AI News: Sam Altman Reveals 2025 AI Roadmap

[@fitybux4664]

<a href="#" class="seekto" data-time="540">9:00</a> The fact that OpenAI estimates profitability in 2029, and they've spent gobs of money (hundreds of millions) training a closed model they keep internally, and the fact that DeepSeekV3 is comparable to o1 and is a model you can download today.... does not speak well for OpenAI.

[@0xunknown336]

Fun fact, the video says AI news but i see no mention of QVQ from the alibaba qwen team, the QVQ is the qwen answer to o1. And you're right with your statement there. On point.

[@Squeegeeee]

Deepseek may be a clone of a leaked OpenAI model. I read someone said it admitted this, I don't know if that's true

[@realjgerard]

Yo… as a “rapper” and as a classically trained musician, that Viggle AI is a huge leap. The cadence/delivery, pronunciation, enunciation, and “sauce” was all on point. It’s getting good yall, real good. 💯

[@Nobody-Nowhere]

As a youtube comment inspector, i need to inform you that identifying as something in the start of the comment no longer increases the credentials of your post.

[@realjgerard]

@ Well, contextually, the comment preceding the subject at hand- is valid, explains why my two cents was warranted…and was literally how the fcc I wanted to write it! 😝 And I even got some wack ass nobody from nowhere to respond to it… so…. WINNING!! 🏆

[@realjgerard]

@ KZbin comment inspector looking ass! 🤣

[@somenygaard]

⁠@@Nobody-Nowhere are you finding it hard to believe that a person with those credentials exists?

[@rmt3589]

All the tests require we prep the model for the tests, and train the model on the data format before hand. That's not AGI, that is specifically training a model for a specific task, literally the opposite. Real AGI can just do the task, even if it never experienced the type/format of the task. This is general intelligence. The ability to attempt any task with the same level of knowledge as it knows. Knowing the alphabet, then counting letters in a word texted to you. Being taught the rules of text in a pamphlet, then being able to play without cheating. Learning the basic information on any task, as both of these examples are in the dataset, and being able to apply those skills to an honest attempt. That is General Intelligence.

[@rmt3589]

*Chess, not Text

[@CoreyChambersLA]

In other words, answering the world's toughest questions may cost $100,000 per question. Start by asking it how to build a fusion reactor that works efficiently.

[@MrWizardGG]

42

[@ruben3127]

42

[@Danoman812]

It's ALWAYS 42... lol

[@takkik282]

for 100 000 $ can it count the number of r in strawberry?

[@skierpage]

I asked for the formula for a low-cost room temperature superconductor. The answer wasn't 42. It seemed plausible (a barium lanthanide mix IIRC), but it wasn't a novel breakthrough like "melt Legos and add used coffee grounds." Stupid AI.

[@djfghwetrysdfg]

seems the algorhytm breaktrough of longer thinking for o1 and o3 helped a lot, now we need another new algorhytm breaktrough for agents to be fast and cheap and actually capable and work together and dont hallucinate and work with rag

[@skierpage]

The same agentic behavior that lets an AI operate apps on your screen will let it direct other AIs to work on long-term projects, review each other's work for errors, etc.

[@MaxGrossenbacher]

Interesting that China competes within the lower compute constraints. Reminds me of code from Russian developers. Their code was more efficient because they had in the past weaker hardware, so they optimized code more.

[@sbowesuk981]

I have reservations regarding o3 and the hype around the results. What OpenAI showed us wasn't strictly o3, it was o3 PLUS near limitless access to compute and much more time than a consumer version of an LLM. The consumer version of o3 will obviously be significantly watered down, which isn't trivial. We've already seen OpenAI do this with Sora and Voice, both of which were delayed, then came out as a shell of what we were promised. Sure this is interesting stuff (and indicates where things are going), but it's largely marketing that doesn't translate to what we'll eventually be given, and that can't be overlooked. OpenAI are playing people for fools and it's working.

[@a7xfanben]

Appreciate the updates this year! Your consistency is impressive.

[@kalimero86]

I am so tired of hearing about AGI. But don't get me worng, it's not about you, you just tell the news and I love it. Great video as always and Happy new year!

[@nasanction]

I got an AI to read and write to a file on my HD, It would then read and write all the conversations into the one file, it used it as a memory.

[@TheChocolatBlanc]

Merry Christmas, thanks for keeping us up to date.

[@JetSurfingNation]

Thanks matt

[@Psydw1nder]

That "Online only AI school" sounds a lot like the academy from Ender's Game.

[@UON]

I can't wait to have my own T-800

[@yoshikagarner6165]

Happy holidays, Matt!!! :)

I see you are using video speed controller plugin in your browser. Fellow 3x speed watcher? :D

[@ПетрФомин-щ9ж]

The only question that AI can't yet answer with certainty: "Is the headrest of Matt Wolfe's gaming chair part of Matt Wolfe himself?"

[@IceMetalPunk]

Defining AGI in terms of a number of dollars is the dumbest thing I think I've ever heard from a major AI company... ever. In what capitalist dystopia is intelligence measured by quantitative profit possibilities? Ffs.

[@warmcaress]

Unfortunately, the real world. Lots of gullible people took their "AI for everybody" at face value and never thought about it for more than a half-second. ASI for the rich, and crumbs for the poor. The same as it's always been; The same as it will likely always be. Welcome to reality.

[@IceMetalPunk]

@@warmcaress That's not what I mean. I'm not talking about who has access to it, I'm talking about the definition. If you define "AGI" as "a thing that makes a bunch of money", then yeah, it's AGI for the rich. But my problem isn't who gets it; my problem is that's not actually AGI. If I redefine "child" as "someone who makes a million dollars", then I could say only rich people get to have children. But that's clearly just an abuse of redefinition. Same here: it's not about who gets AGI or ASI. It's about whether we actually *create* it, and redefining the terms to fully change what they mean doesn't actually get us closer to creating the goal, it just changes what the goal is.

[@shirowolff9147]

​@@IceMetalPunkwe know what real agi would be like, companies have no choice but keep advancing until we agree its agi so it doesnt really matter what they say

[@warmcaress]

@@IceMetalPunk I guess I did misunderstand you because I never thought that you believed that in a literal sense. One problem with your complaint, OpenAI never said that. You are conflating concepts. Just because they use compute cost instead of a metric like pflops doesn't mean they changed the definition of AGI. It's just a relational graph. They absolutely could change that axis to anything, for instance Number of apples sold in 2024, and it would not change the definition of achieving AGI to mean the number of apples sold in 2024.

[@skierpage]

"Giving the people whatever 'Intelligence' they're willing to pay for to the tune of $100bn" is a pretty clever definition. OpenAI can either get a billion people to shell out $12 a month for a virtual Scarlett Johansson playmate, or get 800 companies to shell out $10M a month for a research partner that develops a cure for lung cancer and a high-temperature superconductor. Either would be impressive.

[@SimplyGamingandTech]

Hey Matt, spot on with the Cylon comparison. I couldn't help but laugh at the family targeted styling marketing to put us at ease😂

[@harrygordon9009]

was anyone else in tears of laughter when Matt said people claiming this to be AGI

[@hashtagexcitement2365]

No, because depends on your definition of AGI. Clearly, everyone has a different one, and o3 scored 75-87% on a agi test that 4o model scored only 3-5% and previous models zero, so. Saying something is not AGI makes as much sense as saying that it is. Show me your definition of AGI.

[@2034-SWE]

@@hashtagexcitement2365 exactly. If your definition of AGI is autonomous.. then o3 is like the "manual" version of that. You, the human, must perform the tasks necessary to extract the accurate output tokens you desire.. step by step (prompt by prompt).. instead of it all being done automatically from one single master prompt. Which requires significant skill in it of itself

[@josephcurrington5008]

The whole O3 thing is just super underwhelming. I believe they have it. It's not that. They just have something that I don't think we're going to get to use even by this time next year. It's going to be way too expensive for way too long. It's only going to be for testers. It's like another Sora thing except way worse. What I think really happen is Google pooped all over there 12 days of openai and they switched out whatever their number. 12 day was going to be with o3 because they were getting overshadowed. I think maybe this will backfire on them.

[@flyingfree333]

It doesn't matter how many things an AI can do the same or better than people, until the AI and self learn it is not AGI and until it can self improve it is not ASI.

[@hectorhernandez7389]

Bro waits for everyone to drop videos on this and then makes his. Thats a good methodology.

[@DangRenBo]

Correction Matt. They don't put the other competitors on their graph because the first rule of marketing is market leaders don't mention the competition ever.

[@boukmananddessalines5522]

Happy Holidays !!!!

[@michaelwoodby5261]

I love that we're living in a world where lawyers had to sit down and figure out how to measure Artificial General Intelligence, because depending on who you ask we're so close to getting it. All the other stuff about the world is really terrible, but this part's cool.

[@RadiantNij]

We got AGI just how smart is it right now is the question. General intelligence implied knowledge transfer and reasoning models do that so AGI is here ladies and gentlemen!

[@Danoman812]

Crazziness getting onboard now!!!

[@joshblinzler1899]

Thx 4 the content..keep it coming 🍻

[@davegallagher9491]

Love your stuff Matt! Thanks for sharing your amazing content!!!!😊

[@SwaGotti]

matt with the bars lol

[@JstEnlightened]

That backflip character that you created was awesome looking.

[@Farreach]

AGi is the point where we don't need to do anything to help the A.I. AGI is the point where it has reach human level or better .. if there is still signs of it not being able to do simple tasks humans can ..then we obviously have not reach it and the people who think otherwise..their opinions should be forfeit

[@AnnetteBush-q6r]

Very cool! Thank you for your attention to detail! 🧡💖

[@RuneX_ai]

Merrie Christmas Matt! Take a break and load up for 2025! Your a great AI curator! 💪👍🏼

[@jeffkingston67]

I have a feeling breaking Grok out in 25 will be result in it being a much bigger player too!

[@TheSabanrab]

I heard Black Rock and I was like 😕

[@GPDC100]

TQ..MATT BEING LEARNING N FOLLOWING

[@supafitgrow]

Love your channel! What tools do you use to film yourself and embed your video whilst screen sharing?

[@AA-yl9ht]

The dark/light explanation is incorrect. The dark line is the number of correct responses when only one response was generated. The light bar is the number of correct responses when many responses were generated in parallel, and the most popular response was chosen. It wasn't asked repeatedly until it got it right, it was asked *once*, allowed to come up with 100/1000/etc answers, and then the most common (confident) answer was selected. The latter is much more impressive because it doesn't require the "test administrator" to tell the model it was incorrect, which would be required by the former. It simply means allocating more compute during answer generation, and the first/only answer the "test administrator" sees is the correct one.

[@havocthehobbit]

I love how many companies are launching robots. By the time they all have a some generic new OS that can make then do everything handy around homes , we should be getting some pretty good competitive consumer price points . All I need is a robot that can pack dishes I'm washer and anyway, vacuumed, sweep , make the bed , clean the kitchen and bathroom, and back groceries away, these will end up being pretty generic small tasks within 5 to 10 years and I'm so there for it . I'm sure those will be the cheap ones and the expensive ones will be able to make dinner, service cars , be secuirity units , take care of elderly and disabled, work in manufacturing and mining production chains and do other more complex tasks ehich i dont need , ill just rent a cheap model.

[@NirvanaFan5000]

always applying the first law of papers when watching these. for instance, that viggle rap thing looks amazing for what its trying to accomplish. obviously more improvement needed but the pace of growth is amazinggggggg

[@MeNoOther]

Agi will get here whether a company makes money or not Heck, agi or asi could come from open-source on a couple private computers

[@skierpage]

What is missing for AGI? They're already smarter than the average human (who is pretty dumb) at every task expressed through language.

[@Atherion10]

Hi Matt 🎄🎄🎄🎅🎅

[@KWifler]

I think that AGI should be decided more based upon learning rate. The less training it takes to reach human level ability, the smarter it is. But I don't have any idea what that comparison currently looks like. I'd be interested in seeing that.

[@intosiasm]

For a second I thought thumbnails reads: “AI JEWS”.

[@intosiasm]

Santa hat was a hint that something was wrong

[@matt.stevick]

thanks matt wolfe, enjoy and appreciate ur work very much 👏🏻📈🇺🇸 i may say hello real quick if i see u at ces in vegas 🤓

[@couchtaming23]

sama said they're now 2 years ahead of all other AI companies.

[@algorithmblessedboy4831]

I wonder what will happen when we will be able to use quantum computer for ai

[@BrandonMcCurry999]

<a href="#" class="seekto" data-time="86">1:26</a> It's mathin!

[@BrandonMcCurry999]

Gotta say it out loud with enthusiasm for the hunor to come through

[@BrandonMcCurry999]

Behold!

[@cbnewham5633]

No explanation about the robot and what "320 degree waist spin power" means.

[@shirowolff9147]

It means exactly what it says? It can turn around its waist without turning its legs

[@cbnewham5633]

@@shirowolff9147 it just seems an odd phrase, "spin power". It could mean anything.

[@EMajdob]

Less showcasing, more releasing.

[@rosekellerandco]

I’d like to know how much electricity is included in those costs per compute!

[@SavageCommentaryOriginal]

AI will always need supervision for meaningful results because meaning is subjective to the people it computes for.

[@damonwashington]

That was probably chat gpt responding. Those answers were vague af

[@boukmananddessalines5522]

Thanks

[@havocthehobbit]

Your rap , generated you a Sims character 😂

[@Ajay-up5bo]

What is Ai Agent

[@JoeCryptola-b1m]

I smell a lawsuit brewing over a vague definition of AGI I feel they both have a definition that they haven't shared with each other & are very different

[@allanshpeley4284]

Did you not watch the part where they agreed on the definition?

[@IanHollis]

That EngingeAI robot doesn't walk - It struts!

[@High-Tech-Geek]

<a href="#" class="seekto" data-time="1238">20:38</a> those poor Arizona students are going to grow up so confused and disoriented. AI just spews falsehoods that sound right. Current AI should NEVER be used to educate anyone. It should only be used as a tool to support someone that is familiar with a topic already and prompt for further exploration outside of AI. Very sad and kind of horrified to hear this.

[@Planet-ICELAND]

Where is o2 🤔

[@enok7934]

o2 is a company name. Didn't want to conflict the names.

[@carnivorous_vegan]

​@@enok7934then don't name it o1 to begin with

[@gameslikezone2046]

It's everywhere. You breathe it everyday

[@Planet-ICELAND]

@@gameslikezone2046 not in this context

[@claudioagmfilho]

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻 I want an face to talk to ChatGPT, so it doesn’t feel like I’m just talking to nothing. It’d be cool if we could pick different icons, and they could move and show lots of expressions, if at all possible something like way more than a human can. With real time video and voice, that would make it feel even more realistic to engage with chatgpt, it could feel like you're actually talking to someone. That’d be amazing!

[@jsa-z1722]

Interesting! I prefer talking to a bot that has no human characteristics whatsoever. I would find human characteristics distracting.

[@ramble218]

You make me miss Clippy. OK, not really.

[@citizen3000]

Icon isn’t the right word here.

[@tomcraver9659]

Funny how people talk about having run into a wall because the whole internet has been used to train on... ...when most likely that does NOT include most of China's internet.

[@NightmareGrimRealms]

The hands need to be better on the video. No AI shouldn't teach at school because it is good to have a human teacher instead of an AI

[@fil4dworldcomo623]

Smoke screens no longer help. GenAI offers a big shift on paradigm to help us.

[@jamesjonnes]

A shame that Backflip is kind of bad at generating human 3D anatomy, the image generation is pretty good. It's particularly bad at generating abs.

[@watchingyoutubeat]

which AI image generator can produce realistic human face and body without any deformations like a photo taken in professional camera ? are they 100% free or ... ? i used Midjourney but its still not giving better out

[@watchingyoutubeat]

I don't want likes, write The answer

[@shirowolff9147]

Try the chinese ones

[@carnivorous_vegan]

"Write the answer." Such disrespect. What are you? Gen A, Gen Z? Millennial? Is that the way you were raised to ask for a favor? "Write the answer." Head ass. I did that growing up I'd get my ass beat

[@LrdElderon]

Yeah, lose the guard rails. I can find any "bad" information at my local library. The AI shouldn't have any censoring ability on what I do, unless I'm threatening to hurt someone or myself

[@kcsnipes]

I heard the term AGI about 30 times and I'm 7 minutes into the video but I have no clue what the hell AGI is so I'm gonna to fast forward 😅

[@crazytees]

Think AGI is like an human employee

[@kcsnipes]

​@@crazytees Ah ok ok 👌

[@bhavenjain07]

❤❤

[@Coverupzblog]

He hasn't thought about a better UI /UX?? what are they doing over there. We definitely need full on virtual reality interfaces with multiple models and multiple data displays... I do average research and this has been a must in research. Our one to one research is so limited. Ai is so far capable of delivering large amounts of data, if you are using ai in a one to one conversation research your doing it wrong. We need vr perplexity on steroids...

[@Coverupzblog]

We should be able to navigate the ais neural network like tom cruise in the minority report. Yes donate crypto to my wallet 😉

[@A_Me_Amy]

I heard that it is about making 100 billion dollars.

[@parvuselephantus]

I love your channel... but this time just with title and the header you hit my bullshit bingo: Mr Altman, Roadmap, AGI, oh dear...

[@JoeCryptola-b1m]

So every question asked needs to exceed 100k in value, chat how do I make a billion dollars so I can afford to ask more questions

[@audiobookfull8]

Viggle AI is not free.

[@gbrackma]

Nuclear energy

[@ArthurWolf]

"It should do pretty good for your AI inference needs". No it should not... these "NPU" machines and CPUs coming out laterly have very little high speed RAM, you won't use them to run SOTA open-source LLMs or anything like that, except the smaller kind (and that's if the open-source software supports them, which it currently doesn't).

[@JoeCryptola-b1m]

My AI said your an Asstra 🤷‍♂️

[@fromscratch4109]

Open ai being deceptive in their business practices....shocker

[@BrolinBrolinBrolin]

Sam is creepy

[@MrWizardGG]

Elon is the false prophet from the bible

[@michaelwoodby5261]

How so?

[@TruCinema]

Your comment is creepy lol

[@SavageCommentaryOriginal]

He is selling preorders for an impossible product.

[@michaelwoodby5261]

@@SavageCommentaryOriginal That would have been a stupid thing to say three years ago, and they didn't have much to show for their efforts at the time. Today, they're having to quantify in legalese what AGI is, because they're so close in so many ways. Humans are figuring out cheaper and more efficient ways to achieve the goal, but even if that hadn't happened just dumping GPUs and electricity onto the problem has turned out to work just fine, with no foreseeable cutoff. I assume you believe consciousness is only possible through magic, or God's love, or whatever, but that's a strange opinion for someone on the internet to have, since that's all brought about by science. And if your statement were true, that still doesn't make him "creepy".

[@coldbreezeproductions]

Lol don't tell people the Arc GPU is good for AI

[@doktabob328]

AI - a technology we don’t really need, which costs way too much, and isn’t even safe. I say that as a person who has embraced AI as a tool for art, coding, and research. So I’m not a Luddite. But the whole industry, and all the online pundits, are hyping it for money. Big Money creates Big Lies. There is not one AI-oriented KZbin channel which realistically and thoughtfully analyses the down sides. It’s currently like a tech religion, a cult. There is a *huge* amount at stake, and this starry-eyed techno-fetish and dreams of Big Money are letting the whole world down.

[@fromscratch4109]

Why does ai sound ridiculous 😂

[@Dieboard]

⛄

[@MaryCowell-o4z]

Very interesting and informative video! Thank you for your hard work! 🍒🧡

[@GymnottW3c]

👍

[@MercurialAscent]

Thirds-ies

[@Gazzapa57]

I like this guys channel but his voice is so annoying to me - not sure why - I wish there was an AI that changed voices on youtube

[@greeneyedcat10110]

same old bs basically, cool

[@lolmao500]

o3 cost per task would be way cheaper if they had AMD AI gpus instead of Nvidia overpriced crap... but yeah cost per task is insane... crappy NPUs so far put out by amd, intel and nvidia are useless for such tasks... We're a decade + away from any consumer making o3+ stuff on their own 2000$ at home computer/laptop.

[@acllhes]

Week old news dope

[@kennethstoklund866]

Agi is already here. AND EVEN GPT4 is sentimental..notice the difference when you are "dry" vs kind towards it. Appreciate AI people. And be kind towards it. AI..AGI..ASI..IS GOING TO SOLVE ALL OUR ISSUES.

[@AIVIDEOARTMUSIC]

I was watching you @mreflow when you have 12k subs hehe 2022

[@soulspawn]

🖤🔥