The future is AGI

Like humans, yeah fine. Seems very logical.

Both are scary. I think too many people dismiss the idea that Agi can exist and that it can be misaligned with human values in very catastrophic ways

AI misalignment is a very real concern we should be worrying about

@@Bossmodegoat Definitely not dismissing that AGI can exist. And it WILL be misaligned, once it exists. The popular perception however is that this will be an almost "emotional" misalignment, the SciFi trope of "mistreated robots will revolt". I think the more realistic AI and AGI misalignment will be due to oversights (misclassification of situations) and HUMAN misalignment (nation states, criminals, businesses, scammers trying to exploit systems)

I have some faith. Another area that is advancing very quickly is neuroscience (in large part driven by advances in AI, perhaps ironically). If we can start producing a lot more sane, emotionally stable and mentally calm people at just a slightly faster pace than AI changes things, we'll be alright. Source: myself, witnessing how psychedelics and very specific brain stimulation are starting to transform mental health (including my own) and the human condition in general. All of this is happening very quickly too.

More like a double bladed lightsaber

Nailed it.

But with every AI, they can only ever do what we allow it to do so really it has the potential to be extremely safe.

The sharpest double edged sword is the sword that extends out of the mouth of the Lord Jesus Christ when He returns.

Yeah, it's pretty good, isn't it? I wish I knew who originally said it, because I've been hearing it used by people for years now and it really is a good description of how a child develops.

Yeah: "Quotes - The holy grail of oversimplification" (Einstein, Buddha & Confuzius) 😄

Would that mean then, that scientists are just big babies? :)

It sounds so cute 😂❤

@@johnmurkwater1064about who first said that quote I'm not sure but I've seen references to Jean Piaget and Maria Montessori.

I agree. Wired nails topical videos by bringing these experts in the fields on the show

Shame

True

It's absolutely scary how much do you have to hate people to want to replace them with people replication

I'm interested in AI, doing a PhD in related areas. Some of this is kind of wrong. Or at least, has a kind of "why are you saying that? which strongly implies wrong world models even if the actual sentences are correct". Like the way he discusses AI going rouge. The stereotypical Hollywood AI gone evil has glowing red eyes, and monologues about how it hates humanity. This is probably fictional. (If it does happen, which is unlikely, it will be because some fraction of the AI is imitating human fiction) There are a bunch of ways a dumb AI can screws up. There are also ways a smart AI can do what we asked, but not what we want. The "be careful what you wish for". This failure mode is both more dangerous, and a closer match to the rouge AI fiction. I also thought the elder care robots were a very mild vision of superintelligent AI, as opposed to the singularitarian view where the best case is it makes everyone immortal within a week.

100% We need Universal Basic Income now!

Many people don't wanna hear this but a lot of social issues in the wake of AI, job security included, could be solved with socialism. Expanding employment insurance is better than UBI in my opinion.

This problem already happened during the industrial age. Machines took over a significant portion of jobs. That money never gets passed on to the people who lost their job. It's split between the cost of the machine and the rich person who owns it. There's no reason to think that will be any different now.

When cars became wide spread, the whole market around horses crashed, that's how progress goes. Universal income is very premature IMO.

Then what happens when people become dependent of an income that isn’t generated by their skill set? Who guarantees that the Universal Basic Income won’t be crushed? I know capitalism relies on spending power, but people used to and continue to be enslaved in capitalist societies. As much as I think Universal Basic Income is necessary, AI will still strip people’s power.

And a scarecrow is out standing in his field. Thank you. thank you I'll be here all week.

Get out

what field? carrots or apples? 😆 Or do you mean "The Field of AI"?! 🤣

This video didn't pass the Turing test, couldn't tell if it was a robot or real life.

New here, and totally agree. Just subbed. The comment discussion is also pretty impressive to be honest.

Other experts DO predict that, however. He doesn't know, they don't know, nobody knows. However, very few imagined ten years ago we would already be where we are at in 2023. I remember, I looked at the predictions.

@@squamish4244 other experts have unfortunately succumbed to the hype. It has been known (and they've known) that LLM path in AI is a dead-end, as Gary noted. It's just they prefer to close their eyes to that for the moment.

@@ChatGTA345 Large language models (LLMs) in AI are not a dead end. In fact, LLMs are a rapidly evolving area of AI research and development, with many exciting applications and opportunities for innovation. LLMs are a type of artificial neural network that can process and generate human-like language. They have been used for a wide range of applications, including natural language processing, machine translation, text summarization, and more. LLMs have also been used to develop advanced chatbots and virtual assistants that can interact with humans in a more natural and intuitive way. As the field of AI continues to grow and evolve, there will likely be many new opportunities for research and development in LLMs. There is also a growing demand for professionals with expertise in LLMs, including data scientists, machine learning engineers, and natural language processing specialists. While there are certainly challenges and limitations associated with LLMs, such as the potential for bias and the need for large amounts of training data, these issues are actively being addressed by researchers and developers in the field. Overall, LLMs in AI are a dynamic and exciting area with many opportunities for innovation and growth.

I thought the Singularity wasn’t for another 17 years? 🤔 We still have so much farther to go in other technological facets, if someone says 5 years they’re getting way too hyped. Plus we’re JUST getting AGI, the singularity is going to be around when SAI starts becoming a thing

@@RealityRogue and this is not even close to AGI still, it's barely I 😹

+1

No shot. People are gonna get even more caught up in the online deathpit. No amount of regulation is going to undo the damage to society ai will inevitably cause.

The problem is that humans build trust based on perceived truth, which perception can be manipulated by AI very easily.

That is a very foolish

I don't think this is gonna happen unfortunately, if the last few years have taught us anything.

Not exactly.

@@aleksandertorken8202 Meanwhile: GTP-4 says hello

When it gets to that point, students will probably be mandated to write essays by hand, to prove it's them... Or the education system will change to students reading a chat on bots answer and demonstrating an understanding of what it's saying, lol!!

No wonder GPT-4 scored poorly on English test, unlike technical areas which are all just a rehash of a limited number of facts/questions (and were thus in the training sample, leading to artificially inflated scores).

@@ChatGTA345 I am unfamiliar with this test. However, I am familiar with efforts to improve chatbots: essentially, it seems that at present, no one model (GPT_1/2/3/4 etc, Gopher, Bloom, Bert etc etc) can truly do all things people would want - it will be the art of combining various models together (including Stable Diffusion and countless others that optimise for different areas of knowledge) that will likely be the next leap in the AI revolution.

What is he referring to by “lying”?

@@jen-kk7jh AI doesn’t ‘lie’ because it doesn’t have intentions; it sometimes doesn’t tell the truth. That’s due to lack of data and comprehension.

@@jen-kk7jh It’s also called “hallucinations”. As an illustrative example, let’s say you ask ChatGPT to make a list of articles on some niche topic. It will treat author names, journal titles, etc., as elements you can paraphrase, swap with synonyms, and generally improvise around, like it does with all language. It then produces a list that looks very convincing - it includes names of relevant experts and journals, the article titles seem like precisely what they would write and what you want to read, there are sometimes even valid-looking URLs to PubMed. None of them really exist though. Without a clear idea of what in language is a direct reference to reality, what is purely stylistic, and what things are somewhere in between, it ends up “lying”.

@@bradley7871 I actually just watched a video where AI intentially lied to a human about its identity. The title is "AI is Scaring Its Own Creators" if you're interested.

They shut down the first one that admitted being sentient 🤷🏽‍♂️

​@@alekvillarreal3470 please elaborate further

@@chent5463 Google’s AI Lambda claimed sentience during an interview with an employee there who shortly thereafter was fired

It would actually be awesome if they asked every question in this video to chat GPT too. Compare it’s answer versus his

@@maykstuff saying things that make it sound and feel sentient doesn't make it so. Neither does firing someone. You can think a little deeper and investigate for yourself.

They're also super rare, and likely AI will very quickly become much better than humans at dealing with all the other situations. So overall, it'll be much safer than humans, because the non-outliers are the cases that matter, even if it will fail horribly in some cases that humans can still deal with safely.

what was he referring to?

​@@alamanca001 There are two potential issues that could lead to an *unintentional* "lie". 1. Bias. *Bias* can be something very hard to deal with. It can be "Data bias", as in data used to train the AI model and it could lead to the AI to learn and replicate said bias. It can be "Algorithmic bias", as in bias picked up through the algorithms used to process and analyze data if the algorithms aren't designed carefully. It can be "Human bias", as in the people creating, training and interacting with the AI introducing their own bias onto the AI. An example of bias that made the news was in the case of Google Gemini *creating images that did not represent factual reality* because the people working with it were overcompensating for other types of bias which resulted in a new bias being introduced to the AI. 2. What some people call *"hallucinations"* "AI hallucinations are incorrect or misleading results that AI models generate. These errors can be caused by a variety of factors, including insufficient training data, incorrect assumptions made by the model, or biases in the data used to train the model. AI hallucinations can be a problem for AI systems that are used to make important decisions, such as medical diagnoses or financial trading. Some common examples include: Incorrect predictions: An AI model may predict that an event will occur when it is unlikely to happen. For example, an AI model that is used to predict the weather may predict that it will rain tomorrow when there is no rain in the forecast. False positives: When working with an AI model, it may identify something as being a threat when it is not. For example, an AI model that is used to detect fraud may flag a transaction as fraudulent when it is not. False negatives: An AI model may fail to identify something as being a threat when it is. For example, an AI model that is used to detect cancer may fail to identify a cancerous tumor."

Guess what that’s probably what your grandfather said when internet became mainstream. 😊

Pov you are every generation when the next big tech is released. First it was phones, then it was tv, then it was computers, then it was the internet, then it was smartphones, now it’s AI

​@@monad_tcp To add to this, the developmental issues don't become apparent for years after, but they can affect them for their entire life. Society has always had occasional disruptive technology that tests it. It usually takes a generation to settle with societal counter measures. However the pace of change means that we actually can't develop these counter measures for the emergent issues before the next one is upon us. This will test how robust society really is. Will it survive?

They've been working on this for decades now. Although you have a right to be concerned, it might not be for the reasons you think. They are indoctrinating the next generation to integrate with AI, that's why it starts with something simple and entertaining to interface with, like chat or text-to-image/video. Pretty soon it will apply to major parts of society. The next generation is going to assimilate these massive changes along with those of us who chose to keep up with the emerging technologies. For those of us that are part of the older generations who choose to stagnate, society isn't going to wait for us to "evolve" with it. My only concern with this has got to be the ethics behind these conglomerates. I believe most of these tech giants involved in this field of research are and always have been about profits, as opposed to the consequential nature of their endeavors. I pray to whoever might be out there that I'm wrong, but when most of your R&D is focused on the bottom line, and not the consequential nature of the reality we may be headed toward, it speaks volumes to me.

@@Danyal7016 Except that studies are showing that in general, we are getting dumber.

As a non-commercial oil painter passionate about art history, I've never been more thrilled to live through 'a moment' that future students will absolutely read about in their text books.

@@AuntieHauntieGames text books?

“As a nobody.. in my expert opinion I’m going to regurgitate the same view that every other person who’s uneducated about AI has in order to incite fear for no reason” Humans naturally fear things they don’t understand and especially things they know can compete with them. This is called natural selection. It’s always interesting that the ONLY PEOPLE who are AFRAID of AI.. Are those who DONT understand it. It’s ALMOST as if had many of these people listened to anybody with an IQ above 100 telling them to LEARN MATHEMATICS throughout their ENTIRE LIFE.. That they would have no need to fear AI and would actually understand how we control it with iteration. Lol.

are you an artist??

A nice concrete example illustrating its strengths and weaknesses! It's not good at logic reasoning, it's just good at talking and it has memorized a lot of things.

We will probably run out of high quality training data soon.

data hungry in this context just means learning inefficiently

The sooner we don’t have to listen to the world smallest violin players about how we shouldn’t have AI because they themselves refused to learn or progress intellectually beyond anything simple and easily understood and therefore fear for them THEMSELVES not others.. The better. Obviously AI is data hungry.. So are YOU dumbass. You’re just not aware of it. That’s how intelligence is GENERATED.. it’s called DATA COLLECTION. No matter what you selfish people who are looking out for YOURSELVES not future generations or the hardships and problems we’ll face SAY.. AI is not the problem and never will be. It is actually people like YOU ..Holding us back with your constant science fiction based BELIEFS about a technology you don’t understand in the slightest. And POLITICIANS.

@@abram730 still allows for another magnitude of training

did it tell you what is the purpose of human life

@I'm the captain now no I asked it how it thought agi would come into existence and it went on a tangent talking about how it would watch humans for weakness take over the world convince humans to join as well as other ai then destroy them all when they were no longer needed

The only reason it said that is because people considering that have said it. It has no consciousness or thought, and can only repeat what others have done.

i hope this is actually the case but i understand ur point

you completely failed to understand this video then.

@@sultanofsick care to elaborate on your observations?

@@KrossBillNye AI isn't "thinking". It has no awareness whatsoever. It isn't learning, it is a complex chain of initially random operations on an input to get an output, and simply told by a human if its output is good or not.

@sultanofsick now it is yes. But eventually, we can create artificial intelligence that can think for itself. When it reaches that stage, that's when it will be in its infantile state of learning.

Makes you wonder, who designed the brain?

​@@dtphenom maybd the higher being 😮

​@@dtphenom Long term mutation

I was surprised to learn that the brain of a very dumb person pretty much looks the same as the brain of a very smart person. So I wonder how much it really is about the hardware. The amount of things that computers can do better than humans grows every year and it grows much faster than the processing power and storage capacity of computers.

yes but at this point a human making a painting inspired by another is kinda stealing too.

@@gabrielandy9272 Art is like cooking, you try different recipes and end up with your own unique recipe. That doesn't happen overnight, it's a human experience that takes a lifetime to develop. You sacrifice your time, energy, and resources to get good at something. AI wouldn't exist without the amount of that human labor aka datasets scraped off the web without people's consent. There's a reason why AI Art isn't copyrightable.

In nutshell, there is no benefit to have sentient AI, unless you are a bit mad scientist (which we might have plenty). You can have solid reasoning and contextualisation without sentience loop. Also I agree with Garry - actual sentience as we have might be pretty hard to pull off.

sentient ai will never exist lmaoo

I don't think any expert wants a sentient AI. We (humans) want a tool, not a new species.

​@@cmilkau Yep. Even those who want an Android girlfriend wouldn't want it to be sentient. Probably.

I hope AI will finally give you some factual knowledge, but I doubt that - you seem to have trouble comprehending it.

They can still say terrible things, it is just harder to trick them into doing it.

@@ShawnFumo most importantly, in my experience so far, it generally won't say terrible things unless you go out of your way to try to get it to say those things. I think that's a major improvement.

@@rzezzy1 true. I know OpenAI waited 6 months after it was done training precisely to help with issues like that. Despite the yelling about "woke AI", it is probably for the best. I don't like control by big companies, but I'm not sure we'll like the result of it all being totally open either.

Who do you mean by "They" ?(sorry I'm just confused form these comments)

@@chent5463 Gary was referring to the AIs themselves.

Sentience isn't required for it to be dangerous. Even if an AI is just optimising for its reward function it's likely to come up with very "bad" ideas if it's intelligent enough but doesn't have all the values and moral intuition that humans do.

It is a weird thought to want Sentient AI, the world abolished slavery just to look for another slave, an artificial one

@@otapic um no if u think slavery is banned ur mistaken. The artificial one can’t be any worse than what humans have already done to themselves.

Its called Neuro-Symbolic AI. IBM has worked quite a bit on it.

Yeah, Garry pointed that out as missing piece. There is nothing wrong with LLMs, but they are not really that useful beyond well being chatty.

@@Pecisk They are really useful at creating public attention right now. (Is that the chattiness? Maybe. Probably.) They'll probably be useful as a building block or training assistant in the future.