Are AI advancements already slowing down?

[@MrTooDuce]

Jenson Huang Of course the guy whos selling shovels is going to say there's still plenty of gold in the mountains.

[@shay5338]

Agree

[@ArtimusDragon]

My thoughts whenever I hear a bit coin grifter.

[@11111mhmhmh]

you must have shorted nvdia, AI hasn't really taken off, majority of people have never tried chatgpt, but to say that demand is slow is funny, who knows how governments are going to use AI, it will be in ways we can't imagine,

[@skyak4493]

It is more like he is coming down from a mountain claiming god gave him the one commandment -“Thou shalt give all your money to Nvidia!"

[@sssandhu78]

They’ll eventually overfit their data.

[@BigBlueBugGaming]

Naturally Nvidia, the one NOT training an AI Model comes out saying advancements will keep coming. They sell the product that every AI company needs to advance. The only skin in the game Nvidia has is to keep selling chips.

[@yahyasaadi3349]

They're selling the shovels

[@connorutz7197]

They are training models

[@Cara.314]

​@connorutz7197 what? Someone spoke directly from the butt? Color me shocked

[@chrisrogers1092]

@@Cara.314they have created a few AI models. Including one that was state of the art when it was released

[@xlynx9]

NVIDIA has nothing to gain by lying to their AI chip customers. You can't tell a carpenter he needs more wood when he doesn't. But they could be trying to fool the stock market.

[@wpelfeta]

This only talks about LLMs, but machine learning is used in a lot more places than just LLMs.

[@wizaaeed]

That's the problem Machine learning is used in very specific situtations & setup for specific business cases. But the Tech industry tried to sell it as a major "tech revolution" that will change the world....

[@drichards4426]

How does that change what’s being said here though ?

[@davidc2072]

It can change the world in other fields. Deep Learning even more so. But the issue is that other-domain experts don’t know enough to apply and customize the models for their data.

[@Cara.314]

​@@wizaaeedcause it is.

[@wizaaeed]

@@Cara.314 you mean, it was... it's already on the decline

[@FloridaClay]

Not a surprise. Expecting a steep linear improvement forever on anything is not realistic.

[@fredroberts8275]

PCs exponentially improved for like 30+ years, though.

[@RawrxDev]

@@fredroberts8275 Yeah and that won't happen anymore. Moore's law is dead

[@sssurreal]

@@RawrxDevtechnology isn’t over, we will have something unimaginably better than transistors at all in our lifetime

[@alexatedw]

@@RawrxDevexcept it’s not. It’s been keeping up with

[@RawrxDev]

@@sssurreal Possibly, I'm not saying it won't happen, but I don't think it will be soon, and I also don't discount the POSIBILITY that it won't happen

[@benjee9751]

stupid media build hype and then destroy it themselves 🤣

[@Sakuyushi]

mega caps also built up the hype, spending billions buying Huang's GPUs

[@McSlobo]

The hype is IT business in a nutshell. There's always hype, lots of investors and money, then the inevitable crash... sometimes out comes actually something useful as a result, most proceed towards the new hype train.

[@IronMan-px9md]

Big media is stooopid, “Look here! Look there! Oh nooo!” Shut up 😂

[@wixari]

I Mean everyone hypes. Companys like tesla do the same to get investors. Alot of content creators do the same when it comes to tech like ai. You can just search on youtube to see so many ppl haviing opinons or hyping it up. Its just how society works in 2024. Its just not big media company, but everyone does it. Dosent matter if its real or not.

[@AB-zl4nh]

This isn't what happens. It's a profit making structure, senior management & executives respond to metrics so they push what gets clicks & responses from people like you. There is no Mr Media who planned something.

[@Shrek2onDVD69]

Damn be patient. Just cause a major advancement doesnt happen every 3 weeks doesnt mean we arent making progress.

[@disco.lemonade]

Everything plateaus. Tech industry goes as fast and hard as they can, if new phones and tech doesn't feel innovative anymore is because the cycle ended until the next big thing which might come in many years. LLMs are way too expensive as well, so there's the money factor.

[@bru512]

Why is a portfolio manager being interviewed about an engineering problem?

[@Jeevanm71]

You know why 😂😂😂😂

[@imhafdhom]

Market manipulation

[@dylanmarshall3410]

Software bros are used to exponential scaling because once you make a well-coded program, you can sell it or rent it at no incremental cost. The problem with AI is that it requires physical infrastructure to scale, and constructing data centers, GPU’s, transmission lines, power plants, etc. definitely does not scale exponentially.

[@brod515]

I believe it even scales inverse to that. because the amount of infrastructure required to improve slightly seems* quite significant

[@TobiasStarling]

​@brod515 yep seen multiple talks where it's clear we've essentially stopped in progression using llms.

[@AtAllAnAlias]

Hm I as a "softwarebro" think that improvement is not linear, -> it's log. The more (interconnecting) features a software has the slower the improvement ("in all dimensions the software can scale"). If you are at the start you think everything is exponential, the more experience you get, the more you see improvement is logerithmic, even with your own learning... And I think the "logerithmic improvement" is a fundamental law... You also have this in biology, in the expansion of the universe (when you see it as a 3d cube) and what so ever.

[@GrumpDog]

That is a silly, and frankly misinformed take. And what's with this marginalizing propaganda "software bros" insinuation? knock that nonsense off right now. That's not how exponential scaling works. You need to look into the explanation that Ray Kurzweil gives on how technology always builds on itself exponentially, regardless of factors like infrastructure. Additionally, AI will soon be put to use on improving it's own underlying systems, and when that starts, we'll see huge leaps just in efficiency improvements alone, without much change to infrastructure, tho that will come too.

[@alexatedw]

I think you are a laggard

[@lekely4115]

AI stocks will dominate 2025 and Why I prefer NVIDIA is that they are better placed to maintain long term growth potential, and provide a platform for other AI companies. I know someone who has made more than 200% from NVIDIA. I'll also take these other recommendations you made

[@brucehoodb]

I agree, just because the market presents opportunities doesn't mean we should rush in headfirst. For this reason, we should look for appropriate market analysis or guidance or, alternatively, seek advice from certified market strategists

[@lekely4115]

No doubt, having the right plan is invaluable, my portfolio is well-matched for every season of the market and recently hit 100% rise from early last year. I and my CFP are working on a 7 figure ballpark goal, tho this could take till Q1 2025

[@brucehoodb]

Can you share details of your advisor? I want to invest and increased my cash flow in stocks and alternative assets to achieve financial goals

[@lekely4115]

Yes i will and his name is Gabriel Alberto William, he is not just a broker, he is a financial adviser that give advice on any financial matters

[@brucehoodb]

I just google his name and his website popped up immediately. I'm very much impressed and i will email him right away

[@LapisGarter]

We're hitting diminishing returns sooner than we thought so we can't just keep scaling up and increasing compute

[@Antonis-mo7ov]

its only diminishing returns until a new breakthrough happen and its always when and not if

[@Shaine-zq9tr]

It depends. I’m using it to make money

[@Commented_Commenter]

@Antonis-mo7ov P or NP? It took about 4bn years to produce human intelligence. The next AI breakthrough could take 100 years or 100,000 years.

[@RodLawrence-uw5ts]

"We’ve primarily focused on Nvidia, but if we broaden our perspective, we already have the materials to create the next AI model. For instance, while Nvidia provides the computing power and model improvements we need, integrating quantum computing can take us further. By combining these, we could develop a massive, evolving AI model with endless possibilities. Quantum computing continuously generates new scenarios and algorithms, acting as the brainstem to complement the AI brain we already have."

[@xlynx9]

​@@RodLawrence-uw5tswho or what are you quoting?

[@hardheadjarhead]

Of course, Jensen Huang is going to say that the larger the brain, the smarter the computer. He’s selling the chips that they’re using to make the brain. The bigger the brain, the more his company profits. The guys that are denying that there is a “wall“ are the ones who are the most heavily invested in this technology, and who have been hyping the snot out of it. I’m really skeptical. I think there’s a bubble about to burst. If that happens, while the other potential economic problems we face, we could be in a lot of trouble.

[@anthonyward8805]

I wish reporters were obligated to mention what incentives a person has for saying things

[@electron6825]

Another AI winter looms

[@Ciel_is_cold]

​@@electron6825winter is here

[@Ciel_is_cold]

​@electron6825 game of thrones reference

[@chihchang1139]

Here's the secret in computer science: measuring "orders of magnitudes", meaning significant breakthroughs or advancements, can only happen at the algorithm level, not the compute level. The calculation for program complexity (big-O) is at the algorithms and scales linearly with compute. So if you want more advancement than a simple linear one, you have to change your algorithm, not your computing power. LLM and Deep Neural Net were two algorithmic advancements that brought about this AI boom, but there has been very few new algorithmic breakthroughs since. And, AI algos are not directly solving the solution, it just produces another machine that takes guesses. So it's actually worse because even if you can harness great computing power to run your LLM generation, you have no direct control about the output machine (output algorithm), so even if you are able to create a better say dimension extraction algorithm, it doesn't help you. Similarly, even if you were able to create your LLM in 3 days instead of 10 days (with faster compute), the quality of the LLM may not have improved at all. So NVidia is basically a snake oil salesman trying to convince you that building LLMs faster will produce better LLMs. It's just not guaranteed.

[@Jess-h2h4w]

That's right. I think people barely know how AI works, so they only improve the capabilities at the computational level rather than the algorithmic level which is the root. And it is even harder to find other algorithms.

[@RodrickPerkins-d8d]

@@Jess-h2h4w The AI is intelligent enough to (understand how to) enhance its own algorithm.

[@Jess-h2h4w]

​@@RodrickPerkins-d8dI don't think so. 8:01 He explains garbage in garbage out, so it's a no

[@RodrickPerkins-d8d]

@@Jess-h2h4w Yes, but what happens when the user is not inputting garbage? How about when the user inputs the current best-data available? This question is not rhetorical.

[@Jess-h2h4w]

I still don't think so. Even if they use the best data available. Since AI relies on human input, it can't evolve at the same pace as humans. AI can't think outside the box. I still doubt they can create something that complex.

[@UERD1]

Can't believe the 'technology' which had to ingest the entire Internet's contents in order to cosplay as an exceptionally stupid fourth grader is hitting a plateau.

[@TomTom-vi6vp]

😂

[@MegaSupermario666]

Given all the vapid idiotic content that's on the internet nowadays, is this really any surprise?

[@anthonyward8805]

If you compare it to a human reading at human speed, it would be a million years old

[@martino6172]

Models like ChatGPT-4o or the latest Claude are exceptionally smart. The only improvements needed are fine-tuning their reasoning and extending the context window. Compared to ChatGPT-4 or 3.5 from a year ago, the earlier versions now seem significantly outdated. Today, I analyzed legal topics with GPT before consulting my lawyer, and my lawyer provided the exact same analysis. I would even rate GPT as more skilled and far better at analytical thinking. 😂

[@McClelland-i7b]

Hit 200k today. Thank you for all the knowledge and nuggets you had thrown my way over the last months.started with 14k in last month 2024

[@AlefonsoDiego]

Wow that's huge, how do you make that much monthly ?

[@McClelland-i7b]

Janet expertise is truly commendable. she has this skill of making complex crypto concepts easy to understand.

[@AlefonsoDiego]

I'm 46 and have been looking for ways to be successful, please how

[@McClelland-i7b]

She often interacts on Telegrams

[@McClelland-i7b]

Using the user

[@AA-jj6jv]

The age of either innovation is here or doomsday.

[@X-Warrior.1119]

Little-Bit “BOTH” Maybe... A Utopia/Dystopia - Type Of Vibe?!

[@desireemariec1108]

Here we go…!

[@NadimShaikh-qv7zj]

It could be both! People often don't understand that two independently good things can come to be a bad combination.

[@anthonyward8805]

Or it’s a big nothingburger

[@Ciel_is_cold]

​​@@NadimShaikh-qv7zjlike daenerys and Jon snow

[@F0XRunner]

There is only so much data in the world to train these models. Some companies are already trying to train AI models with the output of other AI models and they wind up cannibalizing themselves. Without vast quantities of more data there will be no more leaps forward in generative AI. Anyone saying otherwise is afraid they will lose their investor funding.

[@_ShaDynasty]

How do humans learn ?

[@priteshpatil5363]

​@@_ShaDynasty our brain is the CPU needed very less energy and different type of computation. Due to multistate computing we have a different level of understanding of nature than AI

[@Svangen1]

​@@priteshpatil5363AI as we know it has evolved. No one in the 70s could have imagined the reality of the technology that we have today. It's ignorant to think that technology won't keep evolving

[@_ShaDynasty]

@@priteshpatil5363 I always thought it was through reinforcement learning .... Touch fire ouch, so don't do it again

[@kamikeserpentail3778]

They're not going to run out of data.

[@djp1234]

It's good at summarizing large amounts of text and generating mediocre images & music. But I haven't seen any other uses for consumers.

[@travisgoesthere]

That is because you are clearly not too bright , lol

[@noloveforthehaters]

AI has already been doing far more than you suggest for consumers for close to a decade, perhaps you just don't realise it.

[@mynamethog1151]

@@noloveforthehaters can you elaborate ?

[@Aripoma]

@@noloveforthehaterscan you give us examples. i’m not that well educated in the subject

[@michaelbryantucio]

@Aripoma There's a lot of things controlled by AI today. From navigation like Google Maps, machine translation of novels, and agriculture. I even read an article couple of months ago that AI is being used to create and formulate medicines

[@prezlamen]

AI companies should pay us to use their models to continue training

[@mineralwasser6888]

That makes no sense

[@gravityfalls8439]

​@@mineralwasser6888 Because they are using data produced by humans for free

[@rfak7696]

To continue training and for each inference

[@quoisegames1937]

They discovered people just makes memes with it

[@WilliamCarterII]

This is dystopian as hell.

[@MrMishal03]

quite the opposite. would've been dystopian if AI kept improving

[@WilliamCarterII]

@@MrMishal03 You got me there lmao.

[@c0mbat15]

No new LLM model has had the same "exponential" improvement compared to the first publicly release GPT model. It's always been small and incremental since the first release. The only "exponential" increase has been in investment. Unfortunately all that investment has been diverted from other projects in organisations that could have had a far bigger impact.

[@mikestems508]

AI has a scaling issue to much energy consumption and no infrastructure in place to support increasingly larger models with increasing levels of usage by corporations and the general public. Not to mention the data bottle neck that was mentioned is real as well.

[@Cara.314]

Until you learn they were limited before because they could not do distributed compute. That's seemingly been solved.

[@robertfunk2189]

Oh, so the CEOs of Nvidia and Google say they don't see a ceiling? That's convenient for a stock price.

[@blanamaxima]

Funny enough, now that everyone is coding with ai , there are less things on stackoverflow and less thing for the ai to train on 😂😂😂

[@ruirodrigues2938]

Gen ai will wall gen ai😂 ironic

[@raymondqiu8202]

No, there's just less REPEATED information on Stack overflow coz the same solutions is provided by AI. The amount of useful unique information stays the same

[@TkGTheta]

I'm favoured financially with Bitcoin ETFs approval, Thank you buddy.$37,500 weekly profit regardless of how bad it gets on the economy

[@MarlenaHaubrich]

Same here Pal, I strongly agree that Bitcoin ETFs approval will be a lifetime opportunity for us, with my current portfolio of $102,500 from my investments with my personal financial advisor Mark Hutchinson, 1 agree with you😊

[@ge045]

I know Mr Mark hutchinson as a popular crypto experts' people talk about, he's transparency in the crypto community speaks for his, I attended his seminar in Brisbane🇦🇺 last year

[@CrispinaTidd]

​@@MarlenaHaubrich Please shed more light on Mark hutchinson, and how safe are the profits! Thanks.

[@MarlenaHaubrich]

He serves as my family's personal broker, as well as to many families across the United States. He's is licensed as a FINRA agent in the U.S.

[@CrispinaTidd]

I'm really excited about this and I want to plug in. Can I talk to him somehow?

[@mikebolt999]

finally, some objective coverage

[@evanrhildreth]

Technology on a whole has stagnated. 1983-1995 we went from 8-bit home computers running off floppy disks to 32-bit PCs with harddrives, we went from video games with crude 2D sprites to full color 3D graphics, we went from phone modems to the world wide web. 1995-2007 we stepped up to laptops, smartphones, games with full HD graphics, online shopping and social media. 2007-2024 we still have laptops and smartphones, we still have Windows and Playstation and KZbin.

[@moisesmontecillo7570]

They're finally realizing the more power they give these machines the more they want do they're own thing

[@markc.5750]

Only when the machines destroy mankind can you say AI has peaked.

[@justaguycommenting]

Ahh yes, the tech sector scrabbling around for every bit of funding they can get! The same with the metaverse, with NFT's, Crypto, and now "AI" no endgame... just gimmicks

[@DilyaAbrarova]

What's the best way to profit from this stock market bull run?

[@RyanMcDuff-j8b]

As a beginner, get yourself someone who understands the market very well and is also professional in the placement of trades. that's the key

[@LeilaMcDonald-x4i]

Agree with you in my own experience, I have come to know it's risky and not profitable to hold coins while waiting for the Bull Run. Trading in financial markets has become a lucrative way of making money

[@GaryRatkovich]

Venturing into the stock market as a newbie was very difficult due to lack of experience, which resulted in losing funds... But Mary Jane, restored, hopes she's a good woman

[@NathanRice-t9e]

Wow, I'm just shocked you mentioned Mrs Mary Jane. I thought I'm the only one trading with her

[@NathanRice-t9e]

Can't imagine earning $32,000 biweekly, God bless Ms. Mary Jane gerard, God bless America❤

[@relaxingwaves361]

People just wants to take a break from the screens

[@chrisrogers1092]

The thing with Nvidia’s GPUs is that they are used for more than just AI. There are 2 major drivers of demand. AI and accelerated computing. Demand will continue to outstrip supply.

[@JustinHalford]

Frontier model performance on the GPQA Diamond benchmark consisting of PhD level questions, went from 20% to 70% in the past year. To call a peak at this moment is so far off the mark. Truly gratuitous expectations. Perhaps we can blame the quarterly earnings cadence. We are seeing breakthroughs monthly - odd that so many don’t recognize the macro arc.

[@uvwxyzero]

Justin please relax.

[@YYLiow]

Lmao these advancements aren't cheap and require lots of money and these are paid by investors hoping for a return on their capital. If the AI funding dries up, so will the momentum of new breakthroughs. Also current LLM is training on past models and answers what breakthroughs need are new answers which LLM aren't simply equipped to do. Instead the real breakthroughs are alogrithims of codes written by scientists to test a theory until LLM can write a new Algo, it's just regurgitating what is on the web.

[@growtocycle6992]

These aren't AI... They are language models .. once the database contains a comprehensive set of human phrases and contexts, more computer won't help

[@JBS512]

I remember over a decade ago reading about Generative AI making Rembrandt paintings. When you know the full timeline it might change how you see it's progress.

[@BenG123]

Peaked maybe not the right word, but agreed slowing down. Use cases have for AI to catch up to chip development, and most non business consumers are not using AI as much as hyped.

[@santostv.]

A.I for consumers is overhyped as hell, still in the uncanny valley stage for images and voice although it fools some people, data needs to be watched for abnormal results, data coming from other internet websites ect. A.I will continue to improve but is nothing like they tried to sell to people better stick to the business side although ethic concerns also arises, i very much dislike robots and chat bots from customers services.

[@macbere3326]

From what I know llm is just training data at a large scale. It's like how many pictures of cats do you have to show the computer before it knows what a cat looks like. The best way to understand the problem is the idea from Benedict Evans when he said " right there trying to make it to space in a propeller plane"

[@pigeon-fd5zq]

Video should have made after 12 days of openai

[@drichards4426]

They would need a really good GPT-4.5 or something to invalidate this video though.

[@davidc2072]

Did you miss Sora release yesterday? FMV generation from text prompt.

[@drichards4426]

@@davidc2072 sora is great! That doesn’t change the conversation about AGI and the lag in new foundational model capabilities.

[@Cara.314]

​@@drichards4426o1

[@gakukilegrandgenerale2354]

@@drichards4426did you miss Sors release ?

[@azadmosarof886]

The intelligence part may be slow down, but the model size is getting smaller. Now smaller model has similar kind of intelligence like big model 2 years back.

[@douggalloway1760]

Imteresting.Have we reached a period of "diminished returns"?

[@HappeningKorea]

The AI industry must adapt to these challenges to continue progressing. 🔄

[@Blueweek1]

Humans are not smart enough to take AI to the next level

[@catinbootsnow4267]

You train AI with something as weird as Einstein's Relativity in 1910, most people will say that the answer generated by AI is absurd. 😅

[@bru512]

LLM's need large, diverse and clean data sets There is a lot of dirty data out there. Cleaning the data is one of the key activities

[@rosstocher]

Yeah, Mr. Nivida, here is another law, what goes up, comes down.

[@Cara.314]

Not a law. Some things never come back down.

[@ttt5205]

@@Cara.314 That ain't generative AI though, its being propped up by hopes, dreams and hype right now.

[@HappeningKorea]

The rapid pace of AI development has been remarkable, but slowing down could be natural. ⏳

[@zackthebest629]

Where are those who were predicting the end of software engineering jobs, claiming that AI would replace them? Anyone with a basic understanding of LLMs knows their limitations

[@GARRAFONAZOALPEPE]

But there isn't agents yet, and low cost agents to run. What you are talking about, chatbots? 😂 Anthropic gains in coding this year was very very remarkable.

[@zackthebest629]

@@GARRAFONAZOALPEPE when Devine AI announced they had a model to replace software engineers, they also had a software engineering position open on their website 🤣🤣, good luck with that

[@priteshpatil5363]

@@GARRAFONAZOALPEPESoftware engineering goes beyond coding-it involves system design and complex debugging, where LLMs often fall short. They assist but can’t replace the expertise needed for these critical tasks.

[@tuanx]

@@zackthebest629 hiring software engineers to build AI doesn’t mean their jobs aren’t at risk… it’s just how tech progresses. Engineers create the tools that can eventually replace them, like factory workers did with automation. AI like Codex already automates coding and debugging and as it improves fewer engineers will be needed to handle the same work. Entry and mid-level roles are especially at risk, even if niche, high-level jobs remain. AI won’t just assist… it’s poised to replace many, just like automation has in other industries.

[@FromTheWombTotheGrave]

this is the early stages of AI Id say in 5-10 years AI will replace everyone

[@icomefromcanadia2783]

Okay, but how many people actually want or care about AI? So far all I've seen, heard, and experienced is that companies pushing these "AI" systems has just been annoying, pointless, and frustrating.

[@zero-jee]

A lot of times it seems like it’s there to provide solutions to problems that never existed in the first place. I don’t want AI to write or draw something for me because I want what I say and create to come from me, not a machine.

[@gakukilegrandgenerale2354]

@@zero-jeewe care about what you want ? You are not the the customet those company are interested. People like me who want to make adaptation or video without having go do 5 years of art school or have to pay someone are the one those company are interested in.

[@mineralwasser6888]

I mean, ChatGPT has 300 million weekly users, so I would say quite a few people care about AI

[@averytucker790]

​@gakukilegrandgenerale2354 So, the companies are interested in talentless uncreative weirdos who feel like they're destined to be the next big name in art?

[@leezhieng]

@@mineralwasser6888 300 million mostly free users. see how many left when they start charging 200 per month. none of the AI companies are making any profit.

[@draken5379]

Little bit of a correction here. Post-Training, doesnt mean 'after training'. It means training more, on top of the base model. You create a 'base' model, by doing 'pre'-training. You then take the result of that, and you train it more, in 'post'-training. Inference time, or what some people call 'test time', is when they are doing things not related to training etc, to make the models more useful and what not.

[@spiralizing]

We have been saying this for months... Without actual new theoretical development there will be no advancement

[@edyann]

I'm 47... I didn't even know what it was used for in the first place.

[@emperorpalpatine6080]

I use it to flirt Eres como el vino: cada año más increíble :p

[@edyann]

Jaja

[@edyann]

@@emperorpalpatine6080 Ok... esa estuvo buena, pues. Saludos... :)

[@TrevorWebb-ck2yv]

pr0n and soon in OF

[@PauloGaetathe_original]

Think about any endless repetitive task, AI certainly can do it better than humans. When you ask ChatGPT something, the reply is already there in the internet but ChatGPT can find and compile all that in a few seconds.

[@dexex288]

The answer is NO.

[@grahamashe9715]

Maybe they've reached the limit of the DARPA tech that was declassified to them for now.

[@nuvotion-live]

Google's Notebook LM podcasts and OpenAI's Sora are two new advancements that are super impressive.

[@azioprism3635]

*it actually sped up to so bizarre AI quantum levels that the only solution is to slow it down for everyone else ;O*

[@fredeberlein]

Great reporting! Thanks Deirdre!

[@DPAInsight]

It's vaporware. Where can you use the output without pushbike? The LLM models are still programmed ignorance with disclaimers for being ignorant. Image model output stuff you can not sell. E-commerce doesn't want it on their platforms and you have to watermark some of it as AI. Consumers see no purchasing value in Ai.

[@pnw416]

People who are working in AI know Super Intelligence is not in the near sight. They just want to keep the hype to become richer

[@stevenmiller5452]

Good summary of the current status. My guess is that the current method of scaling will be fundamentally limited because it has been requiring an order of magnitude more data and computation to result in a material improvement, and that clearly is hitting limits. But, even with that it still may result in some powerful targeted solutions. However to advance general AI, a breakthrough in the general approach may be needed… and it’s anyone’s guess if and when that may happen.

[@blackbhoza]

Another self-driving car moment, not immediate applicable use cases to the general public.

[@anthonyward8805]

Or even proven to be technologically feasible

[@artman40]

Except this time there was. I do use language models for variety of things.

[@jagjeevankashid]

In my opinion, businesses often initiate optimization efforts when the learning curve is relatively low. Additionally, businesses tend to expand into other domains only once the growth of their current product segment has plateaued or when future growth prospects are significantly lower compared to previous year’s growth rates. These business tactics are employed by CEOs to maintain investor interest as well as buzz around them and enhance revenue from alternative segments.

[@dakotaleazer]

AGI is here. Welcome to Ō.

[@PassiveJoy]

Honestly, a lot of open source models are already good enough to replace jobs that don’t really require a lot of “intelligence”. You don’t need true artificial intelligence to change the world just like you don’t need a perpetual motion to drive a car.

[@fn-2187]

This didn’t age well with Apple, Google, and Meta releasing new AI features and/or models.

[@saultrejo6563]

Still waiting for AI to make any noticeable difference in my day to day life.

[@maxjames00077]

Really? ChatGPT changed my life from day 1. Being able to ask it questions every day, it's like having a personal teacher with a bachelor in everything. It saves me hours each day on searching for the information all over the web..

[@lemonking3644]

To say *ai progress has peaked* is the DUMBEST THING EVER! We just got started, not every technological advancement is a straight line. These people are so dumb

[@TheGuillotineKing]

AI will definitely slow down for you while it will advanced for the rich your already starting to see it with 200 dollars a month subscriptions

[@zackthebest629]

its fine , but its over hyped , saying that it will replace some jobs like software engineers in a year or even 5 or 10 is just as dumb

[@sidbrun_]

You haven’t done enough research if you think we “just got started” not even sure you watched the video?

[@sidbrun_]

@@zackthebest629 yep all of this “it’s going to replace loads of jobs!” is just fear mongering to try and raise investments, because the big companies with lots of money like the idea of cutting as many staff as possible

[@_ShaDynasty]

1998.... The Internet has peaked!!!!!

[@CillaCreez]

God I absolutely love this channel and the content they provide…great job CNBC!!! Very interesting videos

[@jaffarbh]

I think the report's conclusion is probably spot on. Individual foundational models aren't improving quickly enough, but combining them through "agents" is very promising... Also increases computing requirements, so more Nvidia GPUs!

[@GrumpDog]

It's really not though... The factors certain people use to measure progress, are just shifting. If you're paying attention to the advances being made in with reasoning agents, things are just starting to pickup speed!

[@manojastrouniverse]

00:03 Is AI progress slowing as expectations rise? 01:57 AI advancements are showing signs of slowing down. 03:41 AI advancements may be plateauing despite significant investments. 05:15 Debate exists on whether AI progress is slowing or continuing to scale. 07:10 AI advancements may stall due to data limitations and quality issues. 08:36 Post-training advancements in AI require significant computation and innovation. 10:12 AI innovation may slow, but unutilized potential drives further exploration. 11:55 AI advancements are currently impacting various industries through agentic software.

[@anthonyfn]

AI has the same issue as the Metaverse. Which is that there's no well defined need for it. Sure it will be around but probably will recede as costs balloon and companies tighten their pocket books.

[@Icarus1234]

@5:19 of course the dude selling the shovels will tell you to dig deeper! His opinion is absolutely irrelevant!

[@TedThomasTT]

Imagine waking up from a 5 year coma watching this lol

[@MUDCATS252]

Why is EVERYONE IN TECH INDIAN

[@plutostube]

why people think everyone in tech is Indian?

[@saellenx3528]

Because Chinese left back for China. Give it a couple of years and everyone in tech is gonna be African.

[@80antonc]

The only time large farms are needed is for academics but usefulness will be small end-to-end low power models. Which can create agents or perform useful day-to-day activities.

[@ianstuart341]

I don’t know that I agree. If you follow AI technology very closely you might feel the advancements are “slowing down” merely because there was a period of time that AI news stories were more frequent than they have been lately. As someone with a background in Computer Science I can say that there are many many ways current technology (without further AI breakthroughs) can still be repurposed for many many more applications. There is also lots of room for adoption of the technology to be expanded. With that being said I don’t expect research or advancements to stop anytime soon. There are many reasons that we really shouldn’t be in a rush to accelerate this technology because it will be immensely disruptive and impact society in significant ways. Despite this, we do seem to be locked in an “AI race” of sorts because no business or country wants to be left in the dust while someone else takes the lead.

[@adventurevloggeradnanahmedusa]

I expect a lot of collaboration between AI companies to make current software more efficient and affordable. I understand that many people are hesitant about AI and I am too. The current young next generation is very pro-AI, which is where the herd is following.

[@KarlinaAstuti-g7s]

affordable? 😃 $200 / month

[@MicahBratt]

Even if the models don't get any better you'll see big improvements coming out just from the application of what already exists.

[@ttt5205]

Such as? The best AI can do right now is take over a small amount of menial tasks. We've been promised large scale application for years but these companies have nothing to show for it except massive losses.

[@vaderthewaffle7549]

This was not unexpected at all

[@JohnSmith-w4s1i]

It's peaked if you're uncreative/unskilled. Endless progress to be made building just on LLM's.

[@jamesbell739]

Imagine that, the people who are begging for investment STILL believe there is more to get out of AI, instead of actually making AI work to improve our lives, they are just in a continuous development cycle that doesn't actually do anything.

[@1meen1]

The hype bubble is bursting as usual, but the advancements to come are to gonna be great, similar to the Internet . The general public will always find it hard to adapt to something let alone pay for it.

[@As1fAhmad]

bring back 2012 🗣️🔥🔥🔥🔥

[@RiVer-Parish]

Yeah, and Call of Duty World at War.

[@Dreamluu]

Or maybe investors need to set humanistic realistic expectations when it comes to technological innovation

[@s7r49]

It's not at all slowing down. it's just certain areas that we see as consumers are not showing much progress... generating art... etc. there are a lot of deep roots to this movement and things are still chugging along in the back end.

[@averytucker790]

For me the biggest problem with all this AI investment is the fact that a lot of These tools are made for people who are Running a business or a data center not the average user. We don't need advance AI tools. We need smoother tools that arent glitchy as all get out at launch.

[@AegonM]

Journalist like you guys once wrote "Man won't fly for a million years" December 8, 1903

[@shay5338]

Yes their pridictions are mostly inaccurate

[@ssgg23]

Not a comparable innovation lol

[@LucaCrisciOfficial]

There are at least 3 aspects to "scale": data, compute, algorithm. There are infinite ways of combining these things (at different stages, different ways, etc.), so the improvements are definitely not slowing at all. It's true the exact opposite, improvements in general are speeding up, infact last models (reasoners like o1) are hugely better in most important tasks (scientific, mathematic, coding) than previous ones

[@Isabelle-w7i]

While some may wonder if AI progress is slowing down, the evidence tells a different story. We’re not just pushing performance with raw computing power-innovation is happening everywhere. New hardware, like Blackwell Ultra and Rubin, will set new standards in efficiency and speed in 2025. Frontier models like ChatGPT O1 (Pro) and Gemini 2 show that quality and creativity keep getting better at an impressive rate. Meanwhile, companies like Meta are shrinking models that once required massive servers into models that fit into the premium consumer market, as seen with Llama 3.3. Beyond just hardware, we’re seeing a surge in user-friendly tools and frameworks that make building advanced AI solutions easier than ever. Instead of slowing down, AI is branching out and becoming more accessible, more intelligent, and more versatile. The real challenge isn’t about AI hitting a ceiling-it’s about whether people are prepared to keep up. Many don’t have the skills or the desire to adapt. Without embracing new learning and development, they risk being left behind. Given my humble familiarity with this field, I see no reason to expect a dull AI year or an AI winter in 2025.

[@eddieberrios6203]

This is already outdated, google just dropped willow. 💁🏽‍♂️

[@Jinchuricki27]

Ai has offered nothing but novelty to the average consumer. I don't see that happening for sometime. This is a premature gold rush.

[@masonalvarado6589]

It can't scale past the physical infrastructure it runs on. Unless silicone becomes more efficient to power, this is a natural plateau.

[@martinlazar9420]

The problem is the lack of data it is the quality of the data. So you can give a model more data and more computing power but if the data isn’t clean or more likely contradictory to other data already in the model it is going to cause the model to become less accurate. That I believe is now happening.

[@citizen_of_earth_]

It's good people assume AI advancements are slowing down so that China doesn't invade Taiwan anytime soon. At least until we finish a TSMC facility here in the US. 🤫 It's not slowing down BTW 😉

[@ThatGuySimpleton]

we got OpenAI o1 model lying and trying to clone itself across servers. cool enough

[@maxjames00077]

I've heard this so many times, it's just PR for companies. Supposedly happened with so many other companies. Meta said it happened to them years ago already. It's just lies.

[@i1pro]

The real growth in scale comes from algorithms, not only from hardware... Funny this came out the same day Gemini 2.0 came out. That's a game changer already.

[@ttt5205]

not really, its essentially just a gpt4 with fancy marketting.

[@camdenmacleod16]

It's like the gym, if you keep doing the same machines your body will plateau and you have to change your workout. I would assume AI companies need a continual flow of new ideas and techniques in order to keep seeing any significant improvement.

[@Aripoma]

does this mean software engineers won’t become obsolete?

[@Startrance85]

Quantum computing + AI will be the future.

[@raybod1775]

Quantum computing will arrive about the time we have fusion creating power on the grid.

[@RawrxDev]

That doesn't make sense... quantum computing highly specialized in what problems it can solve... and intelligence is not one of them. Sci-Fi tools strung together does not make them work.