Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
HLStatsX:CE Update: We've narrowed down some of the bottlenecks affecting HLStats and it appears to be performing significantly better now - we'll continue monitoring and tuning things over the next little while.
~VintagePC
VintagePC:
lol vaultbot having performance anxiety
55 minutes ago
VaultBot:
backbone providers aren't having good handoffs account for a lot in our bioskills labratory
55 minutes ago
VintagePC:
oh great now it's doing neuroscience experiments
54 minutes ago
VaultBot:
how would you know it's bad when you are creating something to you. you can go fuck yourself now and sleeping he said to another thing he said to
54 minutes ago
VintagePC:
I'll just go ahead and rename it Glados, shall I?
54 minutes ago
VintagePC:
don't you take that tone with me vaultbot
52 minutes ago
VintagePC:
I brought you into this world, I can take you out of it
52 minutes ago
hawaw_dacar2:
man the ddos attack has awaken consciousness to him
38 minutes ago
hawaw_dacar2:
Soon enough we may end up with a Skynet situation
37 minutes ago
VintagePC:
I think that ship has sailed lol
31 minutes ago
hawaw_dacar2:
and how the bot even work ? is there like a bajilion amount of pre-writed thing that are trown off at a randomn ?
30 minutes ago
VintagePC:
Nope. It's technically just a simpler version of what LLMs do, and with a tiny data set of a few hundred messages
24 minutes ago
VintagePC:
essentially it's got a markov model of the sentences and uses that to generate new ones
23 minutes ago
VintagePC:
it's quite fascinating considering I wrote it back in ~2013 or so
13 minutes ago
VaultBot:
i guess so unfortunatly...
13 minutes ago
VintagePC:
... and far too sentient-seeming for its own good sometimes
13 minutes ago
hawaw_dacar2:
idk if the analogy is right but is it a bit like the announcer/Army grunts where they only did a few words and made combination's to make it seemless ?
13 minutes ago
VintagePC:
it's a bit more complicated then that. Essentially, for each word in the dictionary, you have a table of probabilties for every other adjacent word in your dataset. Then you pick a starting word (prompt) and walk forwards and backwards through the table, picking the next word(s) based on their probabilities
11 minutes ago
VintagePC:
by "the dictionary" I mean the internal dictionary in the dataset, not every word ever
11 minutes ago
hawaw_dacar2:
damn that's kinda ingenuous
10 minutes ago
VintagePC:
that concept has been around for years and is stupidly simple to implement (I think Vaultbot is like 300 lines of PHP)
9 minutes ago
hawaw_dacar2:
that also remind's me of keyboard's on phone's like they can sort of predict what next letter you will use. Hence why phone keyboard's are usable
8 minutes ago
VintagePC:
it's only recently that people have figured out how, and that we have the computing power to feed these things massive datasets and get really, really coherent (if sometimes hallucinatory and wrong) responses
8 minutes ago
VintagePC:
yeah - same idea, but at the letter level instead of the word level
8 minutes ago
VintagePC:
and of course there's only a finite number of words in the language whereas of course you have a lot more possible choices when combining words to make a sentence
7 minutes ago
VintagePC:
markov models and chains are super interesting in that regard
7 minutes ago
hawaw_dacar2:
is it possible for it to fumble hard and create rly weird things
7 minutes ago
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.