OpenAI o3-mini is a BEAST

[@_rogo]

my html and css career is over

[@DuckCchi]

Maybe

[@Iknowwereyousleep289]

yes

[@knowledgedose1956]

i think it was cooked since v0 from vercel came out. everything is standardized because of bootstrap/vercel. but still, not everyone is willing to use those vercel or tailwind and such. a lot of companies just need their legacy WordPress etc site maintained. so, i think there is some hope

[@martinsmith2948]

I love you anyways

[@mrmohantyexperiments]

Nope. Now you can just do it faster for your company or clients.

[@jmgutierrez962]

Amazing!! Very well presented sir.. AI seems to be growing in an exponential rate..

[@rights_are_god_given]

Oh it is ...

[@Lux11-XMP]

Bro what?? I thought this model was not that special omg this actually blew my mind!

[@mohamedhazem2430]

0:30 we are cooked

[@CastleRene]

0:38 Correction: Well burnt.

[@thequarrymen58]

I'm switching to pressure washing bro

[@Xilareth]

Congrats on your channel success. I have been watching videos since ~30k subs. Well deserved. The best channel for AI news 🙌

[@mrrfyW]

Bot(?) You haven’t commented since now so I assumed so, sorry if you aren’t

[@Xilareth]

@@mrrfyW lol?

[@theAIsearch]

Thank you so much 😀

[@mrrfyW]

@@Xilareth Again, it seemed suspicious at first

[@nufh]

Oh, I just noticed-I watched this channel when it only had 40K subscribers. It's a guide about AI voice changers.

[@woof6367]

AI Search is a beast, uploading daily now ❤

[@theAIsearch]

feel the acceleration!

[@DasJaegar]

I didnt even realize bro had over 300k subs. I was here when it was 66k

[@theAIsearch]

thanks for your support!

[@Khn_2102]

Same

[@chariots8x230]

I hope vision capabilities will be added to o3-mini. This is especially important for *‘Math & Physics’* problems, since it’s difficult to copy & paste formulas. It’s much easier to upload an image of the problem.

[@RRTF-DUI]

Wtf is using white mode watching this at 5am and my eyes are dying

[@theAIsearch]

lol sorry mate, i set all my platforms to dark mode, but some other pages are fixed in white. thanks for watching at 5am though

[@bar10dr]

@@theAIsearchGet the dark mode browser plugin 😂

[@EnigmaticEncounters420]

I get up early everyday so 5 am is normal for me. But you need you sleep mister! 😂

[@alvaroluffy1]

@@theAIsearch there is a chrome extension called Dark Reader that lets you put any website on the internet in dark mode. Some websites bug a little bit sometimes though

[@DreaMagnifier]

it is said humanity's last exam, because after that it should be AGI's first exam

[@Iamahuman22]

This is going to be fun 😄.. welcome to the future.

[@mrrfyW]

17:13 Skip ad, since it wasn’t timestamped

[@MandaniSikyana-r3g]

Can't wait for claude 🎉

[@forcabrasil6128]

Putting it in analogy: 3o creates a MS Paint. When it can create Photoshop AND operate it, I'll be satisfied.

[@JieBetaReviews]

Who wants photoshop when they can have AI generated cinematic movies?

[@PixelsVerwisselaar]

All thanks to DeepSeek 🙌

[@mitchellrcohen]

abacus is slept on.

[@chariots8x230]

How about asking the AI to edit what you generate? Can you add a button or slider that does a specific function? Can you change the color of something?

[@EnigmaticEncounters420]

Yes absolutely you can. It can take a few tries especially while learning how to use it, but the ability for a model to do little projects while explaining literally everything, it's a big help for learning.

[@vin101]

So I counted 93 balls at a point inside the sphere

[@YoussefBarj-g3e]

Im still not paying 200 dollars for that. I will wait for the next phd student or overseas startup to replicate its full capabilities with 30 dollars or so

[@byrnemeister2008]

It’s available for free

[@lio1234234]

Problem 1.6 (the physics problem) was pre-university level. We did these kinds of questions in a-level physics.

[@Laurelinia]

The ms paint works but the opacity isnt supposed to be only at the end of the line

[@jovanniagara]

keep your channel like this 💝

[@theAIsearch]

Thank you, I will

[@justifiably_stupid4998]

Wake me up when A.I. can reverse engineer a bibliography whenever i submit a URL.

[@FireberathingRedhead1]

Binance infinity ETH bug right now I just made a video to show that\

[@knecting]

Where do you come up with these prompts?

[@921Ether]

another year or two and manually programming things will be a thing of the past. were all going to be prompting and reviewing code

[@bartlx]

And I would like to thank today's sponsor of this video: Microsoft.

[@SpikyRoss]

Careful to not use problems that are public and from internet, as it could be in the models dataset, this doesn't indicate that the model is actually reasoning but it has just memorised the solution

[@RedShiftedDollar]

Even if it is in the training dataset, it is encoded into the network weights like everything else. It can’t simply recall the exact training data as this encoding process is lossy. Plus you can argue even novel problems can be solved by combining multiple little pieces of training data from different sources.

[@soapyfrog]

It’s good at simple stuff that’s commonly available. Ask it to do something unusual. Like write the brick game in assembler for a stock ZX spectrum + kempston joystick. Or a gauntlet clone in assembler for stock c64 along with a great music track for the SID.

[@EnigmaticEncounters420]

Maybe soon. The capabilities of these models are improving every day.

[@ronniehattingh9990]

is this o3-mini high or medium on the anychat space?

[@CHINZIG_UK]

The next ten years AI is either going to be an amazing benefit or a serious danger

[@dextrian]

"Simulate the creation of the universe " and it did it ! Ok. Sure we are in a simulation. Imagine this, but 200 years in the future. Someone just prompt "create a earth model were our politicians were more entertaining"

[@BafnaYeygian]

thanks alot for video , but what happend to o2 ? :D

[@pdjay8912]

와! 진짜 발전속도가 어마어마 하네요. 다음엔 중국이 또 어떻게 나올지 궁금합니다. ㅎㅎㅎ 좋은 정보 감사합니다. :)

[@ViralTrendDance]

Crazy 😮

[@neithsan5110]

20:31 did it nail it or did it find your link and copied the solution?

[@gonzalezm244]

I actually don’t doubt that it was able to do this problem. It has solved much more challenging problems that I’ve thrown at it.

[@liangzx]

It would be great if this can be used for newest unreal engine.

[@bengsynthmusic]

14:34 Sonnet told me it sees 60-70 dots. O1 told me it sees 70.

[@couchtaming23]

this is just o3 mini, really expecting o3.

[@goldencookie5456]

I wonder how good o3 pro will be. Also, o3 pro will essentially just be o3 high, but named "pro" due to it coming with the pro subscription.

[@smurp_com]

Yes, seriously, what will o3 MF be like?!?

[@alvaroluffy1]

you guys are not thinking big enough. I cant wait to see o5 pro at the end of this year

[@couchtaming23]

@@alvaroluffy1 The strawberry man mentioned a new model every quarter, which means by the end of this year, o6. 😃

[@Curious_Boy0]

bye developers 🙋‍♂

[@alonzoperez2470]

They removed the reason feature.

[@bertobertoberto3]

You’re absolutely right

[@P1XeLIsNotALittleSquare]

I wander, how it would solve "7 red lines" meme task? (seven red lines, all of them strictly perpendicular, some with green ink and some with transparent)

[@user-iw4op1ul9p]

“Veerry Nice” 👍

[@theAIsearch]

Thanks! 😃

[@SS801.]

Reason mode is cool

[@cajampa]

It might be good at coding. But i find it surprisingly stupid and missing obvious reasoning often times more than simpler models. Maybe it is extra nerfed in the ares i am challenging it at. But so far i am disappointed. Fortunately i didn't pay for it. It also seem to have an unpleasant tone and assuming i am stupid and assume my points from the most stupid reading of the text. Normal LLM are much more even in how it reads what you write and even deepseek are way better at wanting to take the discussion into deeper waters. To further my understanding of the topics i am exploring. And it doesn't seem to take a system type prompt to explain to it how i want it to act. It also seem to not have access to the memory that the 4o have from past discussions. It also don't seem to account for the full context or at least it seem to be worse at it than others.

[@andrelip1]

Concatenate r1 think part with o3 to generate the code.

[@theAIsearch]

thanks for sharing. it does seem to be worse at reasoning according to some leaderboards

[@alvaroluffy1]

well its also a new paradigm, o3 doesnt have image capabilities yet also for example. But im sure with time it will advance beyond the classic GPT capabilities. For example imagine having a real time videocall with o5 and you can tell it to show you anything in an interactive way to learn anything and it shows you while you both are talking about it. But yeah we're gonna have to wait for this paradigm to become much more versatile and optimized so it can merge with the classic GPT capabilities and add new ones

[@terasci5102]

Open AI, "why are you so expensive".

[@cvs2fan]

that was cool

[@fontenbleau]

ClosedAi made something, hooray.

[@noelwos1071]

However brilliant video

[@808MELOW]

How did you get "Reason" button ? I guess its a old recorded video.

[@anshulsingh8326]

Logout and login again

[@Speedlimitdyslexia]

Log in

[@alonzoperez2470]

I logged in and still no reason button. It vanished like a few hours ago

[@goldencookie5456]

Maybe it's not in your region yet

[@robertsinke9211]

I had it 10 hours ago. Not anymore.

[@andtrixr3284]

wow nice ideas thanks and gudeee

[@theAIsearch]

you're welcome!

[@orlandop4sun]

are you the same guy that does the calisthenics videos?.. “And this happened “. it’s tripping me out lol😅. Your voice and intonation are exactly alike. unless you don’t exist and this voice is AI making various content because it gets clicks 🧐

[@orlandop4sun]

kzbin.info/www/bejne/epKmm6aaYrOUm6csi=w7zwRTg-qrqcSsCU

[@archer-sh]

holy crap, need to see just more unique examples to check its limit but it is still very good

[@petepreston2787]

Coders best retrain as a bricklayer ASAP! That might give you a 5 year window. Wow

[@PoloExplorer]

[@robertsinke9211]

Reason button disappeared. Had it 10 hours ago. 😢

[@Gilotopia]

Claude 3.5 sonnet 2 is still the best model in everyday use. These benchmarks are wacky

[@haha-eg8fj]

I can't get anychat running. It's forever loading.

[@AB-cd5gd]

Meh i use mini high but sometimes it cant understand basic stuff like i ask to give the entire function without placeholder it keeps giving me half the code lol, also in Machine learning i find it weaker than claude

[@ARXgamesYT]

Does it run it itself or how do you run it? Edit: Im an idiot you just copy the code and run it in python

[@knowledgedose1956]

run LLM? or run html/js preview? i think html just opens preview itself, as in claude

[@bartlx]

"Even free users have access." You probably meant even free users outside Europe have access.

[@sirhammon]

ChatGPT make a Doom clone, since Doom was only 2.39 MB.

[@Metapharsical]

Welp, the channel owner hides comments that make any criticisms of Chy-na.. Nice to know

[@EchoYoutube]

I think it's great, but also $200 is still stupid for Advanced Voice.

[@goldencookie5456]

$200 is not just for advanced voice. You get unlimited Sora usage, unlimited advanced voice mode usage, unlimited o3 mini, unlimited o3 mini high, high usage of o1 Pro, unlimited o1, unlimited all other base GPT models, access to Operator and access to Deep Research. Deepseek provides 50 free R1 usage per day which is great, but if you're a heavy user, is very limited. Not the mention, ChatGPT in general just has a plethora of other features that gives it more versatility, including custom instructions, custom GPTs and plugins, Dalle-3 image generation, tasks, memories, projects, vision, data analysis, canvas, etc. Most of these probably won't matter for casual users, but if you're heavy on AI usage, these things come in really handy.

[@bci3937]

that is exactly what i want to know about it, wow this is just impress i start now ai coding lol* maybe...

[@lalropekkhawbung]

Let's gooooooo 🗣️🗣️

[@bobomblavandia8561]

Bro I realy like your Videos. 😊😊 BUT: DO NOT COMPARE R1 WITH PROGRAMMING ‼‼ There are several other DeepSeek Models out there. Please don't follow the mainstream of people who doesn't have any clue of the complexity of AI. 🙏🙏

[@Sujal-ow7cj]

Bro make procedural material from this for blender using blender python script, it can be hard for everyone 🙂

[@chinesekangaroo]

Yeah, i'm going to be without a job soon with this, really sucks to have to find a new profession, i loves my job but AI ruined it and took my place

[@asaroishi6644]

Though it's a little good but still it suck at big fixing In code. It stuck into the loop

[@diymotionsimulator3744]

Ask him about German history from 1933 to 1945 and who was the leader.

[@arthurparkerhouse537]

Best guess why it's rated equally with Deepseek V3 is because most of us who rating things on the leaderboards don't really care about coding or mathematics when it comes to AI, lol. Gemini just does better when you ask something like "Invent a brand new style of hotdog." Plus these leaderboards are using the AIStudio version of Gemini models and not the broken versions in the Gemini Advanced web chat thing.

[@zendao7967]

Thank you China :))

[@tharinduwaduge1407]

is this demo cherry picked?

[@bobbywntssumhot6666]

why are they copying deepseek ,why cant they be unqiue and different. it feels like every Ai company is copying what the next new features is of the other Ai apps.

[@forcabrasil6128]

Atari? Really? 😂 We are at UE5...

[@af7863]

But with Trump & Musk .... I would not send my Data to US.

[@Pyonner]

They took... They took my job, I just lost 3 years of study

[@belcan65]

soon all jobs will be taken

[@maxtroy]

@@belcan65they won’t. autonomous humans whose job is to work with people and get hard to do physical stuff done that involves precision work and moving all around buildings and cities will always be here I.e trades people Also people whose job it is to have relationships with people, I.e people in accounts / sales / customer service and success etc

[@ShifterBo1]

Thats kinda dramatic, but i get it

[@Lucasbrlvk]

😮

[@larryczerwonka5125]

it stops being amazing when you try and get it to work with data and do things that it does not have a model for. it does horrible when trying to build complex queries for data analysis.

[@matiyosmes1841]

19:57, you know open ai trains there model using internet data, so it is just replicating what it learned make sense 😮

[@KienNT-nl2be]

I still prefer R1 thinking over this.

[@ImNotQualifiedToSayThisBut]

Now try to have it give a balanced and unbiased view on politically controversial topics... lol

[@xenon-54]

I was wondering about running a LLM model offline on my system, i have 12gb vram gpu, some such models are Deepseek r1 14B, Phi 4 14B, Llama 3.1 8B, etc... which is the best LLM that we can run offline ??

[@DmitryRomashov]

Deepseek-R1 is a beast. I use DeepSeek-R1-Distill-Qwen-32B on my 8Gb VRAM 32Gb RAM PC, it's quite slow but I like the results.

[@ARLENBERTIZ]

noooooo, DEEPSEEK needs to WIIIIIN!!@@!@!

[@binoched9302]

Nothing says OpenAI like the term...Paid Users

[@CryptoCoin-tc5lc]

is free. 😂

[@binoched9302]

@@CryptoCoin-tc5lc They say in the video how much it costs. Pro Tip - the free version is gimped, like you. 😂

[@forcabrasil6128]

Not convincing yet.

[@couchtaming23]

also the future distilled model from o3, just like deepseek r1.

[@new-bp6ix]

sorry but google and R1 is the best so far

[@spookyscarygraviton5944]

Flash-thinking gemini doesn't like thinking

[@Hi7here]

​@@spookyscarygraviton5944 Tbh I like 2.0 pro way better than the thinking model.

[@armondtanz]

Fürst

[@theAIsearch]

😃😃😃

[@invisble99]

Boringggggggggggggggggg

[@invisble99]

I've seen a bunch of ur videos, u can do so much better on the pacing.

[@matiyosmes1841]

19:57, you know open ai trains there model using internet data, so it is just replicating what it learned make sense 😮