AGI is Almost Here! What comes next? What's left to do? How do we adapt and prepare?

New Era Pathfinders, my Skool community: www.skool.com/...
Heavy Silver: www.heavysilve...

[@johnthomasriley2741]

The jar your brain is made of bone. Still a jar.

[@DaveShap]

Brand new sentence!

[@antonystringfellow5152]

And what is the function of experience? Reward/penalty

[@Maneyax]

Do you mean "Your brain is in a jar made of bone"?

[@johnthomasriley2741]

Typo: The jar your brain is in is made of bone. Still a jar.

[@Nuggiesoftruth]

@@DaveShaphahaha you finally took my advice and made a skool

[@BAAPUBhendi-dv4ho]

Every time you watch this video AGI is 24 minutes and 15 seconds nearer

[@barry1807]

Unfortunately so is the grim reaper

[@Tracey66]

@@barry1807 #DontDie

[@evil_duck6405]

@@barry1807 there won't be a grim reaper after AGI/ASI

[@etherlord300]

Since I watch most videos at 2x speed, it gives me an extra 12 minutes and 7.5 seconds to me.

[@autohmae]

@@evil_duck6405 that too is cope

[@AbhijitKrJha]

"Most humans are confidently incorrect." best summary of our kind :)

[@DaveShap]

hehehehe

[@stereo-soulsoundsystem5070]

This could be applied to the race for AI. What if we're confidently incorrect about its uses or ability to be safe, or end labor without subjecting people to poverty?

[@r0ck3r4ever]

​@@stereo-soulsoundsystem5070 there will be poverty and there will be less people in the long run and once the down spiral is started. It will take maybe hundreds of years to recover. The techno optimism could be just the next christianity movement that collapsed the decadent roman empire, the result of that was the middle ages.

[@bulltheknicksfan3140]

These are the videos the people want. You are the Chief AGI Preparation Officer for the AI community

[@DaveShap]

Finding my stride. Thanks :)

[@sesamring7065]

What's your Profession? I am CAPO 😂

[@onlythistube]

As AGI is not an empirical term, i.e. not well enough defined or globally agreed upon, it depends on the intricacies of the individual definition of AGI wether embodiment is mandatory or not.

[@Danuxsy]

@@onlythistube Deepmind has already defined the word AGI and I don't take anything else as being AGI, I know David will try to lessen the requirements to make it seem like his prediction of AGI being here in 2024 (LMAO) is true.

[@phen-themoogle7651]

@@onlythistubeSomeday there will be a worldwide vote to determine the year AGI happened, long after ASI is here 😂 Or ASI will define it for us and tease us for being too slow😅

[@RetireandGo]

I’m amazed by some of the O1 results and reviews. A 30 year oncologist said he is impressed with the treatment plans it’s creating for patients… even detailing aspects he did not think about

[@adampenbrook5751]

Damn. I’m both excited and scared by this lmao.

[@takeuchi5760]

​@@adampenbrook5751 ikr? If this thing can do doctors, engineers and scientists' work this well then it's smart enough to do any intellectual work in any field.

[@saadahmad438]

we're gonna get AGI before GTA 6 😭😭😭

[@Mropinionated3]

@@saadahmad438 we gonna need AGI for gta6 atp😂

[@vangildermichael1767]

The (system) will show AGI before 2026. I think people are looking at the issue naively. We are not going to (get agi now) or (ever). We (got) agi a long time ago. A long, long, long time ago. Like over 500 years long. This kind of thing will make this current system stop working. The system where everybody stays in their little box, and gives me the cream of their work. Now it's all gonna stop whenever this technology is finally unveiled. (before 2026). maybe even before 2025. maybe? Maybe before the sun goes down today. All the work is already done.

[@robsteinbock]

It's absolutely wild that we're in the process of choosing a new leader of the free world and not even a blip of this and its implications are being discussed.

[@denjamin2633]

The government has and will always be reactive. I personally think we'll get to a promised post labour utopia, but there's going to be a few years of riots, recession, and starvation first from the mass unemployment.

[@concernedindian144]

I think this is the reason Elon Musk, Peter Thiel are so hell bent on reclaiming power through Donald Trump at any cost, they know what's coming, AGI is going to create a new world and they want to hold on to the power and abundance it will create

[@higreentj]

America is an oligarchy so it is more about what does corporate America and Wall street want. Technologists like Musk and Thiel want a universal basic income (UBI). Predictions for a GDP of forty one trillion by 3034 seem way off if we have AGI and superintelligence, so a UBI could be easily funded by borrowing.

[@markupton1417]

"free world"....

[@MrPiperian]

Oldsters who can't understand Email are going to dictate AGI policy to the US.

[@Dragon-ul8fv]

2024 was always the year.

[@jaywv1981]

And September was always the month.

[@Danuxsy]

There will be no AGI in 2024, GPTo1 proves it. You are completely delusion if you think otherwise, there is no evidence for any form of AGI whatsoever in these systems. I KNOW that David is wrong lol

[@Danuxsy]

AGI will not happen in 2024, I bet that David is COMPLETELY WRONG and delusional. Mark my words. Nothing about GPTo1 points to AGI, in fact it does the opposite. OpenAI are unable to overcome the hurdles that still persist in all LLM's since the beginning (which was expected since OpenAI did not invent LLM's and the people that did have no idea how to move beyond it either).

[@Xune2000]

Wake me up when September ends.

[@null7936]

:|

[@jippoti2227]

It's crazy, if your prediction of AGI this September turns out to be correct.

[@thebeezkneez7559]

He is certainly a lot closer than most people gave him credit for despite being off as well.

[@Danuxsy]

he's completely wrong lol

[@mmaslav6176]

​@@Danuxsy how is he completely wrong?

[@Katatonya]

@@Danuxsy He might be, he might not be. But if you say anything with certainty, like your comment, you're never to be trusted because your logic is flawed from the start.

[@mtdfs5147]

How is he "completely" wrong? He predicted agi within 3-6 months, 3 months later we get strawberry.​@@Danuxsy

[@briankrane5546]

this channel has become one i fear to click videos for how extremely optimistic the comments section is about "how close are we now?". This place is like it's frozen at the peak of the hype cycle everyone is talking about.

[@dot1298]

i‘m not - AGI is a pipe dream of hopium

[@FlorianUniversität]

@@briankrane5546 It's because we know much more about these topics than the average person.

[@efraimmukendi7137]

Yeah I used to think this as well but it’s getting crazy. I also want these ai take take our jobs and we get ubi

[@cautionroguerobots]

And it’s absolutely vitriolic to anyone else. An echo chamber like anywhere else online unfortunately.

[@ShaneMcGrath.]

@@efraimmukendi7137 Need safeguards first! I do NOT want UBI if it means there are conditions, It needs to be unconditional no matter what.

[@nijario9690]

I think we can safely say that no one Exact knows what's gonna happen, this will drastically change the economy

[@zbird2007]

"I'll be surprised if this event occurs before 2005 or after 2030." - Vernor Vinge on superintelligence

[@sagetmaster4]

Damn the game developers for AAA games have just been using the Sam Altman release philosophy, it all makes sense now

[@PrecioustheMovie1]

“Your brain just sends a signal saying I understand it” Dave, you drop wisdom that it takes most people decades of self reflection to reach. I’m digging it. You truly embody your Star Trek ideals.

[@NScherdin]

The embodiement argument is just people moving the goal posts again.

[@othercryptoaccount]

GPT-3.5 is AGI if you ask most people in 2015 Its not the narrow AI models that we've had. It can do many many things. But then people were like "it can't do ALL the things and do them as good as a human" If AGI meant human level intelligence why do we not just say human level intelligence or "HLI" As soon as AGI gets to todays goalpost of what AGI is I would consider that the beginning of ASI.

[@jacobnunya808]

I think it is because most people define AGI as having the mental abilities to do most jobs people can, or at least the average person. Being able to explain the general steps to do something over text is very different than actually being able to carry out those steps, especially in real time.

[@I-Dophler]

AI is similar to Matt Damon's character in "The Martian," confined within a jar, yet capable of extraordinary feats despite its limitations. Like Damon’s portrayal of an astronaut stranded on Mars, the AI operates in a controlled environment, relying on data and systems to achieve remarkable results. Though it lacks physical presence outside its constraints, much like the astronaut dependent on technology to survive and adapt, AI can still solve complex problems and contribute significantly to advancing human potential. It’s a blend of isolation and ingenuity, showcasing just how much can be accomplished with the right tools, even in a confined space.

[@Danuxsy]

none of this is true for LLM's

[@I-Dophler]

@Danuxsy LLMs can indeed operate similarly to confined systems like Matt Damon's astronaut in "The Martian." While LLMs don't have physical presence, they still leverage vast data inputs to solve complex problems, much like the astronaut using tools and technology in isolation to survive. The confinement does not limit their ingenuity or usefulness, especially in tasks like scientific modeling or software development. The Martian analogy holds, as both thrive within their boundaries through resourcefulness.

[@Danuxsy]

@@I-Dophler Yes but that is an illusion, that's the entire problem with LLM's and has been a problem since the beginning, it's the elephant in the room. LLM's cannot understand, they do not know right from wrong.

[@Katatonya]

@@I-Dophler Did you talk with LLMs so much that you write like them? Or was that written by one?

[@I-Dophler]

@@Katatonya As you are human, you sound just like one. AI doesn't take away human expression, it amplifies it.

[@bigbadallybaby]

I think Orion will be a LLM trained on a massive amounts (10x gpt4) of synthetic data (made using GPT 4 and strawberry). Clever techniques will have been used to ensure the data is accurate and varied (exploring every topic from multiple angles, explaining things from 10000 of viewpoints ).

[@szlagtrafi9115]

So we may consider as something similar to Intel's tic-toc cycle but here tic is the model training and toc is refining it to be used in a way similar to o1. Than this new o1 is used to generate synthetic data. But I wonder if such a trick is enough to get through another tic-toc cycle. It is quite possible that Orion will be so good that it is almost pointless to train something larger. And we will be stuck, ie. AI winter may come and without a significant breakthrough no new progress will be achieved. Still, we have a few years of exploiting these ideas to exhaustion. And maybe just maybe these new models become so good that they figure out another breakthroug for themselves...

[@szlagtrafi9115]

Shit, Dave just said exactly what I wrote. Note to self: watch the whole video before posting comments :).

[@VraserX]

They already revealed that they have critique gpt to verify synthetic data. Orion and OpenAI o2 is going to be insane. With agents they are AGI.

[@bigbadallybaby]

@@szlagtrafi9115 I think if this approach does start leading to diminishing returns and a “wall” is hit then I think the only avenue left is embodied ai learning about the world (physics, movement, warmth, light etc. ) through robotics- this will be slower but maybe only a delay of 5 years ….

[@justinv6132]

@@szlagtrafi9115I think at a certain point it’s limited by how fast we can give it the tools to figure new things out.

[@angelstyro]

When Sam Altman said that ChatGPT 4 was the stupidest model we'd ever have to deal with, and that it was mildly embarrassing at best, it was obvious that they already had much better models and had already had those models for a while - based on what happened before and after he was fired. And the fact that they've been showing their latest to the government says the same thing. People are critical of 'Strawberry' but are they taking into account that it is probably rather stupid compared to what they really have up their sleeves? Are they complaining that these models aren't being released fast enough?

[@criticalmass1884]

I didn't realize this genre of videos existed! I have been thinking about this type of video content for a year. I think you're in the right space. This will become more and more important as time goes on, I think.

[@lostinbravado]

Don't bet against Ray Kurzweil, unless he's not being optimistic enough.

[@coder5199]

For those who believe David Shapiro's prediction of AGI. Ask yourself this: Can GPT-5 drive my car? Can GPT-5 cook my meal? Can GPT-5 wash my dishes? Can GPT-5 build a house for me?

[@lyndonsimpson1056]

@@coder5199 it's all buiseness. He's advertising a book. Always be carefull Ith people that have money to gain from you believing them. Julia Mccoy has a similar channel where the predictions are completely out of touch and she needs that for people to buy her book/courses.

[@screwsnat5041]

Great question and most people would say most humans can’t do all three . But atleast we can learn with one hardware . We don’t need £2billion to learn how to cook 😂

[@coder5199]

@@screwsnat5041 Haha. If we can't do all three GPT-5 certainly can't. Great point we don't need billions of dollars to just run an LLM 😂

[@ecoandrei328]

I will mark it in my calendar 1 year from now to laugh at all of you.

[@lyndonsimpson1056]

@@ecoandrei328 same lol. I can't believe the youtube comment sections on some channels.

[@FourWaveforms]

​@@lyndonsimpson1056Pretty much no one is talking about what a s**t show it would be to have AGI at scale. "I can't wait for UBI!" say people who think any government is ready for this, when in fact there is no plan at all. And by the way, where is the electricity going to come from to deliver AGI at scale? Maybe it's in someone's lab in 2027, but then every business on Earth is going to want it so they can fire 100% of their workers.

[@sirinath]

When it surpasses humans, the accuracy is paramount, as even expects will not be able to validate certain part of the answer. When these are used in critical systems one slip can be disastrous with large losses in terms of money and lives. To realise the economic benefits we have to get the accuracy and correctness really up where it matters. Any increase in intelligence should be backed by a much larger leap in accuracy as it becomes harder to verify the accuracy. Accuracy is paramount when answering question on: Banking finance and risk, Engineering, Medicine, etc.

[@antonystringfellow5152]

I have to disagree just slightly... Accuracy is paramount if the model gets to make the final decision, however, a flawed model can still be a very powerful tool when used to make suggestions, in the role of a consultant. There are already a few examples of this, where o1preview has made good suggestions that the specialist using it hadn't even thought of, not to mention the comparative speed, which itself can be critical in such situations as medical emergencies. In this respect, the AI is a very powerful tool but still just a tool, not a controller.

[@sirinath]

@@antonystringfellow5152 Many many fields some have a speciality. When answers cross cut multiple areas no one person can verify it. Also inter dependencies are hard to spot. If the answer falls with the preview of an average expert where most ca verify it is OK to be in accurate. When models get more complex, the importance of accuracy increase. I nearly used results which look accurate for a critical project to luckily notice an error. To increase personal productivity and being competitive one needs to use AI and when what is possible is pushed the time, ability and resources verify thoroughly goes down. So the onus is on the AI creators to make the solution robust and accurate to avert any disasters.

[@sirinath]

@@antonystringfellow5152 Also sum suggestions are not found in pear reviewed journals. In such case you don't know if this is bull or not. In such cases it need to prove what it says is correct.

[@renatonascimento9306]

I am not so sure. Humans in these fields have been fairly inaccurate. The way we have dealt with human error is that no single flaw should break the whole system, there are always checks and balances. In medicine, no single study (unless very strong supported) changes actual clinical practice, rollouts of new practices and treatments are watched closely and monitored by different govt agencies and medical societies. In engineering, there are redundant systems, overprovisioned stuff, tests, norms, checks and even audits. In the aerospace field, incidents are reviewed by at least three different entities. These areas aren't failure-free, they are just built to be resilient to failure.

[@sirinath]

@@renatonascimento9306 In many of these cases the end user is in a position to understand and apply these. If a single average partitioner can understand and verify an AI solution before applying it, then it is OK that there are are errors. Then if something slips through, it is the negligence of the user applying it. Accuracy becomes paramount when experts can verify only parts of the answers and it becomes infeasible or uneconomic to verify an answer because each answer requires multiple expects in different specialities. Currently there are journals and text books which go though a lot of vetting. When AI produces answers and inferences which are found in neither without holding the process which these were to a very high standard is where the problems start.

[@lexington.]

The collective human intelligence argument is hilarious to me. A person is smart, but people are dumb, panicky animals. We, as a collective, cannot hold a candle to an AGI that will have perfect recall, works faster, and doesn't need to worry about disagreements.

[@IA_Latinoamerica]

I still stand on the hill that your prediction was correct. Orion is coming, and that is just the public release. High compute is just a limit for mass serving. Q* was leaked one year ago. And o1-preview uses GPT-4 level tech, which is two years old. If there is any advantage for AI development that can be taken from these models, OpenAI is WAY ahead than what we imagine.

[@Mephmt]

I had to pause the video to say that I'm loving the Pathfinder plug! At time of writing, this video has only been out for 2ish hours and has over 5100 views. Great work reaching more!!

[@_bassmentdweller]

Look Dave. I’m a union worker. I am concerned about my job and the prospects of my children. I see the potential of post work economics, but we have to get the society of every country from A to B. There wasn’t a single mention of Artificial Intelligence in the US presidential debate. I have absolutely no confidence that the AI entrepreneurs are going to “take care” of us until post labor economics are tolerable for the people. If the government and the entrepreneurs have no defined plan for the world’s population as this future barrels toward us, then general strikes are guaranteed. I’m not a ludite. I want to live a decent life. Let’s not belittle union workers for demanding a future that includes them. Wouldn’t you while you still had power?

[@DaveShap]

There actually was. Only Kamala mentioned AI and quantum computing. Anyways, I didn't "belittle union workers" I just said they'd fight tooth and nail to resist change. Neo-Luddites are another faction.

[@I-Dophler]

It’s remarkable how near we are to reaching AGI, yet the true challenge is in how we manage and shape its influence. Like all groundbreaking innovations, it demands vigilance, foresight, and a thorough grasp of its wide-reaching effects. We must proceed with caution, making sure that it remains a tool for humanity, rather than allowing ourselves to be guided by it.

[@Voorhees94sg]

I saw this title and thought: "Damn, the first thing I have to do when I get home is watch this video."

[@azhuransmx126]

The Current Semi AGI creature with 120IQ is a Genius inside an Egg 🥚it still needs senses (Multimodality), arms and legs (Robotics) to go outside 🐣🐥🦅 and start with the last stage of the Take Over Process.

[@Danuxsy]

No we are not close to an AGI, you are simply delusional. There is nothing about GPTo1 that should give you the conclusion that we are close to AGI, if anything it should do the opposite.

[@cbnewham5633]

It can't do maths. It's not anywhere near AGI.

[@The_Long_Bones_of_Tom_Hoody]

Im not sure whether Dave is fresh thinking and open minded, in respect to his seemingly changing views on things. Or whether he just all over the place....

[@cbnewham5633]

Just all over the shop I'm afraid. I just think back to that woodland walk and his claims of being able to simulate the human body at the molecular level. I've no doubt David is very intelligent, but some of the things he sometimes comes out with are just jaw-droppingly silly.

[@gabrielmestre3623]

I think OpenAI is still holding back something much bigger than the model they just released. In my opinion, they are about a year or more ahead of the competition, as they have only recently reached the performance level of GPT-4. Knowing this, they are not at all concerned about releasing new models or features anytime soon. They launched an "Omni" model to surpass the competition. The competition managed to catch up with that model, and now they've launched a reasoning model. But none of this would scare someone working inside OpenAI on a daily basis, as they were already aware of these advances and knew this would happen. Now, combining: 1. Agents (a base already launched in GPTs), 2. A multimodal model with audio, image, and video, 3. A reasoning model, 4. Long-term memory and planning, All of this together in a single model, which I believe could be a GPT Next, a full Orion, or whatever name they choose to give it. That, indeed, would be something that could surprise someone like Ilya.

[@jeffsteyn7174]

We wont recognise that AGI has arrived. Goal posts for agi keeps getting shifted. Next we going to need it to fart ice cream for it to be agi.

[@paprikar]

People still nearly don't know what consciousness is and using this AGI/ASI categorisation system, no shit. And that's actually concerning, since delusion is definitely not something that you want to see in relation to such tech

[@Nanohamage]

what people rly care about anyway is jobs and there is no moving goalposts there once people start losing jobs then they'll accept it

[@BusDriverGames]

I wouldn't consider anything AGI that has to be pre-trained only. if it can't be shown something new and learn how to do it on the fly and then apply that knowledge going forward across different situations then I don't think it's actually AGI.

[@dante0817]

read again the definition of AGI, you missed something.

[@null7936]

Sakana AI from Japan.

[@othercryptoaccount]

Oh look we're moving the goalposts again

[@BusDriverGames]

I said I wouldn't consider. I didn't say that is the definition. It's my opinion.

[@othercryptoaccount]

@@BusDriverGames I know but AGI definition has been pushed back so many times what the hell is the point of the term? We should come to an agreement on what it means not push the goalposts even more

[@Matt97554]

GPT4o was a proto-AGI, now the 'prewiev' O1 version shows clear signs of progress especially on physics, programming and chemistry problems. I have seen several videos in which a physicist tested the model ON A CLASS OF PROBLEMS NOT PRESENT ON THE INTERNET and the results were nothing less than incredible: about 2 minutes to solve them. What the O1 model seems to lack is abstraction capacity, yet it manages to generalise well enough over a huge class of academic and PHD problems. I am curious now to see the full model capabilities. The future is clear, we are getting closer and closer to an AGI. If not by the end of this year, by 2026.

[@phen-themoogle7651]

I’m also in the by/in 2026 camp 🏕️ Hopefully it’s life changing enough so that UHI(universal high income) comes🤞or makes it easier to survive for struggling people

[@Katatonya]

Naive people keep saying this is all still a "next word predictor" and it will never truly be able to reason. What do you say to that? Imo they're completely misleading themselves and not thinking critically enough. We as humans, when we talk, can also be called "next word predictors", that means nothing. Before we, and before an LLM predicts anything, it has to go through a black box, the MLP block after Self-Attention, we've no idea what happens there, could very well be an attempt at reasoning, where future models might be true reasoners. Same as we, before predicting what to say, are also going through our own black box, which we barely scratched the surface on how it works.

[@John-il4mp]

Great video, Dave! You're truly one of the best on KZbin when it comes to talking about AI. Congratulations on your success! 🎉 Keep up the amazing work!

[@nathanielacton3768]

I was an decel, but I'm moving to neutral. Not because I think AI developments are more valuable but because I'm now able to see what the adoption model consists of. We're not talking 5 years the world changes. We're going to see concurrency for decades and tons of implementation restrictions. I work in Ai implementations so know these first hand. So, we don't need to deliberately slow things down. If you came out with an army of 200IQ models next month people like me would still have to deal with impossible to solve problems. Put it another way, if a 200IQ person walked in to a company would they really change much? It would be a flower in a garbage dump.

[@justinv6132]

Dude people forgot what exponential meant. We might just be a little earlier than we think. But this feels like one of those first big jumps.

[@Rox_Pyros]

Real

[@RezaDarehshooti]

for years, AI scientists have promised us that the advent of AGI (Artificial General Intelligence) would lead to an intelligence explosion and the onset of the singularity. However, with the advent of ChatGPT, which is often referred to as AGI, neither the intelligence explosion nor the singularity has materialized. So, what exactly is AGI? AGI represents a system with an extensive array of skills-ChatGPT, for instance, possesses a broad understanding of multiple languages and extensive scientific knowledge, surpassing many individuals in those areas. Yet, despite its vast array of capabilities, it falls short of inventing novel concepts. This discrepancy has led some AI researchers to revise the definition of AGI, but is this the right approach? We have indeed made strides in creating sophisticated AI, but the kind of AGI we have now does not seem to foster the intelligence explosion we anticipated. Perhaps what we need is ACI (Artificial Conscience Intelligence) or a different paradigm of machines to ignite the intelligence explosion we’ve been aiming for.

[@justapleb7096]

"neither the intelligence explosion nor the singularity has materialized" bro relax, its not even 2 years after chatGPT was released on November 2022.

[@paprikar]

Or just use Strong AI / Weak AI categorisation system. I still do not understand why not use this one, since it is actually focused on the internal part of the system

[@christiandarkin]

to paraphrase you,: "embodiment" can be re-interpreted as "embodiment within an environment" - and that environment can be a virtual one - and any stream of numbers can be an environment if it allows you to make predictions and compare them to your world.

[@DaveShap]

it's just contextual data, yeah

[@alexzarifeh8848]

What moved you from being initially dismissive of O1 to within 48 hrs declaring that AGI is nearly upon us? Struggling to understand this polemic shift?

[@ct5471]

Regarding statutory friction, regulations etc. I would argue that we have a static force, regulation, vs. an exponential (at least for the foreseeable time) development. The longer regulators try to maintain the old system the larger the pressure buildup. They either continuously adept or try to maintain the status until the dam eventually breaks. If another nation introduces full speed automation and post labor economy earlier its economic power might threaten to crush all slower countries. Without maximum automation (relying on humans) your productivity will fall behind, without post Labour economics mechanisms in place, your demand drops. Both to the advantage of more progressive counties.

[@VadimDumbrava-u5r]

I’m a business owner of two clinics and I use chat gpt all the time whenever I need advice for business decisions. It’s mind blowing, I use it like it’s a business consultant

[@dustinbreithaupt9331]

I work in construction as a bridge inspector. I think construction blue-collar work will be incredibly hard to automate for a long time. I absolutely think it's possible but many other jobs will succumb before say an ironworker or carpenter on heavy civil bridge projects. Many industry tribal knowledge nuances that are not easily acquired, except through years of experience.

[@DaveShap]

I suspect soon you'll have a bunch of robots to help you gather data, but yeah eventually the robots will take over

[@Katatonya]

For sure. I say in case the UBI reality doesn't hold true, begin stacking a large amount of backup cash. You'll be one of the ones that will have the longest opportunity to prepare. My IT job will be automated much sooner. Could all be in vain if something like UBI comes to fruition. But better safe than sorry.

[@dustinbreithaupt9331]

@@DaveShap I sure hope so. With that said, I could see inspectors like myself being replaced with just a camera long before the workers are. Either way, exciting stuff.

[@frankbeveridge5714]

There's a lot of talk about chain of thought reasoning. When we have "chain of decisions," then we will get closer to machines performing well in construction and thought leadership roles.

[@NirvanaFan5000]

The downside is that if even just 5% of people who lose jobs to AI move into the construction industry, that can create an oversupply of labor and wreak havoc on wages. (Not to mention that this will be happening in many countries simultaneously.) Also, whatever industries are hard to automate and require costly labor will also have huge economic incentives to automate it. The only safe industries are small, niche, complex, and inexpensive. And once we get ASI, all bets are off.

[@PakistanIcecream000]

I believe the idea that having a body is necessary to be intelligent comes from the belief that a body can appraise reality and the world better than an isolated brain without a body.

[@Williamsl99]

I think it comes from the same part of the brain that say's, "Humans are the only animals that use tools...." when there are so many examples of birds building nests and animals building tools. Humans don't like competition. We believe we are so special that it's not possible for anything to be better than us.

[@CodexPermutatio]

@@Williamsl99 I don't understand your point because we (humans) are not really different from other animals (crows, for instance, are intelligent embodied beings just like us). Having a body helps to understand the physical world better just interacting with things. It's a learning tool that also helps to create some "common sense" of the world.

[@paultparker]

@@CodexPermutatio yeah, I think this is interesting. Embodiment enables the rapid performance of experiments to test hypotheses. This enables rapid growth in practical knowledge.

[@Williamsl99]

@@CodexPermutatio AI can learn from videos and inside 3D simulations. With robots and vision they have a “body”. ChatGPT 4 already analyzes real time video in real time while you hold the camera.

[@I-Dophler]

AI's intelligence and wisdom are expected to far exceed any human counterpart. As these systems advance, their ability to process immense amounts of data, learn from diverse sources, and adapt rapidly will grant them a unique advantage. Unlike human intelligence, which is limited by biology and personal experience, AI can access a nearly limitless pool of information and continuously enhance its understanding without the restrictions of memory loss or mental exhaustion. This suggests that in the future, AI won’t just think faster but will also reason more accurately, making decisions with a depth of insight that no human could match. While intelligence varies among humans, all AI systems are designed to be inherently intelligent.

[@jacksonmatysik8007]

Vast majority of economically useful humans don't reason by first principles. They just apply memorized knowledge. For example how many people could actually derive Pythagoras theorem by first principles thinking? Most people just memorize it. Like DS says as long as the outputs are useful they will start become part of the economy more and more while getting better and better eventually surpassing all humans.

[@mmaslav6176]

Exactly. Most of us dont contribute anything really "original". We only do remixes of things. LLMs are already on this level.

[@advaitc2554]

Unions and other working class people will have much less resistance to AGI if there exists a much stronger social safety net that has full health care, elder care, child care, transportation, housing, K thru PhD education, full pensions, internet, basic utilities, basic income, etc. This is called UBS aka Universal Basic Services. If AGI takes my job and me and my family are thrown out onto the freezing streets to starve, then darn right I'll resist AGI. And I'll strongly resist the plutocrats who block UBS and are happy to see homeless families starve. If you're watching this video, there's some chance your job will soon get taken over by AGI. 5 to 10 years. A rising number of white collar families made homeless? Quite possible.

[@bellch87]

I bought "Heavy Silver." I look forward to it!

[@magicsmoke0]

For me, the one thing stopping o1 from being AGI, besides being agentic, is learning: humans can learn just by using their senses out in the world. I'm not talking about experiences like you say in the video. Here's an example: a student can attend a lecture at college, and after an hour (hopefully) have learned something. AI models today don't learn from the millions of prompts and chats it has daily. It has to go through a training run. That there, in my mind, stops it from being AGI. It can still be smarter than all of humanity and provide economic benefit, but it won't be AGI without being able to learn continuously, in any environment it's in.

[@denjamin2633]

That is precisely my thoughts on the matter.

[@Katatonya]

I don't think that's what AGI entails. All it entails is that it can be as good as humans at doing tasks. If it is as good as them, it is AGI. Humans NEED to evolve to be able to be good. An AGI is already "evolved" to that level, so it is skipping the evolve phase. An AGI won't need to take any classes since it already took them in its training phase. Can it learn and evolve further? That's irrelevant. Irrelevant to the term of course. Now an AGI that can also further evolve seamlessly. That'd be insane. Pure hypothesis: but what if an AGI can evolve for your use case on the spot using its context window? Say it's not up to date to "quantum metaphysics" and you need it to solve a task in regards to that, you just give it the book for quantum metaphysics. And since it's already an expert at all and every subject up until that point. Why can't we assume it assimilated that book and it's now evolved, for the duration of that context window? Which could be configured to be indefinite for your own personal agent. This issue will definitely be challenged, AGI or not, they will gain that ability eventually imo.

[@magicsmoke0]

@@Katatonya You're speaking more on the academic, which is fine, but let's consider something simpler or novel. I'll give two other examples: 1. Let's say I invent a new board game that has never been thought of before with new rules (think chess or Go). The rules are easy to learn, but playing the game at a competent level will take some practice. I'd like a partner to play with me each day and we can learn new strategies together. If o1 is my partner, it can't do this, because it never learned this game in its training run. It could possibly reason through some beginner strategies, but since this is a novel game with novel rules, it would need time to practice. AlphaGo did this, but again, it was dedicated training runs. A human can do this no problem. 2. In the future, we may have AI go out to distant planets for us instead of Humans. It gets to these new worlds and discovers very different lifeforms unlike our own here on Earth. Let's assume these lifeforms are intelligent and evolved enough to have their own society, culture, economy, etc. Assuming these lifeforms are friendly enough to let our AI hang around, how does our AI diplomats learn from these new lifeforms for us besides just recording video? Could they learn to integrate themselves into the alien's society assuming it is theoretically possible (there is a way/path, they just need to find it).

[@Katatonya]

@@magicsmoke0 Yeah I get you. I'm just wondering if it'll already be able to do that, given insane core reasoning abilities. If it's AGI level, and its IQ is say, equal to the smartest IQ recorded in humans, that's 276. What if we just task it to create a really good strategy, and you give it a day. (Now OpenAI chooses for you, and it only thinks for a minute at most). But a day? That'll be the equivalent of you yourself learning the game and creating evolving strategies, for months. And then once it found some, it stores them in the context window and uses them to play. Could repeat the process and there you go it "evolves". Though I'm just hypothesizing. We can play around with ideas like this when we get Orion or something. Or even the full O1 if it won't cost 2000 per month. Karpathy said that in the future, models might be really small, opposite to what we have now, because they'll just contain the core reasoning abilities, and anything else it can just access off the cloud.

[@Williamsl99]

These systems are learning from all the senses... Multi-Modal exists today. No one has pulled it all together yet in one AI system or robot, but it shouldn't take much imagination to see it's about to happen.

[@StevenSSmith]

looking forward to an open source "o1" reasoning model

[@UserErrorGille]

The rationales these models provide give us deeper insights into their decision-making, which we can use to enhance RLHF and fine-tune the models more effectively. By understanding the underlying reasoning, we can offer more targeted feedback and adjustments, leading to better performance and more aligned AI behavior.

[@sassythesasquatch7837]

I watched Dr Waku's most recent video yesterday and he said about o1 intentionally trying to decieve the user one in every 250 thoughts or something along those lines. Even if nothing comes of that and there are guard rails in place to stop the final output being deceptive, that type of information is going to be a huge setback for widespread adoption in big companies

[@parrowman25]

The US government will have to accept and push A.I when National Security is involved. If another country gets there first it may not be a good thing

[@TheSuperColonel]

Every big country like China and US, and Russia is likely racing toward the best AI even if just for military power. So no one wants to be left behind. Maybe we are already in Cold war 2.0 I assume because they race they will bypass as many safety measures as possible as that would slow them down.

[@noctarin1516]

I think a more fair comparison to make is Strawberry is equivalent to a mediocre but not completely incompetent grad student

[@angelstyro]

Why do we need robots for things like construction tasks? Why couldn't AGI just discover a better method that doesn't require any of the so-called human resources? And what will be our purpose when work is no longer necessary? We need a common HUMAN GOAL. Very soon, the human race will no longer be striving for something better, for something more. I can't think of a reason to improve myself if a machine can do everything I can do, but thousands of times better, I don't have to learn a language, I don't have to do my own homework, I don't have to walk or talk because machines can do those things for me... Is this really the world we want? What will happen to us??

[@Mephmt]

Dude, I haven't been this excited in 9-10 months... I'm tampering my hope with this though. Like you said, we just have to wait and see.

[@TonyDiCroce]

I dont think its agi. I believe its agi in a few dimensions but it is also missing a bunch of dimensions. I think its missing a world model.. and its missing emotional intelligence. I think it'll get there but its not there yet.

[@pierce2001]

The first company to fully intergrate agi will be incredibly successful no matter what business it is. When that happens full adoption will happen.

[@Juttutin]

Abilities required to be considered decently intelligent by me that no one (including David) ever seems to talk about: 1. Able to identify when it is making stuff up (hallucinating, imagining, fiction, assuming) vs knowing or remembering. 2. Able to react to interrupting additional input while working on previous input (closely related to embodiment). 3. Able to reach out and request additional input (clarification, stuck, uncertain, low confidence) which feels closely related to 1 and 2 All of these are things pets do to some extent. All of these are problematic to some extent for all humans, but are correlated with how smart (holistically) we judge other humans to be. On 1. the extent to which 'hallucination' and 'imagination' are the same thing is fascinating to me. I suspect they are essentially the same concept. On 2. It's the opposite of intelligence to be fully and totally closed off to interruptions (which may be helpful) while doing thinkwork. Conversely, intelligent people actively limit interruptions (which are generally unhelpful) while doing thinkwork. On 3. Smart people are very good at identifying (or perhaps balancing is better) when they might perform better shifting from imagining and creating to seeking further input to avoid heading down rabbit holes and other wasted effort leading to failure or subsequent rework.

[@h-e-acc]

Simply put, we need to replace our politicians and government with AGI and ASI. We would be better off that way. And those with our cause need to be enhanced the sooner, the better.

[@zvorenergy]

The problem is you think the corporations that own and operate the data centers will be on board with that. They have very cozy relationships with the government.

[@fatboydim.7037]

AGI is almost here: Time to get Gary Marcus on here.

[@johnthomasriley2741]

Now we need a prestige team with an AI member to develop an action plan for our climate crisis. Say in one week. After that the process must be repeated for all major AI improvements. The plan must include: (1) what to do to address the physics, (2) how to enroll the general public, and (3) how to deal with the deniers. The results will be judged on new good ideas and excitement generated.

[@neptunianman]

Seriously underestimating how easy it is to roll out fully automated alternatives to human labor in our economy. Self checkouts have been around since the late 80s, all the hardware and software is here and is manufactured and developed at a massive scale, but it's still took such a long time to catch on and even now there are humans staffing the majority of tills. Even if the software is 99% perfect, that 1% in which it doesn't succeed is going to make it, at least in the near future, more financially viable to use human labor, and that goes for 0.1%, 0.01% etc etc. That's not to say it won't ever make an impact, just not in the near future.

[@keeganpenney169]

Really enlightening video today david, thanks for sharing the hard work!

[@blackup4]

So 4 days ago strawberry was “unimpressive” and something you did 18 months ago, now it’s the harbinger of AGI? Pick a lane dude. Easy to look like a prophet when you offer up every possible prediction

[@PakistanIcecream000]

I want to know why David Shapiro is crying out for the arrival of Artificial General intelligence?

[@DrEnginerd1]

People trying to define AGI as something that needs consciousness are missing the point. We need to look at this Pragmatically, as in, what are the end results. If the end result is that it does stuff most humans can do in their daily life then its AGI. If anything, it being totally unconscious would be better because then you wouldn't need alignment at all.

[@FourWaveforms]

AGI need not be conscious. It need only predict the behavior of other things which are conscious.

[@phen-themoogle7651]

Thanks for your video! This relieved me of some anxiety for some reason. Have a great day 🌟

[@georgytioro]

There is no AGI until fundamental changes are applied! We can yell wolf wolf.. as much as we want

[@chiasm-vj8vn]

Outstanding (as always), Mr. Shapiro!

[@ivarborthen7320]

Hi Dave! If the quality of the O1 model increases when they let it think for longer, wouldn’t OpenAI be better off not making their next model public? Instead, they could use all their resources have it to help them with the next breakthrough/optimizations by letting it think really long. That way, they’d be the first to AGI+ and would control everything. It seems like a poor strategy to spend compute on subscriptions at this point (I’m assuming Sam is aiming for world domination and might not be too concerned about safety).

[@Jwoz7]

Congrats on releasing your book, Dave!

[@coder5199]

My take on this is if an AI can only do virtual-world tasks and not real-world tasks, it is not general enough to be considered "Artificial General Intelligence".

[@FourWaveforms]

It's also nowhere near replacing programmers, unless all programmers write trivial barebones apps. I've seen zero evidence it can code anything beyond that.

[@coder5199]

@@FourWaveforms Exactly! On top of that, it's nowhere near even a self-operating computer. It can't even use a mouse and keyboard like humans do. I doubt GPT-5 is able to do that.

[@MichaelLaFrance1]

If open source models remain 6-12 months behind foundation models, that's not bad at all, considering how amazing the tech is.

[@I-Dophler]

You could think of "embodiment" as just being part of a setting, and that setting doesn’t have to be something physical-it can be virtual too. Basically, any stream of data can work as an environment if it helps you make predictions and check them against what’s happening.

[@ChadZLumenarcus]

Everyone is always, 100% logical. The problem is the information they have has them making different conclusions. Calling people "emotional" doesn't solve the problem of disagreement, it only makes you look like the bad guy who refuses to understand their risk management concerns. If one person says the bridge isn't safe, the other says it's fine, if you want to determine the state of the bridge you have to not just test the bridge, but also figure out how they came to their conclusions and solve that problem. The goal is to get everyone up to date on proper information so we can make the best decision forward. Arguments and conflicts lead to lose lose scenarios.

[@FourWaveforms]

People are not always 100% logical. If they were, the world would be atrociously boring. You are right in criticizing him for calling people "emotional." Governments have no plan, and have barely learned how to stop their clocks from flashing 12:00, and the AI companies are certainly not offering to pay for the livelihoods of the people they will make obsolete. It is unbelievable to imply that there's anything wrong with people saying "okay, but I don't want to eat rats and sleep on a slab of cardboard under the freeway."

[@ChadZLumenarcus]

@FourWaveforms Logical doesn't mean correct. If you have bad information, you're logical, just wrong. If you're being malicious, you're still logical, with malicious goals. Emotional behavior is our safeguard to the reasoning we have against threats. Our perception and understanding baseline our logical behavior. The only reason why I'm pushing this narrative is because from what I found, it helps direct the conversation towards the variables that need to be discussed and figured out. The reason behind the conclusion. Agreement due to understanding and a scientific method is boring. That's why people who commit fraud succeed in controversy and controlling the narrative. Which is what we have to work against.

[@JournalofSean]

Congratulations on finishing your novel!

[@DeimosSaturn]

Well, you know how the singularity works, David. One moment, it's "almost there", and the next moment it has sprinted ahead of us. There will never be any such thing as "AGI". It goes from "almost there" to "god-like" and we probably won't notice for a while. Is this months away or hours away?

[@maureenboyes5434]

😊 I am stunned! I never thought I would live long enough for either S or I.

[@idontexist-satoshi]

I see your ego's calmed down now that you’ve been humbled on X. Keep it up-some of your ideas are solid, but you have to consider that for big companies, product and profit are the main priorities. Unfortunately, open-source is going to fall even further behind by February 2025. It’s already lagging by a year and a half, and no one has figured out how we trained o1 even if they like to think they have.

[@torarinvik4920]

Why doesn't anyone talk about the fact that these LLMs does not have continuous learning. It's not close to AGI unless they have this, memory and autonomy.

[@thedannybseries8857]

I do think AGI has to be agentic. And it has to be able to learn, continuously so, with minimal to no human supervision.

[@torarinvik4920]

@@thedannybseries8857 If find it odd that people rarely talk about it whenever AGI is mentioned. The ability to learn and adapt is what made humans so dominant.

[@FourWaveforms]

That, and LLMs don't seem to be wired in to the most basic, dirt-simple strategies. Even modern ones have to _predict_ the number of Rs in the word "Strawberry" because they can't just _count them._ And they can still get it wrong!

[@torarinvik4920]

@@FourWaveforms Exactly, O1 is really impressive and awesome. However if it cannot learn, it's not close to AGI. Pretty crazy that people even claim this.

[@paultparker]

I’m not so sure this is a downside. Yes, the AI doesn’t learn until the next training run but when it does, it incorporates the experiences it has had from thousands or millions of interactions. So the cycle time is longer, but the total amount of knowledge acquisition would be staggering.

[@lakshyatripathi8624]

AGI is already here. If we think about it, models like o1 answer almost anything at a level of experts or highly trained people in that field, even better sometimes. It's just that the form is not what most of us expect. We could get to see it in near future, or it might need some more time. We have to wait for AGI in a form we more expect, but AGI per se is already here in a sense.

[@cbnewham5633]

No it is not.

[@KraszuPolis]

I think what it lacks to be AGI is ability to learn. It seems impressive sometimes but sometimes fails off at pretty simple task, and lacks ability to learn itself how to solve it.

[@perrier1986]

machine learning street talk has a Deep and balanced view on AI.

[@markkuykendall5475]

The worry about "red tape" is probably a bit over done. It'll be less of a problem than you think, Dave. The reason I'm speculating this is because the money to be made is so huge, that, like AI dev so far, the adoption rate will be driven by a Moloch-y dynamic. If the tech can be implemented in a given area and ISN'T because of red tape, then the jurisdictions with the red tape will be economically crushed by those without.

[@blindfaith8777]

I think it might be a good idea to create basic AGI levels and tests. It will help breakdown progress and help us avoid being weighed down by other people's varying definitions. I put one together as an example but it could probably use significant improvement. AGI Levels (Each level includes the previous) *AGI-0* Interpretably answers questions in text form and remains vaguely on topic. *AGI-1* Correctly answers most questions that don't require deep multi-step reasoning. *AGI-2* Capable of answering questions with deep multi-step reasoning. *AGI-3* Engages in deep multi-step reasoning and relates it to images or video. *AGI-4* Recognizes and requests missing information. *AGI-5* Proposes tests to generate data for missing information. *AGI-6* Interprets data generated from tests. *AGI-7* Performs thought experiments or integrates interpreted data into deep multi-step reasoning. If it is connected to the right equipment it should be able to perform novel closed loop science. This is also where AGI intersects a significant amount of narrow AI. Perhaps this is true AGI. This level also includes postulation and axiom development. *AGI-8* Store learned information for long term future use (online learning). This level should be human level or better in all ways related to reasoning. *AGI-9* Ability to play any video game (like portal) *AGI-10* is anything beyond our comprehension and where the scale must be constructed by the AGI itself. ASI. (Stockfish vs AlphaZero) Some Tests AGI-0 Ask it the heart rate of a lizard, and it responds with sentences about hearts or lizards. AGI-1 Ask it where the Eiffel tower is, and it responds with Paris AGI-2 Ask it a known hard physics problem that requires deep derivation, receive proper answer (Ideally with steps shown) AGI-3 Show it an image of a clotted artery and ask why, receive an answer that relates the condition of another organ to the cause of the clot, and explains the chain of events that is occurring. AGI-9 Have it beat Portal I'd argue that on this scale we were at late AGI-1 with GPT-4o and now early AGI-2 with OpenAI o1-preview. Interestingly, conceivably there could be intelligence beyond what basic ASI can interpret, so we could also generate another intelligence scale (IS) of IS-0 (AGI), IS-2 (AGI-10/ASI), IS-3 (ASI-10), etc etc., where each intelligence is beyond the grasp of the previous.

[@blindfaith8777]

GPT-4o Revised AGI Levels (Each level builds upon the previous, with increasing complexity in reasoning, interaction, and capability.) *AGI-0: Basic Interpretation and Relevance* The system can respond to questions in a way that is readable and vaguely on topic, but it may not be fully accurate or logically coherent. The focus is on producing interpretable text, even if the connection to the question is loose (e.g., asked how the cold virus replicates, responds with information on the number of deaths from the cold). *Test:* Ask a broad factual question, receive a vaguely related answer. *AGI-1: Factually Correct Responses* The system is capable of providing correct answers to most straightforward questions that don’t require complex reasoning. It can rely on a large knowledge base but doesn't perform deep logical operations. *Test:* Ask a factual question (e.g., "Where is the Eiffel Tower?"), receive a direct and accurate answer. *AGI-2: Simple Multi-Step Reasoning* Capable of handling questions that require basic logical reasoning, but limited to shallow reasoning paths (e.g., can solve simple arithmetic or single-step cause-and-effect problems). *Test:* Ask a multi-step logical question (e.g., simple word problems in math), receive a correct response. *AGI-3: Advanced Multi-Step Reasoning* Handles deep multi-step reasoning problems that require layered logic, abstract thinking, or deriving conclusions through more intricate cause-and-effect relationships. *Test:* Ask a complex question requiring a multi-step solution (e.g., "Derive this physics equation"), and receive a solution with proper steps. *AGI-4: Interpretation Across Modalities (Text, Image, Video)* Can apply its reasoning to visual or other sensory inputs (e.g., images, video) in addition to text. It can reason about and interpret information across different types of data. *Test:* Show a medical image (e.g., of a clotted artery) and ask for a diagnosis; receive an explanation of the condition with potential causes. *AGI-5: Information Gaps Recognition and Inquiry* Able to recognize when it lacks sufficient information to answer a question fully, and can request additional data or clarification. This demonstrates a higher degree of self-awareness and problem-solving skills. *Test:* Pose an ambiguous question that lacks key data; it responds by asking for missing information. *AGI-6: Scientific Reasoning and Experimentation* Proposes and designs tests or experiments to gather data and fill in knowledge gaps. The system can identify gaps in its own understanding and suggest ways to resolve them through empirical means. *Test:* Ask a complex scientific question; it proposes a testable hypothesis and experiment to gather data. *AGI-7: Data Interpretation and Integration* Able to analyze data generated from experiments or other sources, draw conclusions, and integrate this data into ongoing reasoning. It can perform thought experiments and synthesize data into larger knowledge structures. *Test:* Ask it to interpret experimental data or conduct a thought experiment that incorporates this new information. *AGI-8: Long-Term Learning and Memory* Capable of storing and retrieving learned information over long periods (online learning) and applying it to future situations. This level marks human-level general intelligence or better in reasoning, memory, and knowledge application. *Test:* Ask it to apply knowledge learned in a past scenario to a new problem or situation. *AGI-9: Complex Problem Solving (Games, Dynamic Systems)* Able to engage in and solve highly complex, dynamic problems such as those presented by video games or real-world interactive tasks. This level requires adapting to new rules and environments in real-time. *Test:* Successfully play complex video games (e.g., Portal) that require reasoning, problem-solving, and real-time adaptability. *AGI-10: Superintelligence (Beyond Human Comprehension)* Capable of operating beyond human cognitive capabilities and constructing its own scale or metrics for intelligence. This level represents a form of Artificial Superintelligence (ASI), where it can solve problems and create innovations far beyond the grasp of human thought. *Test:* No direct human test-this is intelligence beyond our capacity to evaluate (e.g., the difference between Stockfish and AlphaZero).

[@paultparker]

I seem to recall hearing that open AI has some different AI levels that they talk about internally? I wonder how these relate to your AGI levels.

[@paultparker]

@@blindfaith8777 I don’t believe that each of these levels requires all of the previous levels. For example,o1 can do multi step reasoning already.

[@blindfaith8777]

​@@paultparker 100% agree. It's a bit tough because there isn't necessarily an inherent order for some of this (learning persistence, multi-modality, reasoning depth). Even so I think it would be nice if we had something to break down progress and avoid ambiguity.

[@NandoPr1m3]

I'm currently trying to incorporate AI where I work. I feel like most people are slightly familiar with where AI capabilities were 6-8 months ago and have no idea how much progress has happened since then. I often hear people say "AI has problems doing that" when in fact, it's that they haven't learned how to prompt the models appropriately to maximize the output.

[@inspectorcrud]

What if we are the simulated agents being evolutionarily selected, for intelligence and safety before embodiment?

[@jondo7680]

Considering the fact that set another genocide is going on right and humanity can't prevent or stop it, it's a failed project, I would restart the simulation with a new seed value.

[@dot1298]

@@jondo7680 the „project“ has not failed - it rather fulfilled its purpose: proving that humans can‘t rule themself successfully. Soon, the original rulers will take over, when this „experiment“ is finished …

[@stevenkies802]

The scale then optimize is universal. All systems of thought, including technology, develop concepts, explore all the permutations, then optimize by streamlining, eliminating inefficiencies, and keaning in on the strengths.

[@yoyo-jc5qg]

that real-world experience will be like jumping from text models to vision models, thats the next step

[@DecentGradient]

Good points as usual David! I've been in a few conversations about whether or not OpenAI o1 is AGI. I feel people are not comparing apples to apples. This model's knowledge and ability to think and reason is "generally" on par with the average human. It has gaps that are different than ours for some obvious reasons, but generally is able to think and reason more or less how we do. The things people are waiting for such as persistent memory, the ability to continuously learn, and agency is mostly just infrastructure and tooling around the model. We don't really even need any breakthroughs in those spaces. It's mostly just grunt work to stand them up using tools we already have such as knowledge graphs, databases, APIs, etc. I feel society won't ever agree that we have AGI in the midst of it. I'm convinced we'll only recognize it retrospectively, after we have already had ASI for a while.

[@nexoofficiel7915]

Franchement j’ai hâte que les robots nous remplace au travail et que on ne soit plus obligé de travailler ,j’aimerais passer mon temps libre à jouer aux jeux vidéos,profiter avec ma famille et apprendre les sciences et technologies et peut-être un jour construire un TARDIS 😅 et voyager dans l’univers et rencontrer la fille de mes rêves sur une planète qui parle la même langue que le français 😊

[@FourWaveforms]

How is the French government's plan to deploy UBI coming along? Are they ready to go?

[@polygon2744]

Musk claims that version 2.0 of Optimus hands have 22 degrees of freedom. Human hands have 25. Actuators in this version are in the for arm as it is with humans. This should allow pretty much anything a human hand can do.

[@bradshelton2397]

Guaranteed AGI has already been achieved

[@jaitanmartini1478]

Keep hem coming Dave! Great stuff.

[@mugin11223344]

When someone is critical of AI, they are "acting emotionally" or maybe they have a point?

[@Leshpngo]

Your book is out hooray!! I'm sure it will be super dope. Can't wait to read it buddy :))))

[@I-Dophler]

Incredible moments lie just around the corner! As AGI inches closer, it's crucial that we emphasize thoughtful implementation. Ensuring proper safeguards, ethical considerations, and openness in its rollout is vital. The learning curve will be challenging, no doubt, but the rewards on the horizon could be revolutionary. When harnessed wisely, this innovation has the potential to transform sectors and offer new solutions to some of the world's most pressing problems.

[@lyndonsimpson1056]

Mmmh. The syntax here smells like AI gen text

[@I-Dophler]

@@lyndonsimpson1056 Lyndon, I think AI’s getting better at pretending not to be AI. Soon, it’ll have us all second-guessing!