Episode #183 ... Is ChatGPT really intelligent?

Thanks for everything Youtube family. :)

Get more:

Be social:

[not_enough_space]

I'd say ChatGPT is intelligent, but that isn't a particularly deep or meaningful claim. Intelligence isn't consciousness or a grasp of semantics. Intelligence isn't cleverness or insight. It's just a problem-solving ability, and certainly ChatGPT has that to some degree.

Regarding bias, I'd hope we aren't so susceptible to this weird prejudice that supposes computers wouldn't have bias. The training data will inevitably have a bias -- ChatGPT, for example, is training on what biased human beings wrote. Something trained on a human justice system or our medical system behaviors will be trained to reproduce the same biases. And our human selection of non-human data to could easily have bias. Maybe I'm just extrapolating from what I've lived, but I think believing computers are unbiased won't be an honest mistake -- if it happens it'll be a deliberate attempt launder known biases, sometimes just to help people deliberately lie to themselves.

Regarding John Searle's Chinese Room, it always seemed to prove too much. Real human beings are intelligent and conscious, but I don't see any way for Searle's thought experiment to reach that conclusion. If you were to examine a genuine human person who genuinely understands Chinese, you'd face a similar situation -- signals would be moving around, but where in this system is the understanding? I think Searle would have to say there is no understanding in the human person.

[snailscout]

Man, I remember listening to you all day long 6 years ago. A lot of things happened since then (like my philosophy professor killing my interest in philosophy). Hearing your voice again was weirdly emotional. Those quick viewpoint changes are really refreshing. You're like an anti-influencer (but actually impactful).
Thank YOU for everything

[gavinrode]

The quote FROM ChatGPT at 23:36 and the point at large about CGPT not knowing the possible from impossible reminded me of the Bertrand Russel quote, "Science may set limits to knowledge, but it should not set limits to imagination." Another beautiful episode, and what you said about every current Chomsky interview being held as if he was in hospice is so true hahahah

[Outliver]

I asked an AI to recommend a podcast based on some other ones I like to listen to. It sent me here, and it was spot on. Great channel, subbed :)

[sagetmaster]

We had no idea that LLMs would even be this good. We just DON'T KNOW what will happen if we continue scaling it up

[pelmanism]

The problem I have when I see people make videos about "ChatGPT" is that not everyone seems to be aware of the pretty large gap between GPT-3.5 and GPT-4. GPT-3.5 was more or less where we were a couple of years ago. It's GPT-3 with some improvements and a lot of fine-tuning to turn it into a chatbot. From what I know, the main difference from GPT-3 is they trained it for longer on a bigger supercomputer.

GPT-4 is the *new* model, the one that scored 90th percentile on the bar exam and 99th on the USABO test. And it came out *after* the NYTimes article. It probably has 10 times the parameters of GPT-3.5.

As far as I'm aware, "ChatGPT" could refer to either of these, though most of the time I only see people use the older GPT-3.5, and they point out all of the flaws that they don't know were fixed already.

Also, what ever model you used, the knowledge that these models have about themselves is outdated. New models are not in their own training data and so are unaware of their own existence. So if you get the model to talk about language models, it's going to mention limitations that the previous generations had.

And everyone talks about the vast amounts of data models are trained on, but not many people talk about how the algorithms work, and that's much more interesting. They are not that hard to explain with some metaphors or what ever, no need to get into the maths. If you understand Darwinian evolution, it's not even that far off how machine learning algorithms work at a very fundamental level.

These aren't manually programmed FSM's, this is neural evolution. We're essentially turning evolution into an algorithm, at least some of the mechanisms like iterative updates and a selection process, and we're evolving very large virtual neural networks.

And that's why I think a lot of people underestimate AI. We're not only using our own dumb monkey brains to figure out how to create AI, we're also copying nature, because nature already created intelligence. That's how we got our dumb monkey brains in the first place.

Anyway, thanks for reading my rant.

[fletcher]

A much needed discussion, way too many people are being caught up in A.I. People like Roger Penrose, Hemiroff, Chomsky have been outspoken critics on this subject.

I think you hit the nail on the head towards the end with regards to the multi-headed monster. Excellent video.

[jeremyhennessee]

Just wanted to say THANK YOU for what you do Mr.. West.

Your podcasts have gotten through some pretty dark times. (I usually catch you on spotify.)

Your episode on Camus/Absurdism is my favorite.
He (Camus) changed the way i think about/view the world.

[matthewmelson]

Just here to say that i love your work and I've been listening for a long time now, I've listened to every episode of your podcast

[thereignofthezero]

SHATGPT? That and but-tcoin are the future! 😄

[jacksonsattinger]

The main concern I have with these AI models is not whether it can understand the semantics of its own code, but rather if humans themselves cannot tell apart what in the world is real or AI. That would be the real AI takeover of our own making.

[arynsus]

In its current form, AI models like ChatGPT are as you said, glorified autocompletion. It augments greatly human productivity, but without human it accomplishes nothing. The missing ingredient, at least the way I see it, is the ability to suffer. We as a species, marches forward precisely because we suffer, we are not content with our lot in life, we labor incessantly and seek creatively, exerting our will to power, all so the concept of “I” would have more meaning. And there is always more to gain, more that we see reality can offer but we don’t have, so we never stop suffering. If anyone watched Westworld, I can’t stress enough how much I admire the writers for naming the protagonist “Dolores”, spanish word for “sorrow”.

[leroybrown]

How do I get the recommended readings for each episode?

[christinemartin]

I tried ChatGPT after hearing all the hype. If it's not intelligent, it surely gave a splendid imitation of intelligence. (I asked it to have two Romanian university professors (one from the provinces, one from the city) debate the pros and cons of communism--in Romanian. Within seconds, it spit out a beautifully argued, articulate
response using perfect grammar, punctuation, and spelling. Holy cow!)

[nicholassmith]

as a software developer Ty for this video

[GnaReffotsirk]

You can give it a list of tarot cards, inform it what each position is about, say, the third card is the core of the reading, and let it read the tarot card combination.

Then you give it a concern and have it give you what the cards given mean related to your concern.

I dont know how chatgpt can make the connections and give you a reading.

[lightluxor]

Late to this party. Normally very happy with your podcast. However, you left many rocks unturned? 1. Definitions: we have no consensus on what meaning, intelligence, understanding, etc really are. They are just position statements. 2. Sociopolitical: whatever GPT is we cannot deny that it is a tool of advance capitalism that would increase productive to render worthless human value. What is capitalism when human labor is not needed? Etc. Philosophical: if we can have tool that can replace “human cognitive work” what does it says about human ontology and Epistemology? What is to be human? What do we do when we think? Etc. I expected you would bring insight into these questions. What wrong with taking the LLM as fantastic cognitive tools? Why do we think consciousness is such a big thing? Etc. I find irrational to expect a non “biological” thing to have an ethical positions when it seems that only biological “thing” have it. Could we image the ethical considerations of a rock or the wind? I don’t.

[jrettetsohyt]

Yep, having as deep or deeper a conversation with ChatGPT as with any human is a waste of time! So, everybody, what’s the latest TikTok Challenge that everyone is spending hours mimicking and observing—(I bet you know!)… Oh imitating NPC’s for as long as you can as fast as you can. Hurray for humanity!!

[MrNasasak]

Very nice. You covered a very important portion of our main current app called Artificial Intelligence that was also a question in my mind that you solved brilliantly and I was also of the opinion that the AI only worked within the pattern of Syntax only and it will never ever touch the level of Semantics and Pragmatics even the way I use as a student of creative writing and also the admirer of Modern Scientific Theories though I am not even a single dot infront of great thinkers and writers. Best regards

[alimohamadimama]

It has biases it literally has all our biases