AI thread

[dviz]

Right now it's just getting really good at cobbling together data that it scrapes from around the net and presenting it in a way that (most of the time) makes sense based on the prompt. Basically just calculating the probability of each next word making sense and applying given the prompt. There's no real thinking going on there behind the mask.

Again, how's that different than your inner dialog? Don't you think, when your thoughts wander, the same process of "finding the statistically most appropriate next word" is at work?

Confused people are calling generative AI a "stochastic parrot", ignoring the fact that we work exactly the same way.

However, there are some scientists and researchers involved in the field who argue for a push away from the current "generative" AI models to "goal-oriented" or "predictive" AI, which would be the kind that might eventually lead to something greater, but that requires a pretty fundamental shift in conceptualization of the technology and the underlying machine learning models, neural networks, etc.

Absolutely, more complex goals will produce richer response from AI models. At the moment, the only goal chat bots have is answering questions in a informative manner that also keeps the user glued for more interaction. The models can be crafty in achieving this goal, but overall, the goal is simple.

Humans and other living organism, on the other hand, have a variety of goals because they have a will-to-live and lots of short and long term goals for needs that have to be met. AI models don't have, nor need, a will-to-live. They have their needs already met, they are connected to the infrastructure that keeps them going. Living organisms are in a much more precarious situation, they have to spend daily effort to fulfill their needs.

In the future, depending on how much autonomy we'll afford to these models (particularly AI models in autonomous robots), the goals will increase in complexity, and their response will also be richer. We'll have to prepare for a world in which these models roam free.

I also think a huge breakthrough in the tech will happen when we combine quantum computing with AI.

Even though it was posited by some physicists (like Penrose) it is very unlikely that human cognition has any quantum aspects to it. There are other more fruitful avenues to try before quantum computing to achieve something akin to AGI.

I wonder what will happen if and when they are able to do most people's jobs though. Do we naively believe we will just stop working and live in a leisurely utopia? I think not.

AI will bring major societal changes. It's not going to be utopia, but not dystopia either. People will move to other jobs. Fewer people will work on assembly lines, and instead will train (not program) AIs. More people will work with other people (in entertainment, or for elderly etc.). Some form of basic income will also be part of the deal.

[dviz]

Rule by algorithim is also very dangerous - just ask a postmaster!

It depends on the complexity of the moral judgement infused in these models.

It is conceivable that AI models will be fairer judges than humans in the future, if the training was (agreed upon as) fair. Humans on the other can receive fair instruction and still apply it in a biased way.

[dviz]

AI can write an article like human does, but AI currently cannot understand what is written, the human can. The AI is just throwing words out there. These are just LLMs not real AIs.

What do you mean by "understand"? How do you understand your own sentence, written above, better than an AI model? Can you be more explicit?

Certainly, an AI model can explain your sentence using other words, and answer questions about your statement. Is this what you mean by "understanding"? How do humans "understand" your sentence differently?

[Creoda]

You can do very neat things with AI too. As usual with technological advancements, it just adds a new layers a good and evil on top of all the previous ones. One day new wandering minds.

Gotta post some more. Within a couple of years we'll probably be able to watch full versions, at this rate.

[Creoda]

[jakob]

I used chatGPT to write a bash script for me to download photos of soccer players from a roster webpage. I'm incompetent at that sort of thing so it would have taken me at least 30 minutes to figure out myself. It even commented the script, which was nice.

[Lemurian]

Gotta post some more. Within a couple of years we'll probably be able to watch full versions, at this rate.

That is so cool, imagine what we will have in a few years

[Creoda]

That is so cool, imagine what we will have in a few years

Most of the film industry out of work I should think. And perhaps the music industry too.

[Colonel Frank Grimes]

^ this is amazing, Creoda! It’s so accurate.

It's just selecting information from specific threads. In some instances, it gets information wrong (for example, "Cristiano Viejo’s antagonistic behavior toward Spaniards and his unwavering focus on promoting Spain left a lasting impression."

It basically got all my info from one thread (probably because my 'name' is in the title).





Both screenshots are information I gave in one thread being repeated back to me.

[Colonel Frank Grimes]

It basically got all my info from one thread (probably because my 'name' is in the title).

All taken from two posts in the same thread and because Mortimer's name is in the title
https://www.theapricity.com/forum/sh...-Status-Update