Not from America, but can I ask... Is the SAT that difficult?

The SAT, by design, covers the entire difficulty range. Parts are easy, parts are hard, such that about half of students get less than half of it right, and 0.1% of students ace it. It's not merely pass-fail, they're trying to give a pretty granular rank to each student.

Thus, if the test is worth taking for a student (because they want to go to college), it's probably worth cheating on. Students outside the top 0.1% can appear better than their peers to improve their odds of getting into better universities, and students in the top 0.1% tend to be there due to intense extrinsic pressure, which may drive them to cheat to increase their certainty of acing it.

For a competent student, it's not hard to get an acceptable grade. For every student, it's difficult to achieve an exceptional grade.

It is not hard for many students to get a pretty good score (1350). But a score like this will not help you get into a top school, and might actually hurt you. Even having a very good score (1530) does not necessarily help you much, especially if you are from what is perceived to be a high-resource (wealthy) area. If you have a perfect score (1600), that would help you get into any school.

Admittedly, it wouldn't help that much if you are Asian and from a wealthy area, but if you cheated then you could spend time that you would have spent studying for the SAT instead doing something else that you could put on your application.

The slop sickens me. If deep down there still is a part of you that enjoys problem solving and creativity, then you should always do it yourself. The technology still isn't even that great.

I wanted to write a simple midi chord controller for my phone to make music, emphasis on simple. AI can't do it and couldn't even help me on small code fragments, because it doesn't look at the problem through the lens of my artistic creativity.

Vibe coding creates weak men because you reject the deeper meaning behind what you are creating. Vibe coding creates weak men because they can't code.

A single example being 14 cad a month, with the implication being that you have lots of monthly subscriptions for "small apps" sends shivers down my spine. I know YC has lots of wealthy people, but come on!

The security advantage I see in mtotp is that you never reveal the password to the system you are authenticating with, but that there is also no electronic device that can be compromised

I've lost the joy in programming, the only thing I'm good at, I now make horrible music, but at least I don't exist as the means to an end that I don't control.

I hear AI is better at music and poetry too. Go forth and prompt.

No Offense taken, but what's the point in using AI for anything unless you don't want to do it? I want live my life not consume information, is that really so bad?

Let the guy makes his music, why do we have to put this shit AI stuff in everything. Had to delete spotify for this bs.

Step 1 in this situatoon is to try and see if this is a known mathematically unsolved problem, and if it is, giving up.

Isn't this just trying to find a hamiltonian cycle, isn't this NP hard? That's when I would give up, especially because you put so many constraints in it to make it human walkable.

Edit: Of course you don't have to give up, but it's good to know what you get yourself into

Why would you give up just because something is NP hard if there are good algorithms to approximate a solution, and an approximate but good solution is useful?

Yeah, there are many problems which are np-hard in theory, but then realistic cases give you way more constraints that make them solvable. So many hard graph problems become way simpler when applied to real maps, because you know that if you start getting away from something, your minimum remaining distance grows. But on an abstract graph there's no real mapping to our dimensions.

That just means you didn’t encode the information you wanted into the graph.

There's no "information you wanted" in the plain np-hard version of traveling salesman for example. There's only cost. My point was that things get easier if you have the extra information and aren't solving the plain version anymore.

For constant n even exponential complexity reduces to constant time

I remember asking myself this question years ago, and came to 162 bits. I was just a kid back then so the logic is probably wrong but I do wonder how simple the encoding could be under those constraints...

Edit: Here are the Notes

0 Empty

10 Pawn

1100 Knight

1101 Rook

1110 Bishop

1111 Queen

32 + 32 + 472

2 times 6 bits: position of the kings

30 bits: color mask

120 + 2*6 + 30 = 162 bits

We can store the rest using the methods from the blog post and add 18 bits for promotion, giving 180 bits.

I'm sure this isn't the most efficient way, and I think I had other methods and considered things like the bishops being able to occupy 32 squares, though special casing doesn't make sense because of promotions.

Technically if you got 8 bishops/queens/knights/rooks You would occupy another 16 bits, giving 196 bits

I think the upper limit can be reduced at the cost of increasing the lower limit

EDIT2: I think I made the assumption at the time that to promote one piece you needed to capture at least one enemy pawn, giving the space for the two bits, which means the upper bound is actually 180 bits

Would love to see other people try in the comment section

Each pawn that wants to be promoted either takes: (a) another 'special' piece (knight/rook/bishop/queen), in which case it has already bought enough bit budget to later be promoted; or (b) another pawn, in which case this temporarily saves 1 bit (as the other pawn becomes a space), but then later we need 2 extra bits for the promotion, so we pay 1 bit extra per pawn in total

In the case of (b) there are now fewer pawns that can be promoted, and so worst case, we have to pay a budget of 1 bit per each of 8 promoted pawns.

So I think maximum required bits is only 162 + 8 = 170?

Among 4 pawns like white and black a&b pawns, you only need 1 pawn capture to allow the other 3 pawns to promote.

Great point.

So for each 4 pawn cluster, 1 pawn takes another pawn, and the net result is +1 bit once the captor promotes. The remaining 2 pawns in the cluster each need 2 extra bits when promoted => 2 x 2 = 4 bits. So 5 bits per 4-pawn cluster, of which there are 4.

So maximum representation would be 162 + (5 * 4) = 182 bits?

Yep, that increase the total in 3*3-4=5 bits, and you can repeat it 4 times, so the maximum is at least 162+4*5=182.

I'm trying to prove that is the worst case, but there are just too many cases. I guess I'll try to use a program o brute force it or just forget about it.

Actually, given this, we believe that 4 pawns must have been captured to reach 182 bits. So at least 4 pieces no longer need colors. If we store the color mask at the end, I think we can make it variable length, and truncate when no further pieces need colors assigned.

So then we need maximum 182 - 4 = 178 bits

EDIT: Equivalently, we could suffix each non-empty piece in the sequence with an associated color bit

Considering at least half of all squares are empty, further compression is in order for the empty space.

Also if you're encoding the king as a position instead of a byte sequence you would have to encode their space as empty, that's an extra 2 bits

I thought the same but realized you can retrospectively 'insert' the king positions into the position sequence, shifting the remaining sequence one square along for each king, so no more bits required though the data structure is unwieldy!

Only half of the squares are empty, you can almost make a chechboard pattern with the pieces. I don't expect an easy small worst case.

Honestly, if these image models still use diffusion with random seeds at their core, it might be actually more secure than blurring it yourself.

Honestly this should've been introduced with the new AI Features from the start, it's just shipping slightly too late to fully regain my trust.

I couldn't in good conscience work like that, I believe the risk of bad AI generated code due to the tiniest of output variation is far too high. Especially in systems that need to maintain a large state governed by many rules and edge cases.

I recommend having the AI do the typing while still reviewing, comprehensively testing and even dictating the exact shape of every line that you commit.