Post AQLAmORP53NIDC7Gee by RyunoKi@layer8.space
(DIR) More posts by RyunoKi@layer8.space
(DIR) Post #AQL4uRdEVA6ohR6amm by killyourfm@layer8.space
2022-12-06T19:00:34Z
0 likes, 0 repeats
It's slightly blowing my mind that #ChatGPT explained how to use the Past Perfect tense -- and then used it in 4 example sentences. Sentences that were nowhere to be found in existing articles online.
(DIR) Post #AQL8kVDMj3YEjSWLfU by thinkStory@mastodon.green
2022-12-06T19:44:02Z
0 likes, 0 repeats
@killyourfm Sounds fascinating! Iām probably dropping in in the middle of a conversation, but I would love to know more about this. Where can I read about it? Grammatical things are my jam.
(DIR) Post #AQLA6fwTca52KPl3QG by trezzer@social.linux.pizza
2022-12-06T19:59:14Z
0 likes, 0 repeats
@killyourfm With a dictionary source past perfect tense is pretty easy to generate. Past participle is generally always listed in dictionaries for a given word, and beyond that you just need the correct form of to have in the past tense (also super simple as it is regular). I'm not saying it's not doing something smarter than that, but it could literally also be that simple.
(DIR) Post #AQLAf9wkZWf3EOKSg4 by killyourfm@layer8.space
2022-12-06T20:05:30Z
0 likes, 0 repeats
@trezzer Seems like sound logic to me. Admittedly this stuff (on a coding level) is way out of my wheelhouse. To me it feels like magic. Albeit very dark magic.
(DIR) Post #AQLAmORP53NIDC7Gee by RyunoKi@layer8.space
2022-12-06T20:06:49Z
0 likes, 0 repeats
@killyourfmIt's math. Probability and statistics.Have you touch points with that?@trezzer
(DIR) Post #AQLApCCX2orn7SUqdE by trezzer@social.linux.pizza
2022-12-06T20:07:19Z
0 likes, 0 repeats
@killyourfm It's wild what it CAN do, for sure. I just asked it to write an essay on the book I've had my class read (from 2020) ā with a specific focus ā and it just spat it out in seconds. Sure, it doesn't do page/line references and its introductory paragraph isn't perfect, but it all made sense.
(DIR) Post #AQMZKEzdJ2ZsoUl8qW by Cuperino@hostux.social
2022-12-07T12:16:33Z
0 likes, 0 repeats
@killyourfm We could argue that that information was successfully encoded into the model, but going from there to calling it understanding is an overstatement, because all GPT models are capable of doing is predicting which words go before or after a sequence of given words.