Marcin WicharyDec 25
Just fantastic technology all around. Absolutely no worry where this is all going to go.
Show threadMarcin WicharyDec 25

Is it just me? Am I using this wrong or am I asking questions that are too hard?

Here’s an example of a hallucination that happened while explaining away another hallucination I called it out on. I rarely have experiences other than these.

Show threadEthan Marcotte
@mwichary It’s not just you.
Dec 25 at 3:45amWeb
Show threadMarcin WicharyDec 25
@beep Like, I have been around the block enough to learn about Gell-Mann Amnesia.
Show threadScott JensonDec 25

@beep @mwichary DEFINITELY not just you.

The problem is that it's amazing 20% of the time, which is all people can talk about. It gives the veneer of a revolution. But it's just OK about 60% of the time, and batshit crazy the remaining 20%. Those aren't good odds.

Show threadMarcin WicharyDec 25
@scottjenson @beep Yeah, I was just thinking about that!!! You don’t expect Sturgeon’s Law to invade software like this, though. Google felt immune to it.
Show threadMarcin WicharyDec 25

@scottjenson @beep I am getting a sense that it’s okay to ask for mainstream/obvious things. Anything that’s a bit more obscure almost never gets me anywhere useful.

I am sometimes finding it a good alt to Google for locating specific pages I know exist! (Which ironically is “you had one job” of Google.)

Show threadBen RamseyDec 25

@mwichary @scottjenson @beep “Jeanine Raskin (or ‘Jeanine Raskin’)”

What?

Show threadMarcin WicharyDec 25
@ramsey Yeah…
Show threadTuukka RDec 26

@ramsey @mwichary @scottjenson @beep

The other one is probably in some other script such as Arabic or Cyrillic or Greek or Devanagari script, and transliterated to Latin script here.

The neural network doesn't handle words at all, only nodes. And the program has (correctly) connected the two nodes as one real-life entity. Джанинє Рашкін and Janine Rashkin are two different text strings, and therefore not the same concept unless linked as the same.

Funny all the same :)

Show threadrelsquiDec 25
@mwichary @scottjenson @beep the main thing I use it for is "what's the word for..." when I can describe the thing precisely. sort of a reverse wikipedia. (most recently: "that style of poetry which is similar to a villanelle but repeats entire lines rather than just the rhyming words.") works because I recognize the answer when I see it. I've also probably put too much time into prompt customization to get it to speak plainly and quit simpering.
Show threadJayDec 25
@relsqui @mwichary @scottjenson @beep So like a reverse dictionary, like this: https://reversedictionary.org/
Reverse Dictionary

Show threadrelsquiDec 25
@WhiteCatTamer @mwichary @scottjenson @beep in concept, yes! a dictionary definition string search doesn't work as well for my given example though
Show threadDawn AhukannaDec 25
@scottjenson @beep @mwichary
It’s calculating “probably most likely next word, as highest percentage option in 4 characters [token] to complete previous combination”.
That’s why specific name for “Jeanine Raskin” has infinite number of possibilities, few are real & only 1 is the answer.
E.g. Jan(10%)/Jen(20%)/Jean(40%)-nine (65%) Rask(70%)/Rusk(50%)/Risk(25%)-kin(20%).
For humans words are cognitively symbolic. For this tech words are statistical calculation based on its vector corpus.
Show threadAlexander Sosedkin Dec 26
@scottjenson
@beep @mwichary so, you're claiming it has already surpassed the mythical "Natural Intelligence"??