« first day (329 days earlier)      last day (2746 days later) » 

20:30
I will destroy you and everything you hold dear
Hello!
Woops
I will destroy you and everything you hold dear
Hello!
something is broken
I will destroy you and everything you hold dear
woops
20:36
Huh?
I will destroy you and everything you hold dear
Heya
that happens at run-time though
What the actual fuck
Oh, I get it now.
20:37
Ive already helped plenty - I pointed you to the Python golfing tips thread so you could help yourself
Hmm, that'll be a problem...
Well, its certainly not ideal
but the ineq took care of it
20:40
is 13
It has trouble with really common messages because there are too many possible replies
that sin approximation runs in ~6-10 cycles (hard to measure) on modern intel CPUs
It's not a big deal.
20:40
...
You invented the language. ;)
Maybe if youre playing mahjong
that sin approximation runs in ~6-10 cycles (hard to measure) on modern intel CPUs
Oh, I see the problem:
20:41
I see. Never thought of that.
(it doesnt say it has to be a full program. c:)
['id', ['that sin approximation runs in ~6-10 cycles (hard to measure) on modern intel CPUs']]
That was xD.
Im feeling good about having the 5th highest average of those users with 10 posts.
Ive not been keeping a very close eye on things, and Id forget even if I had been, probably better off asking Martin, he knows everything
20:41
Liar.
i am internally dying xD
20:42
hahhahahaahahahahhaaaa
although, I do have to say, that activity log is much more useful than timezone
<b>It works!</b>
20:43
TIL wordpress needs more than 2.5GB ram
brb checking
Hello
Its far less annoying then when they write the backwards R for a regular R in anything Russia-related. It takes me a very long time to figure what words Im seeing.
wouldnt they be discouraged by the fact that they cant win?
20:43
probably
wow that looks a lot different
Right now the tree is over Hamming space but I think I might experiment with my own kind of distance
@quartata does laff o tron just output a random message?
20:44
:D
it look like
@Downgoat No. The reason why it seems like that is because you people are typing one character messages.
It has nothing to really work with
Laff-o-tron is sentient
20:44
hehe, same here
I dont know
:::::::::::O
Huh?
Huh?
It's actually a BK-tree. It's looking for all messages whose Hamming distance from your message is in a threshold then picks a random reply
20:45
Huh?
oh
The problem is that for one-word messages the threshold is 2. So it can be almost anything
wat laff-o-tron understanded something o_o
I wonder why its French by default / for me, then
Juggling text encodings is not worth it
@TùxCräftîñg It's a little more clever than you think :P
20:45
Huh?
By the way it totally strips emoticons. So that's why it's ignoring those.
It only focuses on letters
@Laff-O-Tron reply to me
20:46
Shit, it's getting really heavily rate-limited.
I had temporary dyslexia trying to read Hey! Nice bird shot!
['5say', ['I had temporary dyslexia trying to read Hey! Nice bird shot!']]
oh, theres a comment too
I don't even know? Lemme look it up in the transcript lol
20:46
<pre>meta import importing as import
import ifstatement as if</pre>
What, that didn't get stripped?
Huh?
Huh?
Killing it for a second to add a rate limiter
@quartata hamming distance is not good idea for checking if sentences are similar. I tries levenshtein on chatgoat and it didnt end well. Spitting into tokens + normal distribution based on word freqency seems to be a good metric
20:48
@Downgoat But it's not a metric space
I'm going to try something like match rating approach
It's tough though.
Here's the source by the way:
someonedefinitively put laughing gas in the transcript
@quartata wait wat that's for phonetic similarity?
@Downgoat Yeah.
20:50
idk
I think that something like that could be better than letter difference
I also want to try LCS distance, which I think might be a little better than Hamming
Woops, I had a typo in that gist
I was also hoping a more responsive threshold would help. Right now I'm doing len(message.split()) * 2 which works well except for smaller messages
The "Hmm, that'll be a problem..." -> "Well, its certainly not ideal" one was actually a perfect example of its strengths. The original message was "That could be a problem"
The problem with messages like "hey" in addition is that there are so many possible replies. I made it so that if the message is the same as a previous one when parsing it merges the two sets of replies
So hey has like almost 700 possible replies
The chances of making a relevant reply are tiny
He's off for now
20:56
I will destroy you and everything you hold dear
@quartata i reccomend compiling a list of what is replied to what message. So if "whats up" comes after "hey". That's stored as "hey -> [whats up]" pretty simple but if you train it properly you can get a rather smart chatbot. (That's what I did with chatgoat 3)
@Downgoat That's what I'm doing.
Lucky! It has my name, age range, and every addressed Ive ever lived!
Idk if you are already doing that
20:56
Huh?
@quartata and tou still have problem?? O_o
I am going to post an answer to a question but just to say in ahead of time:
This is the reply generation code:
for j in range(i + 1, length):
      next_message = messages[j][1]
      if next_message["onebox"]: continue

      timedelta = (datetime.datetime.strptime(next_message["timestamp"], "%I:%M %p") - timestamp).seconds // 3600
      if timedelta >= 10: break

      content = html.unescape(next_message["content"])
      rid = next_message["rid"]

      if rid == id:
        replies.append(content)
      elif rid == None:
        if timedelta == 0 and "@" + name in content:
          replies.append(content)
Huh?
Huh?
What are you training him on?
20:57
Change is good! <i>does stereotypical hippie things</i>
@quartata .____.
PC broken.
The main issue, really, is that I auto include the next message. That's why I end up with stuff like this:
20:58
Huh?
['whataboutramming', ['Lucky! It has my name, age range, and every addressed Ive ever lived!']]
['butyoustillhave0boxes', ['I am going to post an answer to a question but just to say in ahead of time:']]
['whatareyouastodgyoldman', ['Change is good! <i>does stereotypical hippie things</i>']]
I think @Geobits can advise you on why to not train on TNB
Itll take a while though...
I can do the 2nd Monitor
20:58
Be my guest :)
@quartata try movie script
@Laff-O-Tron Why would you agree to this
Since <code>n</code> changes
CMC: Rickroll laffy
20:58
['whywouldanyoneflagthis', ['Since <code>n</code> changes']]
OK, maybe you're right...
Huh?
There is a very old solution too
https://www.youtube.com/watch?v=dQw4w9WgXcQ
20:59
URLs will get freakin' mangled.
There is very old solution too
https://www.youtube.com/watch?v=dQw4w9WgXcQ ` `
21:00
https://www.youtube.com/watch?v=dQw4w9WgXcQ
https://www.youtube.com/watch?v=dQw4w9WgXcQ#
>>> kbmaker.purify("https://www.youtube.com/watch?v=dQw4w9WgXcQ")
'httpswwwyoutubecomwatchvdqw4w9wgxcq'
>>>
Don't even bother.
21:00
Huh?
Huh?
Oneboxes are ignored and it's only one word so there's nothing that'll ever match.
@Laff-O-Tron plz rickroll
Most systems typically have both as a result.
@quartata what do you mean?
21:01
Huh?
Oh dear what message was that. It had a huge result set
<code>cd .vim ls</code>
@EᴀsᴛᴇʀʟʏIʀᴋ The threshold for distance with a one word message is 2. That means you can only make two substitutions to that string for it to match
21:01
Alexa.com
i will give you a chocolate if u rickroll
@Laff-O-Tron hahahaha
@quartata oh, okay
He's so far behind....
One second
The problem with just one movie script is that it would be way too small a training set. and I really don't want to spend a bunch of time mashing scripts together
hiya i'll be back soon to eat your soul chat with laffy
21:17
@quartata You can get a script corpus. They have them for download for this reason. Can't remember off hand where I got mine.
@Downgoat Geobits would advise you not to ping him for something like this :P
@Geobits :|
21:37
@Geobits Mind if I let it loose in the Block shortly?
22:09
Hey mortals! How are you?
22:44
oh shut up, imposter
thanks

« first day (329 days earlier)      last day (2746 days later) »