in my view (very limited though it is), things are moving much faster than the general public - and (again, from my limited perspective) at least some ai researchers recognize.
@@not.strictly.necessary you're absolutely right, we keep moving the goalposts on AI as it continuously flies past them and surprises even the researchers making it. The time to pull back and reflect on what we're doing is now...before it's too late
Genuine appreciation to John-Clark, TED team, and TEDxClaremont organizers for creating the conditions for these ideas to spread. Meeting people where they are, building out fresh analogies, and leveraging it all to subtly convey insightful content - chef’s kiss.
The issue, of course, is that "True" is a non-operative concept to the model. When it says something wrong, it's doing the same thing it does when accurate. That is, it is _always_ hallucinating, we've just convinced it to hallucinate useful things most of the time.
Language like species or organisms while referring to AI are a poor choice for describing how these AI systems work. Anthropomorphism is a known cognitive bias researchers trying to educate the public specially need to avoid. See Eliza effect. Emerging statistic patterns are anything anywhere close to an organism and it shouldn’t be used in this context.
ChatGPT is an application whose capability to Chat with humans is empowered by AI models that use GPT technology. ChatGPT allows humans to chat and ask questions to AI that has been trained on large amounts of internet data.
ChatGPT is an AI software as a service that lets you interact with an AI assistant. It is AI because they utilize AI tecnoglogies to make it work the way it does. The kind of AI they use can take text as input and gives text as output and it was trained to simulate an assistant. Nowadays the latest models can also have images, videos and audio as input and output. So, AI is the general field of research and techniques, while ChatGPT is something built using that research and those techniques.
As someone working on AI safety research and advocacy, great work!! One of the best talks I've seen directed towards the public
in my view (very limited though it is), things are moving much faster than the general public - and (again, from my limited perspective) at least some ai researchers recognize.
@@not.strictly.necessary you're absolutely right, we keep moving the goalposts on AI as it continuously flies past them and surprises even the researchers making it. The time to pull back and reflect on what we're doing is now...before it's too late
Genuine appreciation to John-Clark, TED team, and TEDxClaremont organizers for creating the conditions for these ideas to spread. Meeting people where they are, building out fresh analogies, and leveraging it all to subtly convey insightful content - chef’s kiss.
Really clear and concise, and great analogies too!
This is the best video on AI I've seen.
Geoffrey Hinton, the Godfather of AI prefers “confabulation” rather than “hallucination”.
So in that case, any future predictions by AI that are non-justifiable should be termed as 'Delusions'.
The issue, of course, is that "True" is a non-operative concept to the model. When it says something wrong, it's doing the same thing it does when accurate. That is, it is _always_ hallucinating, we've just convinced it to hallucinate useful things most of the time.
Interesting lecture after a long time
Really a great talk!
I totally agree on that.
Excellent talk - really like it!
i would love to work on this .
💪
what can the average person do to help?
Learn as much as possible about it. Knowledge and awareness lead to insight. Insight to action.
Language like species or organisms while referring to AI are a poor choice for describing how these AI systems work. Anthropomorphism is a known cognitive bias researchers trying to educate the public specially need to avoid. See Eliza effect. Emerging statistic patterns are anything anywhere close to an organism and it shouldn’t be used in this context.
What is the difference between chat chat gpt and Ai,
Chat is chat. ChatGPT is one application which you can have a conversation with it. Ai is more broadly intelligence
AI is a multimodal learning mechanism, with which we train different "AI tools". ChatGPT is AI, but AI is not ChatGPT.
ChatGPT is an application whose capability to Chat with humans is empowered by AI models that use GPT technology. ChatGPT allows humans to chat and ask questions to AI that has been trained on large amounts of internet data.
ChatGPT is an AI software as a service that lets you interact with an AI assistant. It is AI because they utilize AI tecnoglogies to make it work the way it does. The kind of AI they use can take text as input and gives text as output and it was trained to simulate an assistant. Nowadays the latest models can also have images, videos and audio as input and output. So, AI is the general field of research and techniques, while ChatGPT is something built using that research and those techniques.
❤