> Polynomial growth (t^n) never reaches infinity at finite time. You could wait until heat death and t^47 would still be finite. Polynomials are for people who think AGI is "decades away."
> Exponential growth reaches infinity at t=∞. Technically a singularity, but an infinitely patient one. Moore's Law was exponential. We are no longer on Moore's Law.
Huh? I don't get it. e^t would also still be finite at heat death.
Is this really true? I played a few games with it in August. It's not very good.
It's one of those old programs where 95% of the moves are pretty strong. But if you just do nothing and sit back it will occasionally make a random blunder and then you grind it out. I figured it's how they were able to weaken a chess engine back in the day; can't adjust the overall strength, so add random blunders.
I'm only about 2000 on lichess but I beat it pretty much every time, especially once I realized there is no reason to try anything sharp.
My suspicion is that the bot was a fairly standard chess bot, but the difficulties were set based on computation time. As airplane computers got better, it turned into a beast.
As a result, if you tried this on older planes, it might have been “easier”
One of my first paid iOS dev jobs was porting a Go game from iPad to iPhone, don't even think the 4 was out yet. It also used computation time based difficulties. By the time I was done writing it, I knew a few tricks I could eke a win out with on 19x19.
When the iPhone 5S came out, I tried it on a whim to check the UI scaling etc... the beginner difficulty on a 9x9 board deleted me. It was grabbing something like 64x more samples per go, the lowest difficulty on the 5S (instant responses) never lost a single game vs the highest difficulty 3GS (15 second turns)
iPhones had a lot of moments like that. Silly bullshit like "what if every pixel was a cell in a collection view" would go from "oh it can barely do 128" to "more responsive than that was, with 2 million" in a few gens.
One of the minor weird things about iOS development early on was just how fast the transition was from the simulator being dramatically faster than actual devices to the simulator being slower than devices. When I started out you’d get things working nicely in the simulator and then discover it’s an order of magnitude too slow on a phone. Just a few years later and my phone was faster than my laptop until thermal throttling kicked in.
I was maintainer of the Chess app from the early 2000s to about 2015. We first noticed in 2004 that level 1 (which was then "Computer thinks for 1 second per move) was getting stronger with each hardware generation (and in fact stronger than myself).
So we introduced 3 new levels, with the Computer thinking 1, 2, or 3 moves ahead. This solved the problem of the engine getting stronger (though the jump from "3 moves ahead" to "1 second" got worse and worse).
A few years after I had handed off the project, somebody decided to meddle with the level setting code (I was not privy to that decision). The time based levels were entirely replaced with depth based levels (which eliminates the strength inflation problem, but unfortunately was not accompanied by UI changes). But for some reason, parsing of the depth setting was broken as well, so the engine now always plays at depth 40 (stronger than ever).
This should be an easy fix, if Apple gets around to make it (Chess was always a side project for the maintainers). I filed feedback report 21609379.
I found a used copy of Warcraft 3 at the store about ten years after it came out, proudly brought it home, fired it up and didn’t recall the graphics being quite that awful, but the first time I tried to scroll the map sideways it shot to the far end because they didn’t build a timing loop onto the animation and I shut it down, disappointed.
Unfortunately they never released a remastered version of it. They seem to have made some clone of it called “reforged” whatever the fuck that means.
Reforged was received poorly because it was a lazy half assed job that was a blatant cash grab. Not because culturally we have moved on and the game has aged beyond being fun
You probably knew this, but wanted to make sure others knew that the reason they ended the franchise is not because there was no market, but instead it was pure unadulterated greed that led to that situation. In an alternate reality they would have actually done the remake justice and there would be a lively competitive scene
There are various hacks and tools for games (especially DOS games, but for W3 there may exist the same) which delayloop various calls to slow things down enough "to work".
The Dolphin emulator has run into similar things; usually doing things "too fast" just gets you more FPS but sometimes it causes the game to go insane.
This is pretty much the experience of trying to play any game from the '90s on modern hardware. It always requires a bit of tinkering and usually a patch from the modding community. Funniest one I've found is Fallout Tactics. The random encounter frequency is somehow tied to clock speed so you'll basically get hit with random encounters during map travel about once every half second.
I've been enjoying Total Annihilation since 1997. Still works fine on fairly modern hardware with Windows 11. No modifications other than some additional maps that I downloaded decades ago.
Sorry if this is a dumb question but did you patch it to the latest version? I don't know if the in-game updater still works but from memory you could download some sort of patch exe file and update it that way.
The original Wing Commander was like that. Playable on 286s/386s, then Pentiums and beyond showed up and it was unplayable. The game started in the "simulator" to show you the controls, and you'd get blown out of space in about 0.5 seconds.
The original Wing Commander brings back memories! I remember being amazed by the graphics and the story.
These days I cannot stand games with cliched storyline and tend to skip the cutscenes, but back then it all seemed so amazing... like a cross between a movie and a game.
I remember playing it later and running into speed issues too, but usually there was a way to tweak the emulator in order to fix this.
> they didn’t build a timing loop onto the animation
Wow.
1984 (!!!) IBM PC (DOS) port of the game Alley Cat had timings built it. They actually used the system clock if I remember correctly, so it would always run at the correct pace no matter how fast the computer. Last I checked it, decades later, it still ran at the correct speed!
AFAIK the only reason Chess even ships at all anymore is as a burn utility. They'll set it to AI vs AI at max difficulty to stress the system and make sure the cooling/power management works.
Never heard that one (it may indeed be used that way, but if it were the only reason Apple would probably keep it in the Apple internal parts of their OS installs).
It would also be of limited use, as the engine is purely CPU based; it is single threaded and does not even use SIMD AFAIK, let alone GPU features or the neural engines.
They aren't talking about the site, they're talking about their strength (as measured by that site) so it can be compared to the numbers in the article.
> I figured it's how they were able to weaken a chess engine back in the day; can't adjust the overall strength, so add random blunders.
In tom7’s Elo World, he does this (“dilutes” strong Chess AIs with a certain percentage of random moves) to smooth the gradient since otherwise it would be impossible to evaluate his terrible chess bots against something like Stockfish since they’d just lose every time. https://youtu.be/DpXy041BIlA?si=z7g1a_TX_QoPYN9b
2. I played a chess bot on Delta on easy and it was really bad, felt like random moves. I beat it trivially and I am actually bad at chess, ~1000 on chess.com. I wonder if this one is different?
That's true, I'm 2050-2100 lichess, around 1800 on chess.com. Never played a rated tournament but played some rated players who were 1400-1500 rated USCF, and they were roughly my strength, maybe a bit better. Still the Delta bot, easy mode, was much, much better than me.
I think it depends on the pool to which you're comparing. Being top 2% of all programmers is not so impressive if you include everyone who's ever taken an Intro class. Top 2% of people who do it for a living is much more significant.
I'm in a similar boat as the other posters (2050-2100 lichess, 1400 USCF). The median active rating for USCF is around 1200 and likely much higher if you don't include scholastic players, so if we compare against the OTB pool, "2000 lichess" is probably closer to top 50% than 2%
I mean, if you’re in the top 3 percent of anything, yes that’s pretty good, but not unbelievably so, especially in the field of chess. If for instance you randomly put together a classroom full of chess players, there’s decent odds one of them is better than top 3%. Two classrooms and it’s almost a certainty.
Put another way, looking at chess.com users, there are ~6 million people who would count as the top 3 percent. Difficult to achieve, yes, but if 6 million people can achieve it, it’s not really a “humble brag,” it’s just a statement.
It made me smile to hear “I’m only 97th percentile” isn’t a humblebrag. You may be employing an old saw of mine, you can make people* react however you want by leaning on either percentages or whole numbers when you shouldn’t.
* who don’t have strong numeracy and time to think
I heard it's never intended to be the same since initial rating for Lichess and chess.com respectively is 1500 and 1200. So they should have 300 rating difference on average. Quite fitting with what the other commenter claims actually.
I don’t think it would average out to a 300 elo difference simply based on the starting rating being 300 apart.
If everything else was the same, and people play enough games they will average out to the same elo.
The difference is caused by many factors. People don’t play enough games to sink to their real elo, the player pool is different, and you gain/lose fewer points per game with Lichess’s elo algorithm.
ELO is relative. There's no reason why a GM ELO should be 2800 or 280 or 28000. So it's all decided by ELO of every other person. So if the ELO gain/loss calculation and audience of Lichess and chess.com are exactly the same, because of different starting position, I don't think they'd converge to the same ELO but instead will differ by starting position difference.
Also I can't really prove it mathematically but I guess average ELO would also hover on the starting ELO. Because I can't see why it would hover anywhere else and any ELO gained would be lost by someone else.
When I started playing I believe chess.com let you select whether you’re beginner, intermediate or advanced and your start elo was based on that. Could be wrong, and it could’ve changed since.
This was my experience on a long Delta flight, I don't remember if I picked easy or not but it was laughably bad. I took its lunch money for a game and then turned the screen off. I was mostly irritated by the horrible touch interface, it felt so laggy among other issues. (I don't have a ranking, I barely play these days and usually just in person, but my memory says around 1400 back in the yahoo chess days as a teen but it's probably closer to 1000 now.)
I wonder if it's different on different planes? I can easily beat my friend and he won a few games on a flight, I played on a different flight and got crushed for two hours straight. I'm probably 1400-ish
I still have my old PowerBook G4 from 2005, with some not-that-old Debian currently installed. Every time my main laptop goes out commission, I get the G4 back out and use it for a few days. It's good enough for most of my work, though modern web-browsing is a challenge. (Maybe one that somebody has solved, I haven't dug at all.)
Does there exist a person who would make this argument straight-faced? I am a professional mathematician and have yet to hear of anyone coaxing an even slightly interesting new theorem out of AI. I think the day is clearly coming but it's not here.
fair enough, i suppose im a believer that the seeds are planted, the day is soon. and i must say, it seems more worhtwhile trying to figure out how to finetune an llm/implement reinforcement learning that could do some form of pure math, than it is to try and do new pure math by hand
I do research in this field. LLMs can be used as (ridiculously inefficient) implementations of some search algorithms that we haven't yet identified and implemented in software ourselves, but which can be inferred from a statistical analysis of the literature. Sometimes those search algorithms generalise to new areas, but more often than not, they flail. The primary advantage of a language model is that it's one big algorithm: when one subcomponent would flail, another (more "confident") subcomponent becomes dominant; but that doesn't solve the case where none of the subcomponents are competent and confident in equal measure. In short: to the extent it's useful, it's a research dead-end. Any potential improvements we understand are better implemented as actual search algorithms.
You've probably seen that thing where ChatGPT cracked Enigma[0]. It used several orders of magnitude more computational power than a Bombe (even given Moore's Law, still thousands of times more electrical power), and still took two dozen times longer. You would literally be better off doing brute-force search with a German dictionary. Thus is it with mathematics: a brute-force search is usually cheaper and better than trying to use a language model.
Terry Tao is one of the staunchest knowledgeable advocates of GPT models in mathematical research, and afaik he doesn't even bother trying to use the models for proof search. It's like trying to build a house with a box of shoes: sure, the shoe is technically more versatile because you can't use a hammer for tightening bolts (the shoe's sole has enough friction to do this) or foot protection (the shoe is the right shape for this) or electrical isolation (the bottom surface of the shoe is largely rubber), but please just use a hammer if you want to manipulate nails.
It's not possible to write (usefully novel) proofs with an LLM, but we have other algorithms that can do that. Perhaps a reinforcement learning component could improve upon the search strategy in some way, but there's no compelling reason to use a predictive text model. (There's not even good reason to believe that naïve reinforcement learning would improve the mathematical ability of a system: RL says "that was good: do more of that", and mathematics is about discovery: thinking thoughts that nobody has ever thought before.)
Yeah, I have been teaching calculus for ... several years, and it's very clear that the curriculum has been dumbed down over the years across the board. The last place I was at removed infinite series from the curriculum altogether. Epsilon/delta proofs were removed from the AP exam not so long ago. Some calculus classes don't use any transcendental functions because students can't handle it. At the same time grades have gone up, I suspect mostly because exams are usually weighted less than they were in the old days.
It is probably true however that there is more of a baseline expectation that you take it in high school. Whether this is sensible I do not know. It is also true that the very best students are better than ever, because the materials on the internet have gotten better and better over time.
I agree with the first two of these, they are great. (And I bet the third is too, I've just never needed it.)
If I had to submit one tip it would be to set everything up with a Makefile or similar. I keep my transactions spread across quite a few separate files for different accounts, and the actual commands I issue to include the right files are very long. Similarly I have various plotting and summary scripts whose exact syntax I don't usually remember. But with make I can just "make cashflow" "make balance 'A=Checking'" "make balance-plot 'A=retirement'" and so on.
There is ledger-mode and beancount-mode which are both nice (depending on which program you use). I would say the majority of what I do in practice Python scripts to convert statements to ledger; the amount of stuff I do by hand is minimal enough that it would be easy to live without the emacs mode.
I use emacs + beancount-mode + some helper elisp scripts.
On average, it takes about 30-40 minutes per sitting to do a weekly review, and that's mostly checking g what hides behind amazon/ebay/Google payments.
I've found that so few of my accounts support ofxget that it's not worth the trouble of dealing with it. I just occasionally download year-to-date statements from all my accounts and then run scripts to convert them to ledger format and append only the things after a given cutoff date (things like the credit card script are pretty interactive so I don't want to overwrite the old data). To know which accounts are in most dire need of a check, I have a script that shows the last transaction on each statement-generating account.
This basically works for me and catching up doesn't really take long. There are some accounts that I update very infrequently, but I've gotten over my OCD about this and it doesn't really matter. My wife doesn't appreciate being hassled for the statement from her work HSA every month -- nor does it really matter how much is in there at any given moment, unless we have a big medical expense -- so we just occasionally sit down and catch everything up, maybe twice a year.
Overall I find the automation means it's vastly less work than when I used gnucash, and the flexibility in expense structures and ease of assigning things mean I have much better budget data than when I used mint.com.
I'm an academic and try pretty hard to take the train. Just got back from a 19-hours-each way Greyhound+train+Greyhound ordeal to give a couple talks. My destination was only two not-that-big states away.
They don't make it easy. Last year I had a reimbursement rejected because I took a two-night train option instead of flying -- the finance office considered this a vacation. Any time you do anything other than fly you need to provide comparison data showing that flying would've cost more. But those who fly even when the train is reasonable (for a normal person) face no such requirement.
> Exponential growth reaches infinity at t=∞. Technically a singularity, but an infinitely patient one. Moore's Law was exponential. We are no longer on Moore's Law.
Huh? I don't get it. e^t would also still be finite at heat death.
reply