Future of Artificial Intelligence

[Quadalupe]

https://chat.openai.com/chat

This is pretty darn cool.

It's also used by some to solve programming problems on Advent of Code. The human only has to send the problem description to OpenAI, and the AI solves it in a couple of seconds! So cool to see.

[jacob]

EDIT: This is like INTP crack.

That's actually an interesting angle. It's plausible because the AI is internally wired much like the Ti primary of the INTP. This implies that the world's first AI might have an INTP-ish personality.

[ducknald_don]

Write a poem about extreme early retirement:

I retired at thirty-five
With a nest egg that was quite alive
No more long hours at the office for me
Now I'm free to do as I please

I travel the world and see new sights
No longer stuck in a daily grind
I have time for hobbies and passions galore
Life is rich and full, never a bore

Some may say I'm too young to retire
But I've planned and saved, no need to inquire
I'm living the dream, no regrets or remorse
Extreme early retirement, my ultimate course.

The freedom and flexibility are pure bliss
No more stress or worries, just endless possibilities
I savor each moment, each day a delight
I'm living the life of my dreams, an extreme early retiree, right.

[jacob]

That poem passes the Turing test for me when it comes to poetry. Then again, I'm helplessly bad when it comes to poetry.

Here's the made-up AI news: https://nymag.com/intelligencer/2022/12 ... orses.html

Don't forget to read the comments. One was written by the same AI. We're probably not far off from when people start filing entire blogs + comments not to mention forums with AI material.

AI: "Create enough material to correspond to 12 years worth of forum threads in a variety of financial and lifestyle related subjects using about 5000 different accounts in a Zipf-distribution pattern".

I suspect facebook et al, not to mention google, has a new problem on their hands. I might too.

[ducknald_don]

A big problem for teachers as well, my guess is many students are already asking it to write a proof of Euler's formula or a summary of the US constitution.

[jacob]

Good stuff: https://www.reddit.com/r/unsong/comment ... abet_with/ (note that "outputs were cherry-picked". Not quite there yet.)

After KSP, I'll switch to https://store.steampowered.com/app/4040 ... t_Command/

[Divandan]

Some great conversation in this thread although some of it is a bit more technical than I understand. I was made aware of ChatGPT last week by a work colleague and seeing some of the things it is capable of has made me think about it non-stop since then.

It was one of those things where for a while we have talked about automation and AI's role in the future, but then seeing the direct impact of it through a tool like ChatGPT is something else. It feels like my world has been turned upside down with some existentialism thrown in as well.

Some thoughts and observations:
This feels like one of those inflection points where we hit exponential growth in many ways. Each technological advancement affected a specific domain. This affects everything (digital at first)

What skills/tools become necessary in this future? I guess the skills of curiosity, resilience, and adaptability probably never go out of style? ERE definitely feels even more pertinent with these developments with some Buddhism and Stoicism thrown in

Assuming people that have conductor skills have some economic value in this new world? What happens to a whole subset of individuals in school/early in their career that do not have the direct experience to be able to successfully conduct/orchestrate the work?

For the average person not paying attention to this until last week, for the people tuned in, how big of a deal is this development? It seems like a huge deal to me, but I was also extrapolating a lot of pessimistic scenarios in my head, of what kind of disruption it will potentially have

Last but not least and the one that I have been pondering a lot, how do I come to accept this new reality? Talk therapy, gratitude, acceptance itself? Any ideas would be welcome as I have some experience in all of the domains listed prior.

[Ego]

Last but not least and the one that I have been pondering a lot, how do I come to accept this new reality? Talk therapy, gratitude, acceptance itself? Any ideas would be welcome as I have some experience in all of the domains listed prior.

An AI chatbot will soon be available to help you come to terms with the new reality.

My favorite example is DoNotPay, the automated lawyer built on OpenAI’s GPT-3 that fights parking tickets, automatically demands removal of photos from facial recognition systems and renegotiates hospital bills.

Here is a video of it negotiating a discount with Comcast chat.
https://twitter.com/jbrowder1/status/16 ... 5753309195

[Divandan]

@Ego thanks for sharing. It is incredible what the technology will be able to do but it just hits me a bit hard because I go straight to the societal ramifications this will have.

[fiby41]

What do ya'll think about Stable Diffusion and the other application apart from text generation-- image generation? Unlike OpenAI, which gives out an API, Stable Diffusion models are also open source, which means you can host them on your own like

https://beta.dreamstudio.ai/dream
https://creator.nightcafe.studio/create

have done. Above you can see 'prompt engineering' where you give a text input and images are generated by it. The detailed your input, the closer the output will match your expectations. You own the rights to the generated images. Precursors were MidJourney and OpenAI's DALL-E.

[daylen]

These tools had been all the rage in the Twitter-sphere recently. I've played with some of these tools briefly. I am pretty bad at generating what I imagine in my head. What I have heard is that understanding art history and lingo is quite important in being able to generate something close to what you want. This leads me to think that these models will tend to eliminate lots of low-skill art gigs and increase the supply/demand of mid to high skill art gigs.

I am curious about how the next generations of the models could fill in gaps of a procedurally generated world. A game or simulation framework/logic with some stylistic elements could seed an AI to generate an infinite world in that style. Combine that with VR and haptic feedback and we are looking at a new frontier of digital logics and aesthetics to be explored.

[Ego]

An incredibly useful tool. Highlight confusing text, math, or tables in a research paper and get a simple explanation. Ask follow-up questions and get instant answers.

https://typeset.io/

[Western Red Cedar]

I suspect facebook et al, not to mention google, has a new problem on their hands. I might too.

I usually watch a video podcast every day on YouTube. In the last few weeks I've noticed that some of the recently uploaded podcasts have extensive conversations in the comments which are pretty obviously all bots. They've been pushing some financial product or advisor. This was different than anything that I've seen before in the level of detail and conversation among the bots on the same comment. Based on other comments in the thread, I was clearly not the only one observing this.

Are we sure some of these new accounts with numbers in the name are actually human ?

[jacob]

I believe it is already being applied to economics and politics which would explain the unexplainable moves we've seen (ie. the confounding way we pulled out of Afghanistan) and I know it is being applied to war.

I have close to 100 hours in on "AI War", which is the most time I've spent on a single game in a very long time. AI War has the reputation for being the hardest or at least only undefeatable 4x RTS game so far. It has somewhat of a cult following because of that.

It is not a true AI but it certainly feels like it sometimes. As I started doing more research on my enemy (there are 12 different ship armor types, for example, UGH) I came across a series of six blog posts by the lead programmer. The idea was to emulate AI or rather make it appear as the enemy is intelligent. To do so they created a kind of distributed intelligence with three tiers (top, middle, bottom) where the tiers only know about the local strategy-space. Much like a decentralized command structure.

Point being, this intelligence is smart but it does not behave like a human. Closest analogy would be an ant hill insofar the ant hill had ant generals and an ant king. Insofar you don't have the [negative] attention of the ant king or the local ant general, you can walk right up to the ant soldiers without setting them off(*) Since the "vocabulary" is limited to firing lasers and missiles, it is also much easier to pass the equivalent Turing test. The main reason humans are amazed at chatGTP is that it uses human words. However, AI does not necessarily need to "output" in human sentence and paragraph constructs. It could also output in policy decisions as you note in the quote. As such, this game is an example of such to play around with.

(*) The AI War intelligence may only care about human foibles insofar humans become annoying to it. Otherwise, it just might leave or alone ... or accidentally step on us.

http://christophermpark.blogspot.com/20 ... art-1.html

[Ego]

However, AI does not necessarily need to "output" in human sentence and paragraph constructs. It could also output in policy decisions as you note in the quote.

Yes, the current war is a real world lab where the US is gradually trusting AI with freer reign. The horrible truth is that with no US troops on the line, abundant caution is unnecessary. The war-AI no longer has to produce explanations the generals can understand and use in their decision-making. Now it just eliminates the middle man and directs the squads, platoons, companies, battalions... itself. With each success we give it greater autonomy.

Of course, AI is still really bad at a lot of things humans just do automatically. See https://en.wikipedia.org/wiki/Moravec%27s_paradox

AI driven sensor production (swarms and such) as well as AI vs AI wargaming, which is actually a contest of who's data (sensors) are better, is the next big thing.

My concern is that battlefield successes may give (human) planners the confidence to allow AI to make key decisions in the endgame. Just enough carnage to encourage withdraw but not so much to provoke a nuclear response.

Point being, this intelligence is smart but it does not behave like a human.

The predictive power of AI is kind of spectacular. Feeding population-wide data to an AI and allowing it to predict the probability of loan default by an individual applicant works extremely well. Using that population-wide data to predict the future action of one outlier human with their complex psychology, biology, and chemistry, making a decision that has been made by only a few humans in history... that is a different kettle of fish.

[Frita]

Here is a timely article from today’s NYT:

https://www.nytimes.com/2023/02/16/tech ... 32f0bef97b

As a less tech savvy person, I found it a amusing yet disturbing read. I imagine this AI interaction to a good enough substitution for human interaction for many people. It leaves me feeling apprehensive.

[jacob]

Here is a timely article from today’s NYT: As a less tech savvy person, I found it a amusing yet disturbing read. I imagine this AI interaction to a good enough substitution for human interaction for many people. It leaves me feeling apprehensive.

Wow! That passes the Turing Test as far as I'm concerned(*). Indeed, I'd say that was a richer conversation than can be had with a 50%+ of the human population. Just log on to a public chat channel (if that's still a thing?) to verify by conversing with the average human. Based on this, I'd even be able to ascribe a temperamental type to Sydney (ESFJ). I wonder to which degree that is a true (innate) type or to which degree Sydney it mirroring back the type of the conversational partner (or the typical human being) or whether that's just the type of behavior that comes out of any rule-based intelligence (SJ). Insofar the training was done on the typical human, ESFJ is one of the most common kinds of human and as such set the tone for much of human interaction.

(*) I do wonder about the delayed deletion of rule-breaking answers. Best explanation is that there's some kind of human oversight behind the scenes---a moderator team---that gets called in based on certain trigger words. If not, then it implies that there's slightly delayed meta-thinking as in "whoops, I shouldn't have said that". That would be even more impressive.

[7Wannabe5]

Here's an interesting article from the OpenAI organization's Milestones in Research page. It describes the development of AI neurons.

https://openai.com/blog/multimodal-neurons/

[daylen]

John Vervaeke's take, which I am in alignment with:

https://youtube.com/watch?v=A-_RdKiDbz4 ... VrZQ%3D%3D

[jennypenny]

Harari has been making the podcast rounds with his thoughts on AI. Here's a recent speech that sums up his position (I was unable to find a transcript).