# Thomas

Karma: 1,198
Page 1
• Either I have no clue, ei­ther …

30 May 2019 9:51 UTC
2 points
• I’ve done some bench­mark­ing in 2018. I bench­marked an “AI soft­ware” we de­vised, by some bench­marks mostly I in­vented, too. Which doesn’t look very good, I know, but bear with me!

For one, I have given an un­solved Su­doku puz­zle to this soft­ware with two work­ing names, “Spec­tor” and/​or “Profounder”. It con­cluded, that for ev­ery X and ev­ery Y: X==Y im­plies that column(X) != column(Y) and row(X)!=row(Y). (Zero Su­doku topic knowl­edge by Spec­tor is, of course, a nec­es­sary con­di­tion.)
With sev­eral un­solved Su­doku puz­zles, Spec­tor con­cluded that sub­square(X) != sub­square(Y). Just for one puz­zle, the con­cept of “3 by 3 sub­square” isn’t eco­nom­i­cal. It’s eco­nom­i­cal for sev­eral of them, though.
The sec­ond bench­mark I in­vented, was giv­ing the string “ABCDEFGHIJKLMNOPQRSTUWXYZ” to Spec­tor. The string gen­er­at­ing al­gorithm would be sim­pler if the let­ter “V” wasn’t miss­ing. This is the way Spec­tor no­tices some­thing might be wrong with the given string. (Zero alpha­bet topic knowl­edge by Spec­tor is, of course, a nec­es­sary con­di­tion.)

Yet an­other bench­mark was num­bers from 3 to 122. Each la­beled by 0 or 1, de­pends if it’s non­prime or prime. The sim­plest gen­er­at­ing al­gorithm is a sort of Eratos­thenes sieve. Not for num­bers, but for their la­bels. Spec­tor finds and gen­er­ates it, with zero knowl­edge about primes.

Another bench­mark was in­spired by a mis­take some­one made. There is a nurs­ing school here some­where, which sends their stu­dents to prac­tice in a nearby hos­pi­tal for a day or two ev­ery week. Ex­cept for fresh­men in the first year. They teach them ev­ery­thing else in this school, of course, in­clud­ing the gym (boys and girls sep­a­rated there) and they feed them all once a day, too. It’s stan­dard in this part of the world. But the school does not feed them when they are at the hos­pi­tal.

So they for­get to feed girls from 2B de­part­ment on Thurs­days when they are in school. They for­get to in­clude that into their sched­ule. Boys from 2B have eaten while girls were ex­er­cis­ing, but poor girls were for­got­ten and no­body no­ticed.

I asked Spec­tor, giv­ing him the school sched­ule in CSV for­mat if any­thing is wrong with it. Spec­tor did con­clude, that ev­ery stu­dent has a lunch break once a day when not prac­tic­ing, ex­cept for those girls on Thurs­day. Which was (prob­a­bil­ity-wise) odd enough to be sig­nifi­cant.

Spec­tor/​Profounder is all about one mayor and three to five lesser tricks. To find a gen­er­at­ing al­gorithm for ev­ery part of any data it gets. This is the mayor. Then to see if some small data al­ter­a­tion would mean a sig­nifi­cantly sim­pler gen­er­a­tion. Then to eval­u­ate the prob­a­bil­ities and needed com­plex­ities. And then Spec­tor also asks it­self, what data changes are pos­si­ble but which con­serve already ob­served rules. Which is par­tic­u­larly handy in the un­solved Su­doku case for ex­am­ple.

We will do some more bench­mark­ing this year.

• For Eve and her ap­ple pieces. She may eat one piece per sec­ond and stay in Par­adise for­ever be­cause at any given mo­ment only a finite num­ber of pieces has been eaten by her.

If her eat­ing pace dou­bles ev­ery minute, she is still okay for­ever.

Only if she, for ex­am­ple, dou­bles her eat­ing pace af­ter ev­ery say 100 pieces eaten, then she is in trou­ble. If she su­per­tasks.

• I tend to agree. I don’t know is it just a habit or some­thing else, like a con­ser­va­tive pro­file of my­self and many oth­ers, but that doesn’t re­ally mat­ter.

The new site isn’t that much bet­ter. Should be sub­stan­tially bet­ter than this one for a smooth tran­si­tion.

• 8 Nov 2017 7:50 UTC
0 points
in reply to: curi's comment on: AGI

Please, fo­cus only on what has been said and not on how it has been said.

Now, there is a pos­si­bil­ity that all is wrong from my side. Of course I think how right I am, but ev­ery­body thinks that any­way. In­clud­ing this Tem­ple guy with his “don’t code yet”! I won­der what peo­ple here think about that.

One more dis­agree­ment per­haps. I do think that this AlphaGo Zero piece of code is an as­ton­ish­ing ex­am­ple of AI pro­gram­ming, but I have some deep doubts about Wat­son. It was great back then in 2011, but now they seem stuck to me.

• Knowl­edge is in­for­ma­tion er­ror-cor­rected (adapted) to a pur­pose (prob­lem).

No. Knowl­edge is just in­for­ma­tion. If you have some in­for­ma­tion how to solve a par­tic­u­lar prob­lem, it’s still “just in­for­ma­tion”.

There are no hard and fast rules about how er­ror-cor­rected or to what

Those rules are just some in­for­ma­tion, some data. How “fast and hard” are they? When there is a perfect data about the fastest check­ing al­gorithm, then it’s still “just data”.

The field started cod­ing too early and is largely wast­ing its time.

Per­haps. How do you know what peo­ple know and who is cod­ing already, pre­ma­turely or not?

If you joined the field, I would recom­mend you do not code stuff.

I wouldn’t give such an ad­vice to ev­ery­body. I don’t know what some peo­ple might know. Let them code, if they wish to.

Cer­tain philos­o­phy progress is needed be­fore cod­ing.

I agree, that you need some philos­o­phy progress, you don’t know if all oth­ers need that too. At least some may be com­pletely un­known to you or to me.

good non-AGI work (e.g. alpha go zero, wat­son)

Isn’t their cod­ing pre­ma­ture as well?

which they hope will some­how gen­er­al­ize to AGI (it won’t, though some tech­niques may turn out to be use­ful due to be­ing good work and hav­ing reach)

I am not as sure as you are. They hope they will do some­thing, you hope they will not. That’s all.

wast­ing their time

Maybe you are a time waster Mr. Tem­ple, your­self. Your claim that “cod­ing AGI” is pre­ma­ture is just a guess. It’s always pos­si­ble that one is wrong, but say­ing “you peo­ple don’t have the right the­ory, stop cod­ing” … is su­per-wrong. You don’t know that. No­body knows, what some­body else might know already.

peo­ple are su­per fo­cused on pre­dic­tions but not ex­pla­na­tions.

A good pre­dic­tion can only be done if you have a good the­ory/​model about the mechanisms in­volved. So ev­ery de­cent pre­dic­tor mod­els any­way. The best pre­dic­tor pos­si­ble has a cor­rect model. Which doesn’t always im­ply that its pre­dic­tions are right. Some­times there isn’t enough data for that. Even in prin­ci­ple. But to pre­dict is to model!

some even deny there are non-em­piri­cal fields like philos­o­phy

Some are dirty bas­tards also, and some have friends in low places and aunts in Aus­tralia. But you seem to im­ply, that all should share your view about “non-em­piri­cal fields like philos­o­phy”. Yeah, right.

It has been enough. At least my last re­mark I gave, was already un­nec­es­sary.

• I don’t like it, at all.

As you see, my ar­gu­men­ta­tion for that is just like yours. Nonex­is­tent.

# Open thread, Septem­ber 25 - Oc­to­ber 1, 2017

25 Sep 2017 7:36 UTC
0 points
• 22 Sep 2017 19:01 UTC
0 points

There are 143 primes be­tween 100 and 999. We can, there­fore, make 2,924,207 3x3 differ­ent squares with 3 hori­zon­tal primes. 50,621 of them have all three ver­ti­cal num­bers prime. About 1.7%.

There are 1061 primes be­tween 1000 and 9999. We can, there­fore, make 1,267,247,769,841 4x4 differ­ent squares with 4 hori­zon­tal primes. 406,721,511 of them have all four ver­ti­cal num­bers prime. About 0.032%.

I strongly sus­pect that this goes to 0, quite rapidly.

How many Su­dokus can you get with 9 digit primes hori­zon­tally and ver­ti­cally?

Not a sin­gle one. Which is quite ob­vi­ous when you con­sider that you can’t have a 2, 4, 6, or 8 in the bot­tom row. But you have to, to have a Su­doku, by the defi­ni­tion.

It’s a bit analo­gous situ­a­tion here.

• Say, that we have N-1 lines, with N-1 primes. Each N digits. What we now need is an N digit prime num­ber to put it be­low.

Its most sig­nifi­cant digit may be 1, 3, 7 or 9. Other­wise, the left­most ver­ti­cal num­ber wouldn’t be prime. If the sum of all N-1 other right­most digits is X, then:

If X mod 3 = 0, then just 1 and 7 are pos­si­ble, oth­er­wise the left­most ver­ti­cal would be di­visi­ble by 3. If X mod 3 = 1, then 1, 3, 7 and 9 are pos­si­ble. If X mod 3 = 2, then just 3 and 9 are pos­si­ble, oth­er­wise the left­most ver­ti­cal would be di­visi­ble by 3.

The prob­a­bil­ity is (1/​3)*(((1+2+1)/​5))=4/​15 that the first digit fits. (4/​15)^N, that all N digit fit.

Ac­tu­ally, we must con­sider the prob­a­bil­ity of di­visi­bil­ity by 11, which is roughly 111, which fur­ther re­duces 415 per num­ber to 40165. And with 7 … and so on.

For the di­visi­bil­ity with 3, we ren­der out not only one per­mu­ta­tion of N-1 primes but all of them. For the di­visi­bilty with 11, some of them.

It’s quite com­pli­cated.

• In­ter­est­ing line of in­fer­ring… I am quite aware how dense primes are, but that might not be enough.

I have counted all these 4x4 (dec­i­mal) crossprimes. There are 913,425,530 of them if lead­ing ze­ros are al­lowed. But only 406,721,511 with­out lead­ing ze­ros.

if lead­ing ze­ros ARE al­lowed, then there are cer­tainly ar­bi­trary large crossprimes out there. But if lead­ing ze­ros aren’t al­lowed, I am not that sure. I have no proof, of course.

• Well, I said some­thing in line with “peo­ple may need some stuff to live and declar­ing that we should “put peo­ple be­fore that stuff” is a silly way to pre­sent the situ­a­tion”. Maybe not as silly as it’s a dem­a­goguery.

But then I changed my mind and de­cided to not par­ti­ci­pate in a dis­cus­sion at all. But some­how couldn’t erase this now empty box.

• tl;dr

But I saw this:

Time to put hu­mans be­fore busi­ness.

Time to put hu­mans be­fore oxy­gen, too? Silly stuff.

• The bot­tom and the right­most prime can both have only odd digits with­out 5. The prob­a­bil­ity for each prime to fit there is then only (2/​5)^N times that. We can’t see them as in­de­pen­dent ran­dom num­bers.

• Very well. What do you think, are there ar­bi­trary large squares pos­si­ble or not?

I think not. Even in bi­nary no­ta­tions NxN and above, prob­a­bly don’t ex­ist for an N, large enough.

• Con­grat­u­la­tion! It’s es­sen­tial that you don’t tell the al­gorithm, at least for now. You have an ex­tra solu­tion, where ev­ery hori­zon­tal has its equal ver­ti­cal. Which is perfectly okay, but I won­der if that is the prop­erty of your al­gorithm?