Miscellaneous News

BoraTas

BoraTas

Major
Registered Member
sheogorath said:
The bot got really argumentative, though which I thought it was interesting.
Click to expand...
That is the usual behavior of nlp models. There are a lot of secondary pieces of software running in chatbots to prevent them from getting really aggressive and the conversation is usually resetted after 5 mins or so.
Please, Log in or Register to view URLs content!

The exact reason is not fully known but they lack the capability to annotate data as truthful or not and their source is existing texts on the internet. Thus they likely just start using arguments humans use when you present them word patterns humans use while arguing.
 
sheogorath

sheogorath

Colonel
Registered Member
solarz said:
Ok, now ask it about Abe Lincoln's assassination.
Click to expand...
I asked it about Chrystia Freeland's grandpa and pulled a Snopes, "yes, her grandpa's paper collaborated with the nazis but we don't know if he liked nazis"

nazi3.JPG

Could be worse, I guess.

Also the "some" in evidence of collaborationism by the paper is doing quite some heavy lifting

Please, Log in or Register to view URLs content!
 
sheogorath

sheogorath

Colonel
Registered Member
BoraTas said:
The exact reason is not fully known but they lack the capability to annotate data as truthful or not and their source is existing texts on the internet. Thus they likely just start using arguments humans use when you present them word patterns humans use while arguing.
Click to expand...

I think Lex Fridman went to Joe Rogan's show and basically said they don't know why ChatGPT works?
 
X

xypher

Senior Member
Registered Member
BoraTas said:
That is the usual behavior of nlp models. There are a lot of secondary pieces of software running in chatbots to prevent them from getting really aggressive and the conversation is usually resetted after 5 mins or so.
Please, Log in or Register to view URLs content!

The exact reason is not fully known but they lack the capability to annotate data as truthful or not and their source is existing texts on the internet. Thus they likely just start using arguments humans use when you present them word patterns humans use while arguing.
Click to expand...
It is not usual behaviour of NLP models, ChatGPT (at least the GPT 3 and 3.5 based ones which I tested) did not get that unhinged. It is Microsoft Bing's thing and it is unclear why that happened. However, it is important to note that Bing overall was much worse than ChatGPT when it started - way more misinformation, failing at some tasks that ChatGPT solved easily, etc. Most likely the GPT-4 behind Bing's bot had a training dataset of worse quality than the one used for OpenAI models.
 
X

xypher

Senior Member
Registered Member
sheogorath said:
I think Lex Fridman went to Joe Rogan's show and basically said they don't know why ChatGPT works?
Click to expand...
That is true for any complex neural network - be it a LLM, CV models or something else. Neural networks are essentially black box models which is one of the reasons why e.g. banks still prefer classic interpretable ML algorithms like regression or decision trees for stuff like credit scoring since they need to actually explain their decision-making process.
 
4

4channer

New Member
Registered Member
Phead128 said:
ChatGPT is trained on data before 2021, so you can't ask it questions for 2022 events or current events.
Click to expand...
I can ask such questions, apparently. I also can link to news articles. However, ChatGPT replied clearly, regarding Shinzo Abe assassination: "In fact, a quick search online can confirm that he is still alive in the public sphere, including giving speeches and attending events, as recently as March 2023".
Somehow, ChatGPT is aware that Shinzo Abe is alive and giving speeches even in March 2023. How so? ;)
 
BlackWindMnt

BlackWindMnt

Major
Registered Member
4channer said:
I can ask such questions, apparently. I also can link to news articles. However, ChatGPT replied clearly, regarding Shinzo Abe assassination: "In fact, a quick search online can confirm that he is still alive in the public sphere, including giving speeches and attending events, as recently as March 2023".
Somehow, ChatGPT is aware that Shinzo Abe is alive and giving speeches even in March 2023. How so? ;)
Click to expand...
Well shinzo Abe still lives forth as a machine spirit, that is probably why Japan is opting to be ruled by machine.
 
plawolf

plawolf

Lieutenant General
Temstar said:
Taken by F-16 with reconnaissance pod apparently. Compare photo:
View attachment 110492

To actual Admiral Kuznetsov, then named Tbilisi sea trial photo from 1980s:
View attachment 110493

Must be some really really old recon pod or epic level of zoom.
Click to expand...

Yeah, I take it back, Cold War era photos are way above the level of that first picture. Honestly, it would not look out of place in a gallery of WWII combat footage.

Did the Shandong accidentally sail through a time rift in a storm to end up in WWII? Are they doing a reboot of the Final Countdown movie, only this time is a Chinese carrier that sent back in time and did the Pearl Harbour attack? Sounds pretty on brand for the Japanese wannabe historical revisionists in charge of Taiwan to be frank.
 
You must log in or register to reply here.
Top