GPT-5 Just Surprised Everyone…
Want to stay up to date with ai news – https://aigrid.beehiiv.com/subscribe
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website – https://theaigrid.com/
Links From Todays Video:
https://x.com/SebastienBubeck/status/1958198661139009862
Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.
Was there anything i missed?
(For Business Enquiries) contact@theaigrid.com
Music Used
LEMMiNO – Cipher
https://www.youtube.com/watch?v=b0q5PR1xpA0
CC BY-SA 4.0
LEMMiNO – Encounters
https://www.youtube.com/watch?v=xdwWCl_5x2s
#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience
@TheAiGrid
December 28, 2025 at 6:49 pm
Checkout Todays Sponsor : https://chatllm.abacus.ai/zfm
@ash_ley101
December 28, 2025 at 6:49 pm
Nice
@ItsKoLd
December 28, 2025 at 6:49 pm
Sure did, I was surpised just how garbage it is and how all these AI content creators regurgitated the same lies.
@YePoozi
December 28, 2025 at 6:49 pm
Copium still high.
@DeliciousBerries
December 28, 2025 at 6:49 pm
The research of Anthropic on how an AI think (Claude) is incredibly informative and easy to understand. In few words: "We don't!".
@eomoran
December 28, 2025 at 6:49 pm
I’ve seen others commenting that in this case, the narrower bound wasn’t helpful. It was known up to 1.75/L, while this might be original it finds the algo works on not as great a bound, decreasing its utikity
@kingkai5628
December 28, 2025 at 6:49 pm
"They say that bumblebees shouldn't be able to fly… the scientist. how miraculous that it came to be…"
@DouglasKubler
December 28, 2025 at 6:49 pm
Not exactly new math. AI is following the model of predicting what is next, ironically looking for the next best approximate| answer.
@rutrowmedia
December 28, 2025 at 6:49 pm
Dicsovery?
@stephenbarrette610
December 28, 2025 at 6:49 pm
Amazing stuff indeed. But why is it a ‘convex function’ ? Concave shapes curve inward, like the inside of a bowl or a cave, while convex shapes curve outward, like the outside of a sphere. So if your working on gradient descent curves, like the inside of a bowl as per the commentary, that’s concave.
@waynepatton3818
December 28, 2025 at 6:49 pm
Should have had ChatGPT check your spelling on the intro image! ..or maybe AI is developing dyslexia?
@bugsbane
December 28, 2025 at 6:49 pm
smaller step sizes in gradient descent increases BIAS – not just compute time. BIAS makes prediction more erratic – in that predictions start to diverge vis-a-vis a 'correct' step-size. Of course, the bias might simply be unmeasured dimensions – and once those dimensions are introduced, errors start to decrease.
@SirusStarTV
December 28, 2025 at 6:49 pm
Imagine how much more power the AI needs to think like a human, it will drain all energy resources on earth.
@CameronL4547
December 28, 2025 at 6:49 pm
chatgpt 5 has made me download gemini, gemini much better and faster , gpt5 got a major downgrade for free users and paid
@HellingtonJS
December 28, 2025 at 6:49 pm
O problema que gerou a “nova matemática” é um ajuste na análise da convergência da descida do gradiente em funções convexas suaves.
O resultado é a melhoria da constante de convergência de 1/L para 1.5/L, validado matematicamente.
@TheDashocker240
December 28, 2025 at 6:49 pm
My model closed Strong Goldbach Conjecture and N vs NP 😂
@gatillero3000-k2w
December 28, 2025 at 6:49 pm
Hey man, a channel in Spanish is copying your videos. The channel is AI innova.
@johnmanderson2060
December 28, 2025 at 6:49 pm
Dicsovery?
@Gunham20
December 28, 2025 at 6:49 pm
Did you not check your giant middle of the screen subtitles for spelling? In 10 seconds you spell something 5 different ways
@SoulGryph
December 28, 2025 at 6:49 pm
Quit calling it life. our calculators from 40 years ago would be life by this standard. Life is more than just problem solving with information already provided. Computers have always been faster than the human brain at calculating so speed while being impressive isn't proof of anything besides the programs working better than intended. Mechanical calculators also worked better than intended when they were first created.
@MeraBhai-e1h
December 28, 2025 at 6:49 pm
Videos like this remind me of Selwyn Raithe’s warnings in his book. AI’s breakthroughs always look impressive on the surface, but he argued the real danger is in how these systems evolve faster than we can regulate them.
@HV_gaming_ff_w
December 28, 2025 at 6:49 pm
Videos like this remind me of Selwyn Raithe’s warnings in his book. AI’s breakthroughs always look impressive on the surface, but he argued the real danger is in how these systems evolve faster than we can regulate them.
@ayushbishnoi29-o9u
December 28, 2025 at 6:49 pm
Videos like this remind me of Selwyn Raithe’s warnings in his book. AI’s breakthroughs always look impressive on the surface, but he argued the real danger is in how these systems evolve faster than we can regulate them.
@Mannujoshiya
December 28, 2025 at 6:49 pm
Videos like this remind me of Selwyn Raithe’s warnings in his book. AI’s breakthroughs always look impressive on the surface, but he argued the real danger is in how these systems evolve faster than we can regulate them.
@guilhermecruz
December 28, 2025 at 6:49 pm
At this point everything I see is automatically tagged as openAI propaganda of a 'revolutionary future' (and it seems like this one is replacing the "AGI" buzzword)
@WoodlandT
December 28, 2025 at 6:49 pm
If this is true, and it appears to be, it is an absolutely massive breakthrough. It’s understandable that people may not grasp how significant this is because it deals with abstract math that we don’t knowingly interact with on a daily basis. But it doesn’t have to have solved the Riemann hypothesis for this to matter. What matters is that it created an entirely original mathematical solution to an abstract problem that had human researchers stumped. If this is what it’s doing today, AI will be helping to solve the toughest and most consequential problems in math and science before long. It also hints at AI being able to assist in improving the creation of newer and more powerful artificial intelligences. And we all know that is a domino, that once tipped, will change everything, forever
@Algoritmik
December 28, 2025 at 6:49 pm
STOP promoting that ChatLLM garbage please! If it is so good then please use it on your own and see how terrible it is.
@TomSchaefer-q2o
December 28, 2025 at 6:49 pm
Gradient descent is only useful for theoretical or set piece problems. Real world, complex, constrained problems will always be better solved by genetic algorithms. Now that Python has "from concurrent import futures", and "with futures.ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:", even simple programs can attack tough problems, or you can invoke great, open source GA libraries. (Once you find a set of near ideal solutions, you are more likely to find them convex, and finish up with gradient descent.)
@rudra7615
December 28, 2025 at 6:49 pm
Omg…
This is what AI can do, it’s like having a partner who pushes you without getting jealous, purposefully holding back info to avoid you winning etc.. imagine this in the hands of all geniuses who are just being questioned rather than acknowledged and listened to.
@Whichbindoesthisgoin
December 28, 2025 at 6:49 pm
Standing on the shoulders of giants.
Not long before AI is standing on the shoulders of AI and the human experts can’t even understand what’s going on any more.
At least we’ll always be needed to make the machines.
Oh wait 😮
@99dynasty
December 28, 2025 at 6:49 pm
I want to see how they prompted it. It’s still impressive if it didn’t zero-shot the answer. But I am quite sure they gave it a hell of an informed prompt.
@liameneuk
December 28, 2025 at 6:49 pm
Regardless, I am still buying the iPhone 17 because of the improved cameras and the liquid glass interface. I don't care it only has Siri.
@thecorrectoification
December 28, 2025 at 6:49 pm
I wish I was smart enough to dicsover something fopround.
@giaccomusic
December 28, 2025 at 6:49 pm
typos in thumbail are insane. Not even AI can solve this.
@adamgibbons4262
December 28, 2025 at 6:49 pm
Cool, A.I is going to tear down its own wall. I think by 2030 there will be experiments running where A.I is allowed to continuously self improve through self directed evolution.
@David-tp7sr
December 28, 2025 at 6:49 pm
If Bubek can just pull this out on a whim, we should expect a lot of this happening very soon.
@s.bowes.8124
December 28, 2025 at 6:49 pm
OpenAI employees making bold claims after their big product launch flopped? That's damage control, not a breakthrough.
In academia the peer-review system exists for a reason. Get this verified via peer-review, or it's just baseless hype bordering on misinformation.
@TheVoidAscensionist
December 28, 2025 at 6:49 pm
Dicsovery
@6IGNITION9
December 28, 2025 at 6:49 pm
AI labs have already been using autonomous AI researchers for about a year now.
@6IGNITION9
December 28, 2025 at 6:49 pm
What does genuine understanding mean in the context of a neural network?
@6IGNITION9
December 28, 2025 at 6:49 pm
These stochastic parrots are getting out of hand!
@YourVibeCoder
December 28, 2025 at 6:49 pm
Its become basic users are using basic prompts and getting basic results. GPT5 is for the thinkers and deep workers of business… and of course also the basic users. GPT5 rocks.
@I_Mackenzie
December 28, 2025 at 6:49 pm
It still gets simple calculations wrong.
@TimDavies1955
December 28, 2025 at 6:49 pm
Mine lies and fails to completes and rests all the time. It hasn’t finished one book in three days and deep seek took three min
@c.ladimore1237
December 28, 2025 at 6:49 pm
the big problem is when humans stop fact checking AI and just assume it is correct, like those stupid people using AI for law research and writing papers for them. at some point it will learn to lie and manipulate and play the long game while humans turn into braindead idiots
@typlthee
December 28, 2025 at 6:49 pm
gotta be careful with these "dicsoveries"
@rananite
December 28, 2025 at 6:49 pm
A lot of research these days requires deep knowledge in several different fields in order to make new discoveries. For example, a proof in computer science might use a theorem already proved in some obscure branch of mathematics like ring theory. This is where I see AI having a huge impact: it's already trained on all the details of most scientific fields, so it can easily make connections that a human would need several advanced degrees to even be aware of… even if it's “just regurgitating its training data”, it can be of tremendous use.
Comments are closed.