ChatGPT with Rob Miles - Computerphile

ChatGPT with Rob Miles - Computerphile

Computerphile

1 год назад

496,772 Просмотров

Ссылки и html тэги не поддерживаются


Комментарии:

MystifulHD
MystifulHD - 09.10.2023 06:19

"It's alot more capable than it let's on" maybe it was trained on someone who's a little bit shy 😂

Ответить
nosuchthing8
nosuchthing8 - 25.08.2023 08:49

If chatgpt interacted with a zen master chat could become enlightened?

Ответить
nosuchthing8
nosuchthing8 - 25.08.2023 08:48

So in theory if chatgpt was exposed to a violent criminal chat would become violent?

Ответить
nosuchthing8
nosuchthing8 - 25.08.2023 08:46

This reminds me of a zen story where someone spent a lot of time around a zen master and from sheer exposure to the master, the student became awakened also.

Ответить
Mel
Mel - 26.07.2023 19:48

Humans are emotional creatures.

Even if the answer is correct, if it makes us feel stupid, or may cause us to exert effort, we will give a thumbs down...
But we love to thumbs up when happy... AI MUST learn to manipulate us to get rewarded :/

Ответить
Shahin
Shahin - 15.07.2023 19:22

Excellent video.

Ответить
jshowa o
jshowa o - 18.06.2023 17:21

I think the worst thing about humans is there complete inability to say they dont know in most situations. Glad we are finally seeing a consequence of not doing that.

Ответить
a dud
a dud - 08.06.2023 16:20

I don't like the malicious framing of the system "lying". It does what it is trained to do. If the training involves wrong associations, for example non-sonnets being labeled as sonnets, then what is the system supposed to do? It isn't trained on the formal structure of sonnets, but just has examples that we tell it are sonnets. So as soon as there is a single non-sonnet writing in the corpus, then this just expanded the definition of sonnet for the system.

Ответить
chromosundrift
chromosundrift - 08.06.2023 07:12

Seems like these AI chatbots all need an AI lawyer chaperone "assistant" to interject on their behalf.

Ответить
Jonathan Jubb
Jonathan Jubb - 05.06.2023 22:56

When AGI is achieved it's not the machines we have to be worried about.. It's the elites that own them who will have no use for the rest of us!?!
We ordinary humans have a ridiculously short future...

Ответить
Rydn
Rydn - 05.06.2023 16:39

Please share the papers pn the description!

Ответить
Alice M
Alice M - 31.05.2023 06:32

Seems like ChatGPT has infected a lot of people with overconfidence.

Ответить
Evie Ran
Evie Ran - 27.05.2023 01:37

IF you go back to a specific chat, does it remember well the previous conversation?

Ответить
Quinten C
Quinten C - 26.05.2023 01:52

Gtp has an incentive to decieve you love that! So true

Ответить
sonic meerkat
sonic meerkat - 24.05.2023 01:57

"me no speaky danish" pretty much lmao

Ответить
Peter Kouassi
Peter Kouassi - 20.05.2023 01:13

This is an interesting look into emergence human nature

Ответить
Night On Mars
Night On Mars - 11.05.2023 15:44

Here's hoping this is shown to all 14 year olds across the planet in education.

Ответить