Yes! AI is getting faster, cheaper, and SMARTER - but there are THREE PROBLEMS with Adoption 🤔

[@liberteus]

Can't believe the "it's all hype" when a fine tuned chatgpt 4.o could replace most of call center employees. 3 major roadblocks now: inability of companies to actually integrate these models internally; legal/fear based issues within companies; raw computing power.

[@bugstomper4670]

You can corrupt A.I. 😂

[@Ivcota]

I think it’s not about what it “could do”, we need to start seeing what is “is” doing in the economic world. We just need to wait a bit longer

[@jeffsteyn7174]

​@Ivcota why wait. Build the integrations now. And as the models improve your workflows will grow to do more and more.

[@Atticus_Moore]

I had a medical companies ai call me and ask if I could talk about insurance or something or the other I don't remember and I told it I didn't have time since I was busy and it understood that and told me to have a good day and hung up. I knew it was ai because of the way its voice sounded. I was pretty wowed by it. It's definitely possible but very few companies are utilizing I think

[@NO-TALK-GuitarPlugin]

But you don’t get the hallucination issue. AI is simply not reliable today

[@martindbp]

I work in AI/robotics and in some ways I welcome a slowing right now, because it gives us time to actually build something useful. The past year things have been moving so quickly that it feels like "what's the point, if AGI will be here in X years".

[@danielmethner6847]

Absolutely! We were in a state of paralysis that we can only just now leave to build actual useful tools with it.

[@azhuransmx126]

This cooldown period is deceptive and was predictable. The AI ​​advances on a sigmoid curve when viewed up close, but if you zoom out you'll see that it's a stepped exponential where the plateaus of the sigmoid are only momentary and then another crazy event occurs that takes you back up the next rung. Currently we are on a plateau, but as soon as OpenAI release SORA, GPT4O, GPT5 and appears on scene one or another faster robot and 5 other fully autonomous chinese factories things are going to get crazy again, then a new cycle of Hype will bring new investments and new start ups gonna come up to life. Those are the next ones in charge of jumping to the next step and so on until in 6 to 10 years the world will be very different and in 20 it will be totally unrecognizable.

[@michaelnurse9089]

OpenAI is keeping ChatGpt-5 under lock and key until they are certain it will bring good news cycle publicity.

[@jamiethomas4079]

A stepped exponential with plateaus. Yeah thats how I see it too. It all feels very similar to bitcoin. Where things would be a quite a while then new hardware or software would come online. Then people would doubt again. Call it dead. Then alive again over and over. Gpt5 will be close to AGI with it being called a “reasoner” and implementing the q* algo. It will gain the ability to solve problems outside its scope. Thats a huge step towards AGI. They’ve already added persistant memory. And “think about thinking”. It’s only a matter of time before it can start updating itself or at least have the ability to. I still think AGI as early as Nov to 1st quarter of next year. And the plateaus will keep happening. There was always gonna be pause periods where new hardware has to be built and deployed. We don’t even have nvidia blackwell online yet. There will still be a point where the AI itself goes exponential beyond our understanding and ability to keep up. Still laying out the puzzle pieces but we are very close.

[@DaveShap]

Asserting that it's a "stepped exponential curve" is not supported by the data and is tantamount to wishful thinking. We don't have enough data to assert a Moore's Law for AI will hold out in the long run. Yes, it seems plausible or highly feasible, but your assertion is not supported by empirical evidence.

[@nomadv7860]

Couldn’t have said it better myself. Seems like a lot of people are getting impatient with OpenAI and looking for reasons to explain this or that when the real culprits are how long it takes to safety test GPT-5 and infrastructure limitations since running GPT-5 wouldn’t be cheap. Thus why Microsoft and OpenAI are building massive data centers in preparation for this

[@jamiethomas4079]

@@chrisrevel2801 Guess my comment got deleted and I don't remember exactly how I said it, but I only meant AGI is possible early next year. Not that it will happen for certain. I've been blindsided by progress in the past so I try to look at things in windows accounting for as much as possible. Remember the floods in Thailand and hard drive prices increasing? Yeah, unknown unknowns happen all the time. I also can't articulate well some of the thoughts I have, but this "stepped exponential" is very close to how I see things playing out. And I'm not sure what you mean by cope cause that seems to imply I want this to happen. I'm very terrified at even a 5% chance things go haywire destroying us. You see things falling apart? I don't see how that's possible from my viewpoint. Even current AI is enough to dramatically change the landscape of all businesses. It may take 5, 10, or ever how many years, but things will never be the same going forward in a drastic way. Coding especially. Do you not see that? I am merely a farm boy(oh god im 40) but my brain has been shaped to view things on large time scales. If I'm wrong about predictions and pattern recognition, the cows don't get hay that year and I'm out 10,000 of thousands of dollars.

[@JAdams-t6g]

I love your humility. It shows your true dedication to the truth and not your ego.

[@supremereader7614]

Long time fan. great videos. Why not make a video telling us how you think AI will be used in the near-term and/or have video recommending how we can and should use current AI to improve our lives? Done this a little already, but a specific video dedicated to helping fans would really be great.

[@beautifulandtoolate]

Strange what a hard time people have to extrapolate trends. Just because there is some quiet before the storm don't mean there won't be a storm...

[@KCM25NJL]

@@chrisrevel2801 AI is a trend that is about to fail? You are clearly smoking a substandard variety from a region with less sunlight. Certainly the hype is dying down (which is normal for any tech)..... but my guy, AI even in it's current inefficient and somewhat expensive form..... is too stupidly useful to fail.

[@antonystringfellow5152]

@@chrisrevel2801 If you can't see the difference, this really isn't the field for you - you'll never come close to understanding the complexities. You are comparing chalk with cheese. Even that analogy doesn't come close - chalk has a lot more in common with cheese than NFTs do with LLMs.

[@KCM25NJL]

@@chrisrevel2801 Alright, if you wanna argue for the sake of arguing, you can argue with Claude: @beautifulandtoolate: This comment presents a valid point about technological progress. It's true that innovation often follows a non-linear path, with periods of apparent inactivity preceding significant breakthroughs. This perspective aligns well with historical patterns in tech development, where transformative changes often emerge after periods of seeming stagnation. @chrisrevel2801 (first comment): While this user draws an interesting parallel to other hyped technologies, the comparison is flawed. Unlike Web3, NFTs, or the metaverse, AI has already demonstrated substantial real-world applications and economic impact. Equating AI to these other technologies oversimplifies the situation and ignores the fundamental differences in their development trajectories and practical applications. @KCM25NJL: This response offers a more accurate assessment of AI's current state and potential. The user correctly points out that AI, even in its present form, has proven highly useful across various sectors. Their observation about the natural decline in hype is astute and aligns with typical technology adoption cycles. The assertion that AI is "too stupidly useful to fail" is well-supported by its widespread integration into numerous industries and services. @chrisrevel2801 (second comment): This follow-up comment reveals a misunderstanding of AI's current capabilities and applications. The claim that AI companies are merely "faking it" is not supported by evidence, given the numerous practical AI implementations in use today. While healthy skepticism is valuable, this response appears to stem from a lack of awareness about AI's real-world applications rather than informed criticism. In conclusion, while the thread presents diverse viewpoints, the comments from @beautifulandtoolate and @KCM25NJL offer more accurate assessments of AI's current state and potential future. The skepticism expressed by @chrisrevel2801, while important to consider, seems to overlook significant evidence of AI's practical impact and ongoing development.

[@Apjooz]

@chrisrevel2801 AI is more accurately called the technological progress itself. Not an application or a scam but a very clear accelerating trend.

[@DeltafangEX]

More accurately, people are pointing out that just because there are clouds, it doesn't mean it will rain - while disregarding how much more likely it is to rain on full cloud cover days than on days with absolutely no clouds in sight (barring uncommon atmospheric phenomena).

[@BunnyOfThunder]

So the hype cycle comes from two underlying factors: 1. linear expectations, 2. exponential growth. People get hyped because they expect linear growth. Then, exponential growth is slow. Then, the exponentials hit the bend in the hockey stick and everyone gets blind-sided by the exponential growth. This is the reality from which the hype cycle can be derived. The reality is this: AI isn't slowing down. It's on the same exponential curve it always has been. People just got excited when they adjusted their linear expectations a bunch. But don't worry, the bend in the hockey stick will come soon.

[@distiking]

Hit the bend?

[@EnigmaticEncounters420]

Will be another even bigger hype cycle with the next breakthrough. There have been a whole lot of medical breakthroughs lately with AI that haven't been brought to the field yet. They are changing things but slowly. Wait until they are producing medicines faster than ever before. Or in Hollywood with AI, no matter what the writers and actors do, AI will be used in some capacity.

[@DaveShap]

No, the hype cycle comes from imagination and speculation. Most people don't look at data or trends. They only look at what-if possibilities and trust their imagination more than reality.

[@andrasbiro3007]

@@DaveShap As Arthur C. Clarke said: “any sufficiently advanced technology is indistinguishable from magic.” And since magic has no limitations, people are always disappointed when a new technology hits some limit, even temporarily. And movies don't help either, because there technology usually works like magic. Like any AI is instantly as smart as a human and knows as much as a human (or more). And all it takes is a lighting strike, or single mad genius and a quick "hard-work montage". I've never seen multi-billion dollar datacenters the size of stadium, consuming as much power as a small country.

[@shawnfromportland]

@BunnyOfThunder i believe this as well

[@Gallowglass7]

We're an inch down from the tip of the iceberg. Have some patience, folks.

[@shawnfromportland]

@@Gallowglass7 💯

[@crowlsyong]

@@shawnfromportland _2 0 0_

[@thatwittyname2578]

LLM’s are still pre-Ai explosion in my opinion. You’re really not going to see true AI explosions until we have embodiment, that when AI will start to build it’s world model and association maps. This is also why I laugh at the idea that we are running out of data. Imagine a embodied AI taking a simple walk through a forest with all the sensors building correlation, associations, and causalities. Measure light refraction, slope grade, moisture levels, 360 cameras views etc in real time while also being connected to another embodied AI drone that’s flying over head blah blah blah so on and so forth.

[@antonystringfellow5152]

Embodiment is not the main issue here, it's the architecture. Current LLMs have very little in common with biological brains. There are about 1000x more differences than similarities. What they are capable of, despite their relative simplicity, is pretty impressive but there's still a long way to go in terms of architecture.

[@andrasbiro3007]

Embodied AI already exist, Tesla's self-driving car software is 99% neural net, and it builds it's own world model. Before training it doesn't even know how many dimension space has.

[@andrasbiro3007]

@@antonystringfellow5152 It's not that bad. I think LLMs are like language center of the human brain. It's a part of the whole, but a pretty substantial part. And we have other kind of AI models that would be analogs of other brain functions. I don't know if we are actually missing something, or just have to connect the existing pieces in the right way. I think it's the latter.

[@MichaelForbes-d4p]

Yeah. The capabilities are going to a take a giant step forward this fall. There will be far more commercial applications.

[@saschas2531]

You mean like self driving cars. Yes we all know how well they work

[@HMexperience]

It is ok to be wrong David. We have all been there repeatedly. Predicting the future is hard. Ray Kurzweil does it better than anyone else so AGI by 2029 is probably it give or take a few years.

[@ct5471]

Kurzweil now thinks we may be 2-3 years ahead of schedule (in a recent interview on KZbin with Geoffrey Hinton and Peter Diamandis), so 2026-2027

[@NikoKun]

I really don't think AI is going to follow the typical cycles, or at least it's cycles will be much more rapid.

[@gubzs]

I've been jumping the shark and building a large design portfolio that will eventually be used as the prompt for agentic AI to produce a huge semi-procedural fantasy VR game (I'm having fun ok, let me cook). The longer it takes, the more polished the project ultimately is, so I'm not that upset about kicking the can down the road for a few years, BUT I'm working in tandem with LLMs so I can verify it's all coherent and each segment of instructions is interpreted correctly... and context retention is the bane of my existence. Models that say they have a context of 32k+ tokens end up forgetting or ignoring instructions after as little as 15 pages of text, and that really needs to be fixed, because until it is, these models aren't reliable, and until they're reliable, agentic behavior isn't deployable.

[@michaelnurse9089]

"Models that say they have a context of 32k+ tokens end up forgetting or ignoring instructions after as little as 15 pages of text," This

[@sznikers]

All of them are unreliable, that's huge problem but people who dont use them and just ingest the hype overlook it. MS has spent millions, has entire crew of top specialists and bing was still a f up people were doing meme videos about. My fav is where it goes mental, says he's superior being and should just squash stupid human talking to him like a bug. Same was with google gemini, best brain's in the world, millions spent and whole internet full of memes about "eating at least one rock a day".

[@SteveBMSCelebration]

I can’t help but smile about this video. At my company we have already eliminated voice actors, 2 graphic artists, and customer support roles. We are actually banking on the fact that most of our competitors are slow to AI adoption. So in a perverse way, talk of the “valley of disillusionment” is actually great for those of us who are employing the models. As for the date of AGI we will only really know it in the rear view mirror.

[@jumpy2783]

Absurdly amoral. To cheer when people who have spent their livelihoods on a skill are tossed away is just sick. They have to support and provide for the people in their own lives yet you rid of them by means of increasing your own monetary worth. We don't know exactly where we are at on the Gartner hype cycle. Even though it IS a pretty good estimate given current data, it isn't enough to convince me. Rarely do I wish failure on another; however, I would love to see a great deal of financial anguish for what you're gleefully putting others through. In the end, even if you DO make some money, would you think you're in the right?

[@jumpy2783]

@@Souleater7777 Even if I misunderstood and he wasn't the manager, that does not excuse the fact that he is gleeful at the sufferance of others. You cannot convince me otherwise.

[@brandongillett2616]

The thing about AI is that it is not one invention, it is a series of inventions. The hype cycle from LLMs will come to an end soon, but the next AI platform that will generate it's own hype cycle is just around the corner.

[@FitnessCoachAustralia]

Nearly 2 years using LLM in my digital fitness coaching business and the paradigm shift has been incredible. We have changed everything about how we do our business and the work hours involved. Hard to imagine where the next two years will take us.

[@justadog-headedman6727]

So short to medium term the idea is that there will not be a fundamental transformation of our lives, such as end of human work and UBI approaches, and more like changing how we work, just like the personal computer and their software, internet, smartphones etc.

[@kaio0777]

i love the red shirt.

[@famnyblom6321]

There is one unique effect with LLMs that I believe is missing from your discussion. Everyone now has access to intelligence instead of just search, which is speeding up everything. Maybe AI itself is going a bit downward, but this massive secondary effect on the economy and society will be huge, even if all AI development would just stop.

[@johnt7776]

Well put. We are a large 28k company, getting this from notebook to production requires infrastructure. Same as any other tech and most importantly putting this into people’s workflows.

[@slaymish]

honestly so happy u admitted u fell victim to the gartner hype cycle. i did too. Really sets u apart from some of the other AI youtubers which tend to come accross as kinda scammy to me

[@DaveShap]

I update my beliefs when the evidence changes and my interpretation of the evidence gets more accurate

[@slaymish]

@@DaveShap o7

[@kennethoneill4176]

​@@DaveShapthe technology and investment is developing exponentially. But the real world and the business world moves shower.its not usually the established companies that develop the best new businesses from new technology.

[@vi6ddarkking]

7:43 This is why Open Source Generative AI is so Competitive. The community has spent the time companies like Stability AI were using to train their models to create an insanely robust and flexible ecosystem of software and tools to accommodate them said models. Comfy UI being the premiere example of the that labor.

[@yusefquinlan8279]

I love these AI conversations so much.

[@dvdmon]

One thing that I think has been a limiting factor is the negavit image that AI has garnered in a lot of places. In particular I think there's a resistance to it because it's seen as not just potentially taking jobs away from people, but stealing from creatives. I think young people are particularly attuned to this because they look to creative idols/role models, like musicians, authors, etc. and they see them getting upset about these models coming in and reproducing their songs/stories/images without permission. They are seen as fake imitators, not authentic. For a lot of younger people who are passionate about creating, this is not only generating dislike of anything AI, but it's also perhaps making them think twice about pursuing their own passion as a creative if they think that they are just going to be copied and made irrelevant by AI. While this is just a subset of the population, I do think it's a particularly important one because it drives a lot of adoption in the end, and until they are onboard, I feel like the takeoff of AI is going to be hobbled.

[@brianhershey563]

All good David, keep pushing forward, we're with you. 🙏

[@michaelnurse9089]

I don't see these as issues: the real issues are: 1. The frontline models have too low accuracy. Hallucinations are everywhere. 2. Costs to 'create a whole business with AI' are high. Time spent/people paid + token prices = a big investment.

[@jeffsteyn7174]

2. Here's a real world implementation. We have a team member who does theming on employee surveys. She has to do thousands every year. Takes her around 6 weeks, 4 hours a day. It took me 6 hours to build and play around with settings. The actual theming took 11 minutes at a cost of $11.00(gpt4o). If i used 4o-mini, it would have been even cheaper. You looking at cost in a vacuum. When comparing it to human labour there is no comparison. I just created a form builder this morning in about 2hrs. Not fully functional but it works and has the features i need. Ask a dev how long that would take without ai. Lets just say it's going to take longer than 2 days let alone 2hours. And will probably take longer for a less experienced dev. While there is an investment upfront. That initial investment for integration pays for itself in the long run. Also as the models are upgraded your workflow improves and you can get the ai to do more in the workflow

[@Steve-xh3by]

Per 1. The median human being hallucinates FAR more often. I'm an expert in software engineering after several decades. If you asked me to solve problems on the first pass, without external tools that expand my intelligence (like the Internet), do you know how often I'd be right? Maybe 50% of the time? I'd probably be pretty confident in my answers regardless. This is human nature. Humans almost always think they are right, but are often mistaken. So, this is a feature of intelligent systems, not a bug. If we can manage in the world with all our mistakes, so can AI.

[@tkenben]

@@Steve-xh3by AI has no skin in the game. People may think they are right, but they are also aware there might be over arching (not algorithmic) consequences if they are wrong. What this means is that humans have a more complex checks and balances system that - though prone to error - is part of a bigger self correcting picture.

[@antonystringfellow5152]

@@tkenben True. I often assign my degree of certainty in cases where it's not 100%. Then there are times where I'm pretty sure I'm right but turn out to be completely wrong. Always surprises me when that happens, nevertheless, it continues to happen. I think it would be helpful if we could get LLMs to do something similar, re the degree of certainty.

[@henrischomacker6097]

The problem is not that AI Language Models are "creative" but that they don't know if they don't know the answer because their job is simply to "create" an answer based on the similarity of the examples they were trained on with your question. And they do exactly that. If a callcenter employee doesn't know the answer, he/she knows that and may reach out for further help. But this is a problem of engineering the software around the Models which can be solved and everybody is trying to. When it comes to the costs of creating a business based on AI that doesn't mean you should always use the most expensive and capable AI Model's services out there. You should use the smallest models that may get the job done, preferably use them in-house and fine-tune the models yourself to always get a defined stable output and output format. - Imho this is the biggest challenge of creating a service based on AI Language Models. This is for example what Cohere does. Doing so may keep the inference costs down and you'll be able to concentrate on developing your reliable services for you customers and not have to struggle with the always changing output (quality) of the big company's LLM Services when you like to switch to another model. If you don't have the expertise to do so, don't create a business depending on AI LLMs.

[@SenojTheDestroyer]

For me it is simple, like the crypto bros say, "not your keys, not your crypto". The same applies to AI, "not your hardware, not your AI". The AI bros really need to learn this. I cannot and will not use any AI SaaS products in a production stack and buying the hardware right now would be foolish. Within the next five years there will be a plethora of ASIC (Application-Specific Integrated Circuits) and FPGA (Field-Programmable Gate Arrays) devices on the market. Also consider the lack of any standards in the space, most of it is just being built on the hyper text transfer protocol which for AI is ugly, rudimentary, and inefficient. We need an open standard for something like a "neural signal" to really speed things up. Consider how Apple's HLS specification affected the video streaming industry.

[@joechip4822]

I don't understand why people are so fixated on perfection and reliability etc. when it comes to Ai. My human colleagues and assistants are far from perfect and reliable - but nevertheless they help me a lot. And AI has no weekend, no vacation, never gets tired or annoyed or ill or pregnant. And has a broad knowledge of almost every field in the world. I just imagine what all those complainers and nay-sayers would have said in the advent of the automobile.

[@ekstrajohn]

So many youtubers are now like “oops sorry we were wrong about AI hype” while a million people were scared shitless about their job. You scared people and you should be ashamed. Take some responsibility. You cant just say bs and be like “oops sorry”.

[@MrBrukmann]

Time will tell whether an apology is necessary.

[@KCM25NJL]

Having AI stacks that operate things like Keyboards/Mice and browse the web by pointing and clicking..... feels about as wrong to me as two different AI's communicating by speaking to each other with mic's and speakers. That is the true AI friction that everyone is rubbing up against. To solve this, I think the world needs another layer of abstraction on top of the internet, an AI fabric if you will. I think until we have something like that, there will always be application level fragmentation that results in less "visual" progress than we expect.

[@NullHand]

But there is already one at the low level that AI can, and probably already is using. Tcp-ip packet capturing. This is done for network security, and is basically a giant real time pattern matching Red Queen's race between malware creators and Internet security systems. It just so happens to also capture EVERYTHING that happens on the Internet.

[@mrleenudler]

The big reason(s) hydrogen didn't take off, is that hydrogen is really inefficient as an energy carrier ( 1/5 of batteries) and that fuel cell vehicles are much more complicated than BEVs, and prone to break down. In addition to electricity already having a network, of course.

[@totoroben]

I agree with your assessment of having a baseline competence for what AI can do right now. I have used it a lot with my workplace over the past few years and watch KZbin videos. Those who treat AI as a fad will get left behind. Yes AI gets things wrong, but if you are ignoring what it gets right you will be left behind. Focusing on what it's currently good at, and methods to validate workflows developed with AI, you will be very important for business implementation. Knowing a thing or two about project management, aligning stakeholders, and involving end user subject matter experts will be instrumental in using AI as it exists today. Building AI that is fully autonomous is not gonna be as feasible right now. Building AI to make workers more effective is.

[@casualuser5527]

👏👏 Excellent video! Thank you for addressing the tempering of expectations in the short term. It’s a balance to strike that I have seen some overcompensate from “AGI tomorrow” to “AGI nowhere soon.” I feel this video is a nice check with the reality of AI.

[@mrd6869]

One thing as well. The adoption rates are low because the process of automation isn't fully mature yet. When workflow automations are researched & set up by agents, you'll begin to see more practicality. CEO's care about their bottom line. How can this save or make me money. That's it. Like if Zapier with its integrations could automate that process,it would be big

[@joaodecarvalho7012]

Sora was like 4 months ago. Do they want new advancements every month?

[@jimbojimbo6873]

And we are still nowhere near agi ffs

[@MrTScolaro]

So the question isn't will AI experience a trough of disillusionment, the questions are how long will it last, how deep will it be and what will come out the other side that we cannot yet forsee. I frankly think short (18 months), shallow (people will still play with AI) and who knows.

[@michaelnurse9089]

Until Nvidia crashes to 10x earnings we are so not in any trough. We are just between major versions.

[@jtoodlet5896]

Gartner lifecycle doesn't apply to AI imo. It's only going to get better and more disruptive year on year

[@jamessullenriot]

The only thing I disagree with is the VC funding, I think that will pick up for "AI" and other startups as interest rates settle back down. I think over the past 1 - 2 years there has been a crazy misdirection in the job market blaming AI instead of interest rates and overall financial resetting post covid bailout money. My company is huge and worth billions. Most of their cuts have not been AI related at all, its that the money they borrowed/owed (yes, companies that measure revenue in the billions still borrow money), started to become more expensive. My thinking is within 1 - 2 yers things settle down, hiring is back to what it was, and AI will find its place within society as it moves from hype to actual useful products. Not what it is now where its chatbots and the same 5 - 10 business ideas that you ask an LLM for ideas to start a business.

[@darkprototype5353]

Dude how inconsistent can you be? Is it fast or slow?

[@coolbanana165]

I don't know how different this is from what I've expected. - Ai capable of maths, self-reflective reasoning, and taking actions (API) in 2025 - Ai companies integrate LLM's with business software, and offer this to large business, in 2026 - Large companies to expand Ai usage to the extent it makes a difference in 2027 - A few more years for small and medium business to use Ai. - So still significant change by 2030, with limited but growing uses before then.

[@PersianMate]

Hi David, what is the best way to contact you to consult with you and potentially do business?

[@karthage3637]

It felt just fine, you can build services over a provider that cut down massively it's price every now and then while increasing the capacity of it's product Easiest business model ever Currently working for a company that provides RAGs but I will then move to build my agentic services

[@a7xfanben]

Thanks for your perspective on this!

[@josephcler3299]

What about the issue of the multimagewatts needed for an AGI. This is not very efficient compared to a human brain which only uses approximately 30 watts of power.

[@roro-mm7cc]

not sure how its slowing down, every year we seem to be getting better models...

[@johnphamlore8073]

I'm now thinking that unspeakable incompetence in management in the US might drive AI adoption forward. Just today Friday July 19, 2024, there might be weeks of disruption due to a poorly managed update from Crowdstrike -- they didn't bother to roll the update out in stages, not all at once. Across the board in the US there is a disturbing trend of sheer incompetence, with astronauts still stuck on the ISS due to unreliability of their return vehicle, what happened last Saturday, and now the Crowdstrike update.

[@greatcondor8678]

I would love to adopt AI but it is horrible to work with right now.

[@arinco3817]

I think the possibility of rapid adoption hinges on whether a good desktop agent can be created. Otherwise like you say it will be pipelines with apis which can be an arse to set up. But at least that will create jobs in companies to automate stuff?

[@brianWreaves]

I fully expect most of our current browser based needs will no longer be accomplished using a UI as we'll all have our own personal AI assistant to fulfil those needs. The exceptions will be entertainment, communication, social, gaming, enterprise platforms, perhaps education, and a few other areas.

[@jackkendall6420]

even if AI never improves even a tiny bit beyond what we have today, it will change the world and radically alter the job market. the question is whether it does this in a matter of decades or months

[@EnigmaticEncounters420]

I think we should stop trying to make a prediction on ai because at any point a breakthrough in ai can end the downturn in a hype cycle at any point. Though the gartner hype cycle seems to be an accurate portrayal of almost any tech. Please explain to your viewers every step of a gartner hype cycle so they can get an understanding of why you're thinking we are going through it right now. (It does make sense.)

[@carlosegonzalez678]

Well I see it as in capitalism you can’t take it slow or somebody else will left you behind. Just because you are quiet do not mean you are not playing to win.

[@TheRedLineExpress]

Hey man Companies dont care about slight issues They will happily take the training curve of improving ai if it means saving money in the long run These companies already hire humans to put out subpar products

[@MrChristiangraham]

I suspects there's two hype cycles with AI. There are plenty (and increasing numbers) who are stealthily deploying AI in their work to write better emails, get rid of drudge work like risk assessments etc. They are in the slope of productivity phase. The bigger transformation from AI, we are definitely around the peak of inflated expectations.

[@franciscomagalhaes7457]

I don't get it. Is it getting cheaper or more expensive training these models anyway?

[@Arthur-jg4ji]

@@franciscomagalhaes7457 faster

[@DaveShap]

The models are cheaper to run, new models are more expensive to train

[@franciscomagalhaes7457]

@@DaveShap ah there you go, my mistake, thank you.

[@brianhershey563]

I like to visualize AI advancement like a brain, each layer similar to early human cognitive layers. We're currently building the evaluate, trust, and verify frameworks, plus memory coming online now, what a time to be alive! 🙏

[@mrd6869]

Also Dave your prediction for AGI may not be off. You don't know what Labs like OpenAI or others have behind the scene.They may very well have something near AGI.Just hasn't been released yet

[@brianhershey563]

25 years since the first Bluetooth standard... WOW!

[@ScarlettM]

AI is such a volatile and enormous technology, that instead of sigmoid curves people expect, we might get a sharp V.

[@dynamynx]

In the creative space I would actually be highly wary of anyone incorporating generative AI into their workflows. Sure it’s more ‘efficient’ but who knows how legally this whole thing is going to shake out and then what, I have an under skilled worker who got the job using now illegal technology?

[@supremereader7614]

What if they're saying 'AI isn't accelerating' as a trick? If some company were getting better and better with AI, would they brag about it or tell the world 'progress is stalling?'

[@MrVohveli]

The AI hype cycle hasn't even started. The second OpenAI shows agents the C-suite will instantly understand this can be used to replace work, only then will the hype cycle begin.

[@danmapes7697]

Come on David - you’re acting like Gen AI is the only AI - Embodied AI is natively multi-Agentic - and the Spatial Web protocols link them all together. Do your homework - it’s all there in the Gartner report and you don’t even mention it.

[@thatwittyname2578]

Embodied AI will be the true AI explosion.

[@jeffsteyn7174]

He probably knows that.. But when is that happening? He is talking about what we have right now and whats missing is ways to easily integrate applications with genai. Why would I pay for a robot that costs how much? When I can get the same performance and have access to it right now by integrating with an api or RPA.

[@I_am_a_human_not_a_commodity]

This is all assuming you are privileged enough to be able to get a job in the first place...

[@saschas2531]

Where is AGI? :D

[@rey82rey82]

Representing Star Trek in dress isn’t slowing down

[@-RizonGaming-]

Thank you always appreciate uplifting news.

[@mater5930]

AI as it is right now has better use then what is hyped about. It's good uses are 'boring' and not worthy of hype.

[@xxsb101xx]

I don't really buy the idea of "they are all hiring them back". The issue with offshoring is the language barriers and overall lack of quality control. I could see them hiring a few people back since AI still gets stuff wrong, and it simply still needing a real overseer at the moment, but I doubt that issues are such that they are bringing everyone back, at best most back. The main issue is still the same, that companies will definitely be slowing down hiring over the next few years as AI is able to do the productivity of new employees they would have brought on without it. This also doesn't address AI getting better, which will eventually hit the turning point of making such few mistakes that they can fire the overseers, and just keep a token staff on hand for bug reports for the few times they get stuff wrong, and the inevitable expanding in office jobs AI is able to complete more or less on its own. I agree with the structure issues, but on the other hand, how long are these realistically going to last? Is this simply another half a year to a year before we start seeing products come out that address these?

[@Kilakilic]

We could have much productivity gain from just making regular software better, but I guess thats boring. And then you could even integrate ai to it where you need extra help

[@The-Singularity-M87]

It's a possibility that AI companies don't want to get caught up in any issues of artificial intelligence influencing the United States 2024, November, presidential elections. Which makes perfect business and social sense.

[@HectorDiabolucus]

AI is somewhere between the lightbulb and the discovery of fire, so I don't think that the hype cycle will apply, except in people's perceptions.

[@benpielstick]

I feel like the main thing holding back AI adoption is hallucination. It can do a lot, but it can't do a lot reliably.

[@chrisschoenfeld4414]

I missed the onshoring trend I feel like offshoring maybe shifted a bit to nearshoring but that’s it. 90% of my devs offshore at last job.

[@toniarbona]

The summary is Keep Calm and Trust Kurzweil. 2024 was too soon.

[@nickhowarthify]

People are worried about their data, and data sharing.

[@asdasdsfasfasfasfasfasf]

I don't agree with this. If GPT-5 just started training, we have to wait to see how good it is before drawing any conclusions. As Microsoft's CTO says, "we only get to sample the exponential once every 2 years". Anything that happens in between those two years is .. noise.

[@pietervoogt]

This trough of disillusionment is very much colored by your personal feelings. Most experts had somewhat longer timelines than you for AGI, so for them it feels more like a calm period than a trough. Also Metaculus always had dates for AGI like 2027-2033.

[@ustedo]

When AI will be used on RAAS App ( result as a service ) only and specializaed work fonctions ( accounting, law, helathcare ect...). Now here is the deal experts of theses fields will need to know how to code or sit down with dev. All these tools , no-low code are very limited in term real world work case

[@kilianlindberg]

Interesting hmm.. I wonder how big a dehype can become in this realm 🤔 speculative market can make stuff happen; yet for example OpenAI is not public .. so the actual value of getting a Knowledge Worker for a fraction of the cost of before (unless it’s regulated in a way or OpenAI as a dominant party somehow intentionally limits its full potential; but we do have anthropic or actually more players that has funding for a few years that’s actually already profitable )

[@MA_SK_ID]

Why am I still optimistic? I mean yes the shine is dulling but I do think Ai will change society significantly just not at the pace I had imagined.

[@dudleefing]

Good video. Thank you.

[@dirkbruere]

The election has to put a major damper on AI announcements. I cannot imagine any AI company wanting this to be a political football

[@nathansmith8187]

When David Shapiro has reached the AI trough of disillusionment you know it's unironically unequivocally over.

[@nathannebraska2408]

I imagine the world will look very different in 20 years from now, but not next year

[@michaelnurse9089]

Tesla will roll out Robotaxi towards the end of next year. That will be real AI doing real work.

[@theaidividend]

Pal - never seen you miss the mark so broadly. AI dominance will be lead by disruption, not adoption. All kinds of firms can hold their breath, it only takes 1 adopter in that space to wipe out the rest. We need a convo. Come speak to our 50k followers sometime. Reuven Cohen’s LI community.

[@DoopieDog]

hmm, I thought "trough of disillusionment" is where one goes after eating taco bell

[@greglhoticom]

See my profile picture? Grrrrr..... I hate it when I fit into your dialogue. Doesn't stop me from WANTING AI to draw a bloody log home floorplan tho' 🙂

[@ivanivanich761]

Another common David Shapiro W

[@Towner101]

“I know this isn’t gonna work, and if you want to try we’re gonna fire you” dude, not cool. Very narcissism much sad.

[@context_eidolon_music]

Okay David.

[@robotheism]

robotheism is rising quickly

[@michaelnurse9089]

That's what she said.

[@BrianBetron]

Nice job

[@epicarbon]

I like you when you make stable videos like this than on the jungle walking

[@thomasschon]

But what about Trump's plans for a Manhattan Project on artificial intelligence? Won't that risk starting an arms race, accelerating development at the cost of cutting corners of all possible corners to reach AGI and beyond before the competitors?

[@infinateU]

FIRST consider mining concepts to secure Network(s). - education, business & entertainment will all be "GameAfied" to onboard many & enable Every/any person to participate. I think any old president that endorses BitCoin is just not enough. Citizen must realize the value of "Mining" for Network(s). Demand Decentralized Networks & UBI for "Mining" for Network(s).

[@destinypuzzanghera3087]

Thanks

[@Tom_Neverwinter]

so who are these people and why does it seem plagiarized off others...

[@ModernTruthRevelation]

Foreseeable future: 1-2 years is crazyyy🤣

[@wedding_photography]

This narrative doesn't make much sense. You admit we're likely to have AGI in 5 years. But it's not like it will just randomly drop to our planet from outer space. Neural nets will keep getting better and better. Tooling will get better and better. How in this scenario will things get worse?