperScan did some groundbreaking work for DPI, where you (1) have to match a huge number of regexes at the same time, (2) need high and consistent performance, and (3) don't need the star
18.
by ulber 5y ago
I tried to with a Quest 2. Immersed VR was the only option I found for getting multiple virtual monitors with my non-VR-ready work laptop (it has to encode and stream the virtual monitors to the headset), but performance was not good enough
19.
by ulber 5y ago
More likely ease of implementation. The use case shown here is an interactive UI element, so performance isn't really a concern.
20.
by ulber 5y ago
>His thesis is that the US is a world power on the decline and China will soon rise to take its place. Something I've found interesting about these kinds of theses (although I'm sure there's more to his) is that just belie
21.
by ulber 5y ago
I found your pitch really interesting, I'd love to try it. I, however, refuse to install an extension and change my default search engine just to be allowed to type in a search query. I'm sure you have some "growth hacking&qu
22.
by ulber 5y ago
For people reading the article, do read the top comment by Scott Alexander. It provides what seems to me like a grounded summary and opinion on the events.
23.
by ulber 5y ago
It's not an app (although those apps may help implement it, idk). It's a method from a researcher B.J. Fogg, who wrote a book called "Tiny Habits" [1]. I haven't read that one, but I listened to a similar approach i
24.
by ulber 5y ago
The "reverse a binary tree" problem gets often brought up as one of these tricky algorithm problems that you just have to know. However, the "algorithm" is to just swap all of the left pointers with the right pointers in
25.
by ulber 5y ago
Yeah, the internet is bigger now - all of the weird fringe stuff from enthusiasts is still there and there's even more of it. One thing that has happened is that the mainstream is there too now, so if you don't venture outside tha
26.
by ulber 5y ago
Driving is qualitatively different from coding: an AI that's pretty good but messes up sometimes is vastly more useful for coding than for driving. In neither case can you let the AI "drive", but that's ok in coding as s
27.
by ulber 5y ago
I wasn't aware of Lean not being sound and a quick search didn't come up with anything related to that. Could you link a source?
28.
by ulber 5y ago
It still manages to be poor UX: when reading the book it has already trained me to navigate by flipping forward, so if I actually want to go to the previous page I’ll just flip backwards. When I want to exit the book to the previous page I
29.
by ulber 5y ago
Why would anyone take that wager? I see no reason to believe that's possible with either Prolog or NNs when you're restricted to 80s hardware.
30.
by ulber 5y ago
I'd wager hedonistic adaptation is the big culprit. You don't end up saving that much unless you actually plan for achieving financial independence.

More ›