What ChatGPT is and what it's not: A three minute guide

AI chat bots like Chat GPT and its cousins have taken the world by storm in 2023. To a casual user, large language models are creating increasingly human-like responses - and yet this does not represent the reality of what is going on under the hood. In this video we are going to explore what Chat GPT and other like it are actually doing, what the future may looks like for these programs and ask whether it is fair to call them artificial intelligences at all?

[yqisq]

I want to draw a simple analogy of visual illusions which we understand fairly well. For instance, think about when we watch a movie it is really an illusion of motion created by quickly displaying still images, or a magician perfroming a seemingly impossible trick. They are all examples of what can appear to be quite convincing to our perception but it doesn't mean they represent reality.

[mariokotlar]

I feel you forgot to mention that despite all that we do know, the model itself is still a black box, and we don't actually know how it works, how it does what it does. We know how we set up the training algorithm that generated the LLM and how what methods we used to finetune the agents, but that's not the same as having an understanding of how it works. We don't know how it does reasoning, logic, common sense, any more than we know how our own brain does it. In fact, in most areas we understand the brain better.

[nosoyross]

Thanks for this kind of videos! Very clear and understood

[ChrstphreCampbell]

I would like these GPTs to check or produce their responses derived from a really responsible & accurate Encyclopedia of Everything, & then verify if their produced statements are Valid ( ? )

[হিজিবিজিওKnowledge]

For repeated action it's good but AI is still far away what is actually we know as intelligence. AI is marketed intentionally to generate some money in share markets as well as some education company

[Alex_Plante]

Within 5 years, this guy will be replaced by a bot.

[paulwary]

Although all of this is true, it still seems like a cope to me. A system doesn’t have to equal human performance to be classed as intelligent. An approach that is found to be effective may be at mouse-level performance on day one, and blast past humans on day two. The effectiveness of the current crop has surprised everyone, really, and we’ve barely begun things like introspecting and processing the network structure itself, melding neural networks with symbolic AI and so on. We don’t know enough about human intelligence to say whether our brains do something categorically different, so it’s presumptuous to say that current AI is ‘merely’ this or that.

[lucaslouzada]

It took decades to have a competent chess playing computer, even though it’s a linear mathematical problem - however enormously complex. Hardly it won’t take a whole lot more to achieve neural complexity. As AI evolves, so will evolve the conspicuousness of human expression, which is exactly what happened at every new technologic revolution.

[qixxxz]

That was a great mimic but that's all AI will ever be.

[jsalsman]

I'm skeptical that decades of orthodoxy about the Turing test being proof of sapience has been thrown out the window after it's been blown past. One very obvious problem is using the word "sentience" instead of "sapience" -- the dictionary definition supports the argument that a light switch is sentient. Another obvious issue being avoided is that thousands of researchers and engineers have already built agents out of a little infrastructure around LLMs. Nobody denies that such agents lack, well, agency and fully self-aware goal directed behaviors.

[ray]

Business chatgtp sucks, can’t deal with human complexity in interactions of speech. Still need humans. 😡😡🤷🏿👨🏿‍💻

[guldenaydin]

"Understand the world?"...
I don't think so.
The world is crying...
Words and sentences are not enough...