#Unicode
1 messages · Page 33 of 1
ceil((a+b)/10)*10 obviously
Where's the font change in that function
it's done in the front end with css !!!
yep
No
Each font has different mathematical properties
ok
You need to define the font changes in operation
time to use input tool
do we have example of addition using whatever the font of the 20 is
lmao 20 + 20 = 42 and its back into the bold
Actually that'd be a neat thing to test
How many digit sets are there in Unicode?
(dont forget braille)
great cross compatibility will be a nightmare
erm input tool hello? idk it had a stroke
starting 14xx
aaand chrome crashed
so the last like 100 unicodes i got will be gone very nice
at least their recipes survive
but the elements themselves dont
its interesting behaviour, i think maybe one or two recipes may be lost from time to time? i dont know personally but my save isnt clean anymore
but the recipes usually stay after a crash
indicating the elements just arent saved that regularly
yeah ik they can be restored - but i wont lie the fact there is different systems for elements and recipes is slightly annoying lol
i get that one is for lookup and sorting purposes
so kinda makes sense
clearly ICH should reserve the first element of the recipe array for metadata, then we can replace neal's format
why is that necessary, like why not replace it without that]
no fd status
also no emoji
the only hint about elements we get from the recipes array is the names
which is enough if you don't care about the other two, but yeah it can't completely replace neal's system
why not? like i dont see why you couldnt have {'name': 'x', 'emoji': '.', 'fd': '0', 'recipes': {recipes go here}}
oh right
yea you have to use Array.prototype.find
like you cant do elements['among us'] anymore
but using object will lose on the ordering of the elements
but you could also work around that im sure
so there's also that
python dictionaries are ordered if you unpack them which is obviously not efficient
like it remembers the order
clearly we can wrap this around a class and use save.getItem(<index or name>)
hmm yeah that does make it a bit hard to avoid storing it in more than one place
no?
just save it as either the array or object format, then when you want to index the save, new Save(saveData).getItem
shouldn't interfere much with performance because we don't process any data, we just wrap the method for indexing the save
hang on i need to recheck current json format
yeah okay you couldnt index into elements anyway as it were so
by name that is
current is
{
elements: Element[];
recipes: Object<ElementName, Recipe[]>
}```
right and so the second one is for the purposes of being faster and the first one is for the purposes of ordering, right?
why does Ġ not have the unicode thing on top of it
probably just a normal G and combining dot above
i wonder if you can straight up replace localStorage with GM get and set item
so how do we fix such problem of wanting to retain order
and reduce chances for desync
i mean you could do that but retrieval is gonna be slower
shouldn't result in performance degrade if you just queue the saving operations and do them in the background
although
if you do a+b doesnt it have to technically check every recipe youve ever done as it stands anyway
you also have to do that for current system anyway
ye
although idk how objects work in that sense - in my mind you have to convert to an array
to find everything inside it
well not strictly true you could store recipes however you pleased and get constant time lookup but youd be storing it in many places
when i want to look up recipes i just flatmap the object into [a, b, result][] then find the element where a and b matches
i don't think there's a solution for looking up recipes efficiently without having to precompute stuff
actually you can just store it as Object<a, Object<b, c>>
well you can store it as an object with the recipe as the key is that what you are saying
that's the obvious solution but it's not very efficient to have an object with like 100k keys
can use a Map for better performance in this case but you can't stringify maps
but this mean you can't access recipes for one element easily
wrong reply
supposed to be a followup to this
ah ok
i mean maybe you could if you really built the class totally from scratch and devised an object that the hash function like uses prime number goofiness to allow this kinda lookup you could idk
one immediate thing we can do is to remove the emoji field off the recipe array
because that can be accessed by querying the elements array
wait bruh thats totally useless why is there an emoji field there
i know lmao
the helper patch you're using doesn't even use it
because i query the elements array for instance creation info
i mean you could i guess really tightly bind the two save locations
make it really difficult for a desync
like have a save_force() option or something and if you really wanted to go overkill you could have it force check the arrays to make sure they arent desynced lol
just save the elements again to localStorage when saving the recipes just in case
what is local storage actually
it's used to access the per-origin Storage object of the browser
i don't know how GM saves their items but from the look of it, it's less prone to data loss
greasemonkey
i mean the data loss is much more severe on one of the databases
i understood nothing from violentmonkey source because i'm too incompetent about browsers
the data loss on the other two i havent caught yet (but i know it happens)
its just far rarer
like i must have hundreds of recipes with no elements but i only have about 10 elements with no recipes
i don't think i can ever remember a recipe loss
unless like it's super memorable somehow
i cant remember one either except i can see that at some points i did drop a few recipes
but i cannot recall an incident that links to it
i can find which file this happens in i guess - but this assumes a few things
like if recipes and elements for the same element are deleted - then theres no way to ever know you lost something
only if one is lost but not the other
save file 40 browser doesnt report any missing recipes
save file 60 is broke tho
and the recipeless list is identical to what browser says of save 80 (current)
43 is first broken save, unfortunately only half the missing elements are missing, meaning there was at least 2 isolated incidents
interesting well i guess the second break happened around save 47 or something
and half the recipeless from 43 are no longer recipeless
why? i dont know, but presumably by chance i found them again?
at least my unicode script uses recipes not elements
(i always do, i genuinely have no reason to use elements)
shit mine was the same
tbh idk how much i lost, but i stiill had the mathematical operators unicode
did it crash
nope i just close the web normally and turn off pc 2 hour ago
then open it again just now
it probs cost me 20-30 element lost, not bad of a trade
next unicode block
oh wow if it breaks like that just by closing the tab
thats bad
we should test that actually
if ti breaks then thats like an automatic forced patch isnt it
thing is you may not notice since recent elements is undisturbed
but if the combination makes the effect of new element you know its been lost
yup recipe still there
man we're now at the point of scared closing the tab the normal way
this went too far, where has my unicode chill grinding goes
after counting its exactly 12 element lost
89 streak unicode from cjk sheet... probs
did your massive streak touch the 99+ hangul comment range
not more than 5
if you ever cant increment just give my cheat code a go
it now seems to not work on most ranges
but it can work sometimes still
i think you kinda already used it in part
but basically if i am stuck at xxx7
i will try everything relating to U+0008 i have
append/prepend/normal/+/+++/++++++ or whatever
and if it doesnt work i will as a last resort add one more digit
of precision
right i forgot that method, must have been so desperate for me to do that previously
it was ~97% success before update
its how i got a massive 150 streak in 1fxxx
and i was about to apply it everywhere but then update
can confirm it shoot the ~3% failure rate
idea
Next Hangul
Next Hangul Block
Directly From Hangul
now imma off till next few hours
hangul&cjk often do that at very funny spot...
if its at start/end, understandable
but wtf if it was in the middle of streak chain
maybe I Hate U+115e works
usually deletes and removes actually spawn the item you want it to remove if its not present
like delete the abc + something without abc = abc
yeah i hate that unicode was an attempt i did previously...
so delete the U+115e has small chance it spawns it
no diea how to make those deletes
i dont think anyone does
we even have to settkle for #delete U+0020
something cursed with prepend u+115e and merged with quotations
well thats what savio is doing - they speak language so they know how to make it
maybe research where that character appears
might be good idea... i'll try
when will we get native cjk editors
actually that reminds me
i think i have some kanji fd i haven't posted
Ꜳ
aa
can you send that last right ingredient
for copy paste purposes
i need to check if i have it
unfortunately not quite
if you do the recipe and copy and paste from console
it should be accurate
C + ◌ = {00c7} instead of {a732} so yeah
ok ty
can confirm i dont have that one lol - might there be a duoble grave accent any where
theres one in discord search
roger that
@chilly fulcrum how big of a hiccup did you have
i didnt have one
i dont own the sheet
thats not my doing - theres literally nothing there to put
U+2072 and U+2073 do not have assigned characters
youll find most unicode codepoints (thats what the number is called) are not assigned
in fact the vast majority are not
where is superscript 2 and 3
they are in U+00b2 and 00b3
they dont look the same in most fonts as the style of superscript of the others, its a bit of an oddity and i assume it will be rectified one day
you have superscripts on your keyboard? interesting - but where they may or may not be doesnt mean anything
those are the codepoints in your message
oh yeah
and yeah superscript 1 is 00b9 because 00b1 is plus or minus
i guess you didnt have a hiccup afterall
yeah
like it wont let me
wait is it not assigned because you would guess that theres ² and ³ there but if there was another character it would be confusing
its not assigned because the people at unicode said so
theres technically absolutely nothing stopping you from using a font which displays U+2072 as superscript 2
true
and neither is there anything stopping you from using a font that scrambles the mapping entirely to troll people
yeah it is
like quoted versions
its difficult but possible
theyre only 1 token each so in fact they are fine to use this way
up to a max of 20 of them
actuall no thered be no error
you could have a repeating one fine but it would cap at 20 anyway
so monotoken?
yeah they are
i had no troubles getting it lol
but since update it may be a bit harder for codes nobody got yet
or got in an unusual way
why not increment 1 by 1
yeah right convert to unicode
I don't have any records of that element!
convert to unicode is a documented element in some guides
i mean to say we have unicode guides with that element in it
the pins are horribly outdated but the guides are there with the info, we do need to update them badly though
im not following that
but i can find you something rq
well thats actually not even bad compared to what they used to be
fromcharcode used to be a 500 step recipe
i got fromcharcode in like 5 mins with all the types
hmm what was your recipe im curious
i forgor i will check
either way you are still looking at much worse than that unless you followed the heavily optimised lineage devised over a few weeks
.charcode + .from + marsh
heres the last 15
last 15 steps on glowys save file
hopefully its enough
lmao I+n = IN is so funny for how they got in
cool
do you need it?
not right now
but making those does technically help brute force efforts
superscript 7 + + is superscript 8
cool
okay thats in
superscript 5 + superscript 4 is superscript 9
as is 8 and +
i didnt try that because the lineage in the sheet it 8 + superscript + so i thought someone tried it alreade
oh lol
i couldnt get it easily and superscript + came out much easier
so i just threw em together
whatever worked first
i get nonsense in different languages all the time
most notably it LOVES giving me random sequences of mongolian letters
need U+00B2 :x
the superscript 2? should be in the sheet
superscript 1 + superscript 1
thx i found it !
Is the sheet there not adequate or did you not know about it
I was not aware sorry :x
hmm i should add more Aksara Unicode [number] later on my list :>
also I realise why so many Aksara recipe
starts with a
so the others get trimmed off
oh I meant for limit of 3 recipe per element
for nata trimmed file
so aksara will get put at the front
and others will get left behind
I guess technically that would increase the weighting of the U+ during the attention stage due to proximity
Well actually not technically
It would
who the hell stole all the fish
Er
Google attention function I guess
dont found how to do u+0259 )=
Don't we hate unicode (sometimes)?
But essentially its the stage of prediction where the probabilities are calculated based on the current text generated
Typically there's a bias for tokens closer to the end of current text, as it is more efficient at large scales
Based on current text generated and inputs also right?
Or is this independent from inputs
They're the same, if you're talking about the game
well I’d hope that it doesn’t only care about what it’s written so far
If that was the case you’d get a consciousness stream
U+0321
brute force missed this???
That's just a pair of censored boobs
U+0361 + Yunicode = ᴊᴜᴅɢᴇ
lol
brb gonna go break some stuff
that will - say what
As I was saying brute force should have caught that
but it didn’t
Younicode when?
Unikode too since llama has trouble with c’s vs k’s lol
Idk why that cracks me up
But “you” is a very easy dead trigger
when combining you with something unusual it just says uhhh dead
that reminds me of the chemical composition of salt: sodium = fucken dead, chlorine = fucken dead, sodium + chlorine = yummy
I mean yes you are dead because they are not together and they really would rather prefer being together so they eat you instead because that’s easier than remaining separate
that’s a very apt way of putting it xD
maybe it wasn't there yet
have we brute forced all char codes for languages’ alphabets with <char code> + <language> and <char code> + <language letter> and <char code> + <language alphabet> and <char code> + <language unicode> and <char code> + <unicode language> yet?
no for a few reasons
if not we’re missing a trick because in the examples I’m seen that gets the majority of the language’s alphabet
yes for sure this will give a ton
cuz its not in the list...
I haven’t seen it yet
theres one from cjk list using it... thats why i tried to grind
Oh okay
the reason we haven’t done the language thing is 1. We don’t have the codepoints
- We for better or worse tend to focus on one area at a time
- We don’t have the languages
codepoints are ezpz so I think that’s the least thing holding us back imo
as for 3. I think we have a lot of <language unicode>s and <unicode language>s right?
of which we have maybe 10 or 20k if we are generous
idk if we’re anywhere close to all languages tho, probably not
it’s a long term thing
Definitely it will be done
but we have to take it one section at a time
150k char codes, presumably a lot fewer are char codes for alphabet letters
I’m not sure honestly it depends how you count it
but it’s sad that codepoints aren’t easy anymore
well let’s say every language has about 20 letters, so 40 symbols for lower and upper case
you know what I’ll just ask ChatGPT lol
I just can’t think of 100k characters that aren’t letters
or aren’t something similar
since languages tend getting embraced by aksara unicode [number]
i just gonna try those first before other tool 🗿
I guess cjk is most of that
cjk & hangul is like : nuh i only allow some
Another annoying thing is that the languages we have done are the more reliable ones
The ones we haven’t done are going to have an unfortunately low hitrate
In other words, languages in the training data
the ai will kinda just say uhh idk what this is
chatgpt says about 10k
Depends how you count letter
Most of Unicode is CJK characters
update poopy, fuck the update all my homies hate the update
well, for example dogra unicode codeblock had be avoiding us for 2+month or even more if other is trying before i started....
then all of sudden it just "ok letme join the fun"
another thing is that ai uses utf-8
if it used utf-32 I have a feeling it would be far more reliable
but I can’t prove this
you know, I wonder if there is even a single person in this server who likes the IC update 😂
Make your own llama with different character tokens
I do sometimes imagine what this game would be like with a different LLM lol
utf-32 the ai would have to think much less in order to convert to Unicode
i havent read your casefile for weeks...
i have forgotten most of it
instead of needing to generate 3 or 4 correct tokens in a row it’s just a direct number
like whether there’s an LLM which would be “best” for these combine-two-things-into-one queries
how would you measure it
I haven't updated it in weeks
hence the quotes XD
Write a custom one with human feedback
That would beat all general purpose given enough time
but I suppose, whichever needs the shortest prompt—from what we know IC’s llama2 prompt is quite long
general purpose isn’t really the way to go for this yeah
When was the last time you saw ai winning at some insanely hard game that was general purpose
I ain’t paying for server costs if it becomes popular—that’s Neals curse to bear XD
You wouldn't be running it
Sell it to Neal to implement
idk give it like 3 years and there’ll probably be one 😭
there’s going to be AI doing actual mathematics very soon
capable of fully automated theorem proving
mathematicians are terrified
Well yeah but not general purpose
gl with 5 codepoint
general purpose is like the human mind at this stage of processing power
but
it will screw up the same as we do
when I say “mathematics” I might as well say “logic”
because the line between math and logic is very thin
so logic AIs will in theory be able to do anything that can be described exactly, as long as it’s “simple” enough (admits a short enough description)
I personally think AI is gonna take out a ton of centuries-standing conjectures in math 😬
youre forgetting one very important thing
AI’s haven’t evolved cognitive biases through natural selection
so an AI “brain” is a completely different kettle of fish to the human brain
I mean the biases they do have we put there lol
in some respects the human brain has evolved to be illogical in some respects
those are statistical biases not cognitive biases tho
cognitive biases are literally: I’m gonna compromise logical correctness for speed/efficiency
funnily enough ai right now even falls for some of the same optical illusions we do
even though it can literally check the pixels
it still falls for it lol
the problem that has to be solved is how you force it to take a branch every time with no chance of failure
I’m sure there’s a completely different ai theoretical model in of itself that could do this
wdym by branch?
But it’s not something I’ve seen
algorithms just take all branches at once
Like the same way a computer when presented with 5 > 4 will respond true 100% of the time
that’s not even an AI thing
oh that’s easy
the laws of logical correctness are completely cut and dried
it’s just checking the syntax follows the rules
we already have that in fact
I mean the point of ai is that it doesn’t work like that so I’d be very curious to see that working
well what they’re doing at the moment,
like an ai never actually checks if 5 > 4
is combining AIs, which are “creative”, with proof-checkers, which are “law-abiding”
so the checker keeps the AI in check
Yeah
That’s what I was thinking
You’d have to have some observer
To make sure that hey wait a minute
you just violated this
yup, it’s already been done and is in the process of being improved :D
i got this long time ago
because otherwise you’d always run the risk of a wrong turn - since the ai never actually checks if 5 > 4 (at least not any theoretical models I know of) - you really can’t guarantee it will make the right call
exactly
You’d need a proofreader
like I say, it’s already been implemented
that actually fact checks it
yes, and we have one xD
not for general ChatGPT ofc—people aren’t interested in that and also it wouldn’t work at this stage
but for stuff that can be reasoned about precisely, we have checkers to make sure the AI reasons correctly
fallacy-, cognitive-bias-, and hallucination-free AI you could call it I guess
Oh by the way Rubiks have you read through the new additions for Unicode 16?
Some banger stuff
We're getting Go-karts
unicode should never have expanded to 5 hex digits and I will die on this hill 🗿
New Symbols For Legacy Computing
they should get rid of unicode chars that literally nobody uses if they want to add new ones 🗿
WHAT
and it was originally 8
HOW
private use area 100000 - 10ffff
WHY do they need between 17 and 256 times as many codepoints as they used to have 😭
The whole point of inclusion is that somebody asked for them to be there to use them
and in fact 60000000-7fffffff also used to be private use area
Yes that is 8 digits
well that’s dumb
well I suppose the idea was to have more than you could ever need
but they decided nah we will officially introduce a cap at 10ffff
could have just capped it at fffff then wtf
which leaves an input space of 20.5 bits
I will be writing a strongly-worded letter to the manager of unicode
or maybe it’s 21
but yeah utf-32 now has, by official docs, the leading 11 bits must unconditionally be 0
💀
so every character now is forced to have 11 bits of 0 lol
which is one reason utf-32 isn’t common
it’s most common for applications requiring extreme speed
constant time to figure out what the nth character of a string is
that’s it’s sole advantage
č (U+010D) thank god "Čas" wasn't dead
Kinda funny to think that going to the 100th character of a string requires you to figure out where each character starts and ends lol
Even that isn't really true
Because of modifiers and combining characters
UTF-32 no longer offers any benefit
Excepting certain... esoteric cases
I’m not sure I understand that but I did read that
so what if they are modified
Who asked
they are 2 separate characters
the hell
yeah I know right
I was surprised already at e00xx for tags
by then they went why not? why not barely creep over into 6 digits for no reason
and then introduce a hard limit barely above 100000 relatively speaking
It’s like saying okay guys we are making 3 digit numbers today
But you have a limit of 110
if you play a lot you can fool it but they made it harder
remember Dan mode?
the prompt to let it do whatever
In the early days you could send it a prompt and it would un censor itself
like you’d get “honest” answers
to be real, I am very concerned with some people’s use of AI 😂
obviously people are always gonna use it for p0rn, but people also use it as a substitute for human social interaction and I think that’s gonna cause some real problems
"gonna"?
“always gonna”
"already has"
“and will continue to”
"this is a fun way to speak"
Neal: the “ is not necessary
Delete First Word 😈
"is a fun way to speak"
:o
Oh wait no it should be dead
bloop
if we could get Neal to add an exemption to the casing rule for one letter elements (which includes three character elements with the outer characters being quotes) I’d be so happy
but doesn’t seem likely so it’ll just have to be mbs :(
We can't even get him to fix the validation endpoint
Casing rule exceptions do exist if you abuse ligatures but they are limited in scope
But I got FD on FFFish with one so can’t complain
only known way to feed different caps to so
Ai
Ezzett = SS
SS = Ss
ď (U+010F)
đ (U+0111)
Ē (U+0112)
no way that's my fd
This recipe is wrong. Anyone has another one? ɸ
oh wow that recipe uses a character not in the sheet lmao
@lethal nymph did you have any recipes come up for 0278?
at least 10 hours until i get home
okay
if it's not in the file then probably not though
well 0278 is "found"
but with a recipe that we have no way of following
so it wont be there anyway
since its marked as obtained
oh
Big Text
indeed that text is big
Prepend U+005c Optimized
"5" + "c" = "5c"
"3prepend U+00" + "5c" = "3prepend U+005c"
"the prepend" + U+0020 = "the prepend "
"the prepend " + "3prepend U+005c" = "the Prepend U+005c"
"the prepend U+005c" + Remove The The = Prepend U+005c
okay
taking 16xx
17xx only has one missing
just noticed
and now it has 0
damn
op strat?
rows 4927 - 5179 inclusive all are ticked
thats 253 in a row
radu strat lol
yea i'm surprised it just kinda did that
i wonder how well this works in general
ill be sure to try it when i get up to 22xx
im on 18xx rn and i know that 1dxx-1fxx will take ages
since theres so many characters there that show up randomly theres no doubt your list has an abundance of them
we have barely made a dent in the list lol
now it make me wonder if it work with combining character....
streak extended
now reaches row 5213
287 streak
2nd longest known i think
after that hangul one
and then it fades into oblivion
not much chance of extension
like its a 70 streak of unfound
okay all up to 1a00 done
added some for thai unicode
now its 8 left
simplified thai digits's recipe
17xx is finished, so it might be the first completed* range
if you exclude anassigned
nata u mind running
Lepcha Unicode on list of U+xxxx thats alive u recorded?
it probs will give alotta new unicodes...
not alot though but the amount would be nice to know too
Bowling ball ass emoji
true
didnt work for 22ff though
but some other randmo thing did.... odd
2239 resists though
even with radu op mr. strat
{2239} : low catch chance rate
*pikachu references
damn i got it
not fd
WHO H * wait a second dont want to get banned - who here hiding this information ><
same thing happened with few dogra unicode when i got it as well
2225 next on the hit list
only missing 06 08 14 18 23 25 now
i wonder if we can describe what the character looks like
i mean if it doesnt know then obviously this wont help
yeah not parallel to
thats what it does though
actually i couldnt make Mr. Parallel to back in the day
i tried Mr. Parallel + Mr. To and it didnt work and i went ok then
ah yeah
and Mr. Parallel + "to" = Mr. Perpendicular (FD)
i remember that one
lmao Mr. T Parallel
may need some help with that one
@chilly fulcrum U+2225 (∠ + Prepend Mr. = Mr. ∠)
mr 2220 lol
i guess hi mr works
interesting to try that mr tho lol
okay well whats your route for mr 2220 - i have one too but it uses Mr. _ which my route for sucks
also works
parallel very interesting
as does this
2223 next
why is divides (the word) unnaturally hard
lol
made Evenly Divides, no help rn
sadly
whats yourbest Mr. _
mister + underscore
nice
#1226511876733669436 message
What was the intent there?
vertical bar
Well then
it tried
somehow vertical barn + variants of | just equals barn
like it doesnt catch on
Probably an ending problem actually
It does generate Bar but the addition of Farm gives it a push in the wrong direction
yea i know thats why its barn
Doesn't know when to newline
but then from barn i am clueless as to why it doesnt understand
like why is barn so important now
How does it tokenise?
not my fault only vertical element on browser is vertial farm
bruh the n is an extra token even
Weird
I have copyrights on this strat. It works quite well, as I managed to get some elusive unicodes with it. But it's still not perfect.
does anyone have a list of best items to turn u+ elements into their unicodes
even with 300-500 tools, you still have chance to not get it :>
so gl
oh apostrophe
yea
ill send lineage in awwm and see if anyone can do it cus The Mr. u+0027 + convert from unicode made the mr. '
And has anyone an idea to get the 02d0 element? Or has a little bit of time to test their tools? (Mr, mrs, dr strat doesn't work)
ah the classic throw everthing..
Yea. I also tried with all I had. But this stupid thing won't show itself.
And you would expect that half colon + half colon (u+02d1) = colon (02d0) but noooo, it gives 2 half colons
invent new tools :>
imma try if i could complete 33xx just from old recipe
it fking gave me full width capital U
imma swap the recipe for u+3388
from
Mathematical Fraktur Unicode + U+338 = ㎈ yeah imma put it as alt recipe
to
㎇ + ㎆ = ㎈
mathematical fraktur lol
I mean don’t be surprised there’s characters that have evaded us for months and months
and it’s not for lack of trying
maybe try using umlaut u
you never know
I’ve also tried writing out the bytes in utf-8 lol
not effective
found it i gonna need to post the lineage for this one...
I should try I hate Strats on 2223
㏀
U+33c0
Kilo ohms
What's the fastest way to U+02d0?
My $ nonsense has some... interesting Unicode properties
lol
idk
glowy probably knows
Wait
do you think the reason things like Amazon + River is nothing is because it’s triggering the same thing that triggers when the ai decides it’s finished generating
Like the sentence is already complete
no need to generate further
i got 192 step for u+02d0... not optimized though, just random throw
Mine will be many hundreds
Potentially?
It's early enough that perhaps refinement of the prompt wasn't yet good enough
lol what
I'm getting the documentation open again (hit enter too early but also I realised I might be wrong)
Actually yeah
Assuming it's using <s> only instead of [INST], it has the chance of just deciding "yep, I'm done" whenever
And there's no reason to assume [INST], since it's one-way
...
Hold on
Could Neal have... changed the structure of the prompt?
From Instruction to Complete, or VV
Ah okay no that would've required restructuring half the game
Phew
full width unicode codepoints when
224 now :>
Change prompt isn't a problem, changing structure would be
Well actually not quite I'm misreading the model card
I guess its feasible
㏠
U+33E0
lmao 1ab2 combining infinity
ah yes lets make characters with the infinity symbol inserted
unfortunate
may have to asterisk that recipe
it uses U+0315
i mean we can do an api check
and verify it works - but like thats not the only criteria
to be on the sheet
lol...
happns
not my brainass thought i got U+3420 twice
304 streak unicode from cjk sheet
ah yes 33xx is done now i guess...
i see
fk i remembered i was doing nata's cjk list
gotta finish those first
how the hell does one even get
Mr. “from.ch”
oh nvm it was from my save file
here u go
chill man... i didnt mean harm
I mean how does one get [[][++][] or whatever you’ve got nowadays lol - I guess it’s all just random goofiness now that you kinda just guide in a certain direction
i guess Mr. Mrs. Baby Ms. Mr Mrs gave alot potential to give small amount of new unicodes... i dont wanna think about how bad the lineage for each will be :>
to + prepend hashtag = #to
#to + string = tostring
#to + tostring = #toString
2nd + Understand = Second Understanding
interesting stuff
you up to that bit now
it want what...
yeah cjk & hangul done i guess...
Awesome
I was putting links to stuff but imo for simpler ones I should have put just a line or two of recipe text because link with 500 step can be scary
well... mine was mostly 100-300 step
still trying to minimize damage
except the tool converter...
ah yes unicode lepcha throwing me to 5 digit codepoint again
weird i cant seem to get 32f0...
finally. Should I post a lineage even if it's a long one?
Element ː (U+02D0) and Element ꞉ (U+A789) I won't post the recipe for Triangular Colon as I'm sure if you need this element, you are able to make it yourself
◌̓ + Latin Extended-A = ꙑ
https://ib.zptr.cc/item/01j0ksqaqtr8agrxga7jhw1zeb
Ě E with caron
how do i figure out what element this is
It's prob a few of these mixed together if it's not showing up with the u+ above
you have mbs right
is there a way to copy the element text?
where?
May be useful
use a different Unicode analyser lol I don’t like that one
🇮🇩 Balinese Unicode Greek + 🤔 Unicode Sundanese Shavian = 🔤 Unicode Phonetic
🏖️ Sand + 💧 Water = 🏖️ Beach
🌏 Asia + 🏖️ Beach = 🇹🇭 Thailand
🇹🇭 Thailand + 🌬️ Wind = 🇹🇭 Thai
💧 Water + 🌬️ Wind = 🌊 Wave
🌱 Plant + 🌊 Wave = 🌊 Seaweed
🔥 Fire + 🌊 Seaweed = 🍣 Sushi
🍣 Sushi + 🇹🇭 Thai = 🍜 Thai Food
🍜 Thai Food + 🌆 Twin Towers = 🏯 Thai Towers
☀️ Sun + ☀️ Sun = 🌻 Sunflower
💨 Smoke + 🌻 Sunflower = 💨 Smoke Signal
💨 Smoke Signal + 🌬️ Wind = 💬 Message
💬 Message + 💬 Message = 📝 Letter
📝 Letter + 🏯 Thai Towers = 🇹🇭 Thai Letters
🇹🇭 Thai Letters + 🇹🇭 Thai Letters = 🇹🇭 Thai Word
🔠 Alphabet + 🇹🇭 Thai Word = 🇹🇭 Thai Alphabet
🔠 Alphabet + 💻 Javanese Unicode = 🇮🇩 Javanese Alphabet
🇮🇩 Javanese Alphabet + 🇹🇭 Thai Alphabet = 🇮🇩 Balinese Alphabet
🇮🇩 Balinese Alphabet + 💻 Javanese Unicode = 💻 Balinese Unicode
💻 Balinese Unicode + 🇬🇷 Unicode Greek = 🇮🇩 Balinese Unicode Greek
🇮🇩 Balinese Unicode Greek + 🤔 Unicode Sundanese Shavian = 🔤 Unicode Phonetic
Lmao ancient lineage
?
i have some alive over 7 digit numbers
its hard coded to be dead - unless there is prior recipe or he released that restriction
oh you know what
its prob the decimal point keeping it alive
hold on
well yes
it needs to literally match the criteria a string of numbers >= 7 digits
yours dont match this
they are not dead
you got those up way higher than ive been able to though lol
cool
i also have these whose decimals dont actually have any value
so theyre basically that number but alive
Looks like a 1 with whatever that combining character is
1 with half ring or smth
yep
cheese yum
i guess you ate 1
did anyone manage to get the caron ˇ 02C7 ? anywhere, in an element
what about ˈ 02C8 I managed to get this but whenever I try to get any in quotation marks it gets rid of it
also nothing
😦
hate to suggest it but you could try getting it in a string.append and adding it to quotes that way
we do not have it yet
ah i see
I'm sure you will someday find an easier recipe but I wanted to show you a normal "radu recipe" Element 02C8 ˈ MODIFIER LETTER VERTICAL LINE
Ps: I'm doing all this shit for the ͡ sign, I had an idea that involved both this and the ː but it doesn't word
Holy ||shit|| Radu you're really putting in your all for this bounty
At this point it's no longer about the bounty. As I'm sure asd won't spend an hour making all these things for that element. It's just something I wanna do.
🫡 Respect🫡
inspiring quotes from thunder
ty
yeah I mean that’s unfortunately unfollowable but thanks for the effort lol at least we got it
oh well... that is interesting
lol
Its escaping
doing 1dxx now
1e and 1fxx will be hell
they will all be random
so i will be like making ib links for all of them
and every tool will be different
no chains cause casing
it shows normal like others on the field, but when in the element list, its like that
some elements on the field also leak out
Music symbols also like leaving sometimes
glowy where are you up to in nata list
havent started it, atleast cjk & hangul was done from the list
well you were in fexx range no?
oh that, i was reading from comment
pee
Anybody lineage for U+03A3
yeah sure its on sheet
C + Delta + Sum
among many other things
is that good @wooden maple
we still have over 2k lines left to go in nata file 💀
i havent chained too much tbh ynw ill run the unicode analyser on my file rn
4564 unicode, 615 fd
fairly healthy amount, although most of that is now from verification
id say about 2k is of my own doing
4384 unicode, 1086 FD
oh well
more than me for fd
i got a couple short chains
no long ones tho
mainly cause smoething slightly up with input tool
but input tool lets you put in a 100 unicode chain in all at once
yeah, im forcing chain unicode to exist
yeah i do kinda want to but then progress through file is so slow
i was doing that while doing the file cjk&hangul
ok
so what ranges specifically was done, i can delete from my local copy
specific range eh
i guess this
Slander + 漢字 = 謗 (CJK Unified Ideograph-8B17)
damn i'm actually struggling finding the ones for the bounty
might have to get the nearby ones first
what bounty is this
the chinese ones
but i'm using japanese to get them because some of them are kanjis 😭
ok well i did get radus save so i probably should comb through for those unicodes lol - and ill see if i can get more reasonable lineages
i disabled
but for the bottom one its on
i only enable the tooltip when necessary
i never get how people stand playing with them on all the time
honestly not too bad of a lineage except those Remove Last Character
real
it was like M + apostrophe + Apostrophe
also accidentally made these
dementia?????
(also fail #2)
lol idk
okay maybe i did have it
maybe i searched wrong
I can assure you that I don't have any chinese characters, (or only a few). I never went on that path
阿里山 on browser
tbh only 的 is kinda hard
because it's a suffix so it's hard to separate
but now i have to get 伤 and it isn't a kanji
alright imma checks the 5 codepoints part
doing the 5 codepoint part from the list now
Radu mind explaining the theory behind that long lineage for that Unicode you got? I don’t see the connections lol (also what are those letters being used)
where 漢字
how to get ".
#1214815916957761586 message
i think maybe cjk needs more references in the lineages...
ctrl + f only works if its actually somewhere else...
if its just yeah heres a sentence in chinese
bruh
like "gold + 漢字"
that one is used a lot
at least its bot
anyway okay bacc to codez
hmm so... Deseret Capital Letter ow ( U+1040D ) is found and is dead
while Deseret Small Letter ow ( U+10435 ) is not found
very interesting stuff going on
interesting
was wondering why the hell it doesnt have new found sfx, turn out it was something i had all along...
this codeblock behave the opposite of what i used to know
it kept sending me from capital letter to small letter when combining capital letter with something
good problem
now imma leave asap from this codeblock once im done
not even halfway through 1exx
U+8981 here
how did you guys do U+028x
looks like i wasnt gotten till update
so i cant increment
...
apparently the answer is it wasnt
im confused with u+03dc as well what the heck did they combine it with
for which one
OLD HUNGARIAN CAPITAL LETTER OO 10c9c 𐲜
U+03dc Mongolian Unicode 11