Just saw this Google engineer interview I suppose the media made out to be a hack. His arguments and criticism on ethics and alignmnent are on point and reasonable.

https://youtu.be/kgCUn4fQTsc

This is such a great thread. I've been on 2+2 for many years, and never have seen OP of this thread start one.

Very well thought through and definitely food for thought if one is of intellectual bent.

My only addition to the conversation is this - I heard that the google AI and one of the others both passed the Turing test. That seems like quite the rubricon. I haven't looked into detail, but my (admittedly rudimentary) understanding is that the "human" traits we have are sarcasm and, for want of a better term, "tone". If the AI can fool the judges in those regards, we're all pretty much ****ed.

We're beyond that issue now with deepfake. If somebody wanted to they can take say an influencer on Instagram with 50k to 1 mil followers and re-create their entire personal likeness with deepfake videos of their face and voice. Essentially create a fake realistic alt account.

A lot of people depend on social media for business income, marketing. If bots could take over the process it's big trouble for these social media companies.

The deepfake video is maybe a couple years away from 100% being indistinguishable from real. That Paris Hilton, Tom Cruise vid makes me think it's there now though.

Then there is what is admissable in court if all video can be faked and incredibly hard if not impossible to ID as fake.

We just saw Twitter push professional accounts into paying a subscription to keep their verification. It was as much to provide another layer of security as get more revenue for Twitter.

They are trying to get ahead of any bot storms on ID theft if and when they come. Twitter spaces has a lot of good AI spaces on the tech and philosophy. Twitter is the best breaking news source on AI.

Then you have the route these big tech companies like even Snapchat now(they have a LLM chatbot) are taking it. They are training the LLMs with all the conversations it's having with the general public to immitate a human perfectly. The Snapchat bot even calls itself a person usually.

I think a better path for LLM tech is smaller focused models that can solve more problems than be chatty. We're already seeing this the medical field, astronomy, law, any field with large datasets.

But to stay positive I see these early models as eventually being as helpful as Data the android from Star Trek but without the human mobility. Its getting close already. I would say it is already ahead of Data in understanding a couple emotions.

Multi-modal input is advancing quickly too. Faster than i expected. These models are learning to understand the world like we do through video and pictures. It can already tell u why a meme is funny. Multi-modal is the future. ChatGPT can already work with images and 3d.

AI listens and Does Magic
https://twitter.com/Scobleizer/statu...tBuQPE0dg&s=19

Good OP. It's been interesting to see the move from AI denial to a bit of a panic in the press and from governments. I'm a fan of regulation in applications but we will make a terrible mistake if we think we can curtail it now.

AI (or Alife) might replace us but somethign more like cybourgs are at least as likely. Neural implants and augumentation will be awesome and coming to a place near you real soon.

Space travel aka the great ****ing off will become a reality. VR will be epic - nothign virtual about it (the V will quickly be as redundent as the A in AI)

I've been banging of for ages how ~all the jobs are going and that the capitalist era is drawing to an end. We wlll either find a new way to distribute wealth via some form of democratic socialism and or UBI. Or democracy will be under serious existential threat. Warfare is the stuff of nightmares - forget terminator, think swarms of cheap, small AI drones/insects. Printed on mass.

In all respects the future will be awesome. For good and bad.

There will be sex robots.

The big philosophical debate will be when this stuff has rights.

It's hard for me to see the prospect of the celebs and influencers market being destroyed as in anyway a bad thing.

It's just a question of how long

There's a bit people tend to miss which is that the world get's modified alongside advanced in handling the 'old' world. So not only will AI get rapidly better at legal reasoning, the laws and legal systems will co-evolve to be more 'AI Reasonable'

I just used that as an example. It applies to any public figure or news media. Politicians, news sources. The trend for sometime has been for news to be harder and harder to verify on social media where most people get their news. This will push it to the extreme of lack of authentication. I haven't quite wrapped my mind around this outcome. In hindsight bot misinformation already happened in the last U.S. election. Doubt it is good for a democratic republic. I love the AI though. there's just a lot to work out.

I'm dubious that the accuracy of what they get isn't so bad that they wouldn't be better informed if they knew it was made up.

Everything will happen in the very long run. When I say that I have difficulty imagining something, I mean that I have difficulty imagining it happening in a time frame that is easy for me to conceptualize.

For example, you may be right that laws will co-evolve to be what you call "AI-reasonable." But I highly doubt that you (or I) could explain what that really means, or what form it might take, with any specificity.

This is definitely true. And it raises an almost existential question that has bothered me for a decade or more.

The plasticity of the human brain is exceptional, but I'm not convinced that we can accommodate an accelerating rate of technological change without profound psychological distress. Even in today's world, older people are extremely susceptible to feelings of alienation and being lost in the modern world. It's hard to adapt to a world that is very, very different than the world you grew up in.

The children who are being born today may live in a certain type of world when they are 15, a very different type of world when they are 25, and a profoundly different world when they are 35. Changes that previously occurred over a period of 50 years may occur over a period of five years or less--far too fast for most people to effectively adapt to them.

Another area in which AI is generating a lot of debate is education, which I didn't mention in the OP. For many people with school-age children, education is probably the first thing that comes to mind when someone mentions AI.

The first-level concerns are prosaic--namely that AI will facilitate cheating. The deeper concern is that education will move away from teaching kids how to think analytically and express themselves coherently, either in writing or verbally, and move instead in the direction of teaching kids how to extract information and build arguments using AI.

In the late 18th century, most educated people had beautiful penmanship. Now, most teenagers can barely write in cursive. Very few people view the loss of penmanship as a big deal. And that's understandable, because it simply isn't that important to be able to write like a guy who signed the Declaration of Independence. But if the ability to think analytically and express yourself in words with minimal assistance went the way of old-fashioned penmanship, the implications, civic and otherwise, could be profound.

On the plus side, many people view AI as having tremendous potential to benefit education. Here is a TED talk from Sal Khan, the founder of Khan Academy, which makes the positive case:



Khan imagines a world in which AI effectively will function as a personal tutor, with all the usual benefits you would expect from personal tutoring. And he gives a lot of examples of how his education AI, Khanmigo, functions as a personal tutor.

I don't quibble much with Khan's vision of how an AI could benefit education, but in this TED talk, I think he mostly elides the question of whether AI in fact will have a positive impact on education. In other words, he doesn't grapple with the fact that students won't be limited to using Khanmigo.

I came up with the with what I call BMD, Brain Modelling Disorder a decade or so ago. It's a problem we will face if we conquer aging to a great extent and even if brain plasticity/etc suffer no degredation (or are enhanced) we will hit the the problem you describe. A fast pace of change will make it even worse but it as you say, that who we are is deeply bound to the model of the world we make of it while growing up. There are fundamental problems when we keep updating those models.

On education we will learn different things. Learning facts, writing essays etc will hopefully be junked alongside the slide rule (yes I am just about that old)

I'm sure the same scared of change arguments were made when the printing press was invented and got widespread(less memorization), computers in the 80s and 90s, internet. Cheating has always existed. The biggest gateway for cheating was created with the internet. To say the internet then was bad for education is funny. Person in the 1980s and 90s. "look computers are teaching kids how to extract vast amounts of information faster! Bad!" Technology is never the evil doer. it's how it's implemented and used.

If we conquer aging, what's preventing us from creating some "ideal individualized world" or in 80s Hollywood terms a "total recall" or anything that solves this particular issue. I mean, AI literally can create an artificial world for those who feel alienated to current times.

I think a lot of people forget that these large LLMs are in essence just the internet and books. Bing's LLM was trained on over 100 million books. This video breaksdown the basics of the large LLMs. Nice lamen explanations. I especially like the tokenization explanation because I never quite understood that.

https://youtu.be/bZQun8Y4L2A

I remember my mom talking about my grandmother who lived in the 1930s and 40s was initially turned off by microwave ovens because she couldn't understand how they worked. Thought they were dangerously radioactive or something.

Not without an absurd amount of processing power, which requires a lot of energy. Energy limitations are a real impediment to the creation of a Ready Player One level VR.

Sure we may be able to create a reality to live in that's based on when we were young. Or at least spend most of our time in it. Better than life (as in Red Dwarf etc) will be brilliant. A;lso a possible answer to the fermi paradox - VR will be so much more fun and interesting to explore than the contingent near empty lump of rocks we happen to be in.

As long as we dont have to earn money to pay for it and look after us in the outer world.

It's hard to envisage a future where energy isn't basically free.

Because you think Dyson spheres will be cost free to build?

If only we could find a way to harness the power of chez' imagination, we would be a Type II civilisation in no time at all.

There's no shortage of low entropy energy (to put it very mildly) and we're rapidly learning to scoop it up

I have a decent imagination but this is barely in the realm of imagination

Threat feels real - already serial lying from public figures to the point where
lines between truth and deceit are blurred, no doubt aided by their use of AI.

Nothing creates believable bullshit quite like ChatGPT.

We've been making a push lately to get all of our websites brought up to date and create a backlog of blogs to assist with SEO. We've hired a writer who came from print. Apparently this oldschool news man isn't fast enough in researching and writing articles, so we're being guided to utilize AI to speed things along.

It's still a small business, so I end up looped in on a lot of this stuff even though I know jack **** about SEO or web marketing and it's not really my focus. I have now seen some of the material as to how the sausage is made in the modern SEO/blog industry, and it's honestly disturbing.

I played around a little bit with ChatGPT on things I do have some expertise in. The primary takeaway is that the AI is slinging out complete and utter horseshit as fact, and countless communications majors around the country are copy/pasting it onto websites and curating high rankings on Google.

There is absolutely no way for a layman to know if they're being fed text with any basis in reality. This isn't really any different than human writers making **** up and posting it Facebook to rile up the deplorables, but AI can spit these things out as fast as you can feed it prompts, and it won't have most of the telltale signs of having been written by a crazy person like most awful content. If the internet wasn't already filled with garbage, it's about to get much much worse.

Inso0,

It's hard to not feel jaded when dealing with SEO.

AI's intrinsic ability to generate quality content will only improve. But that's cold comfort because enterprising humans always will find ways to use AI to generate misleading content at scale. And your point about an AI's ability to generate misleading or awful content without displaying the "telltale signs of having been written by a crazy person like most awful content" is very important imo.

Just wanted to thank everyone for the mostly sane posting in this thread so far. My hope was to generate a longer-lasting, slower-burning thread on a topic that wasn't as intrinsically laden with partisan priors as some other topics are.

Seems to be working so far.