Did OpenAI FAKE AGI ? (Controversy Explained)
Learn AI Free for the first 30 days- http://brilliant.org/TheAIGRID
Join my AI Academy – https://www.skool.com/postagiprepardness
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website – https://theaigrid.com/
00:00 Initial controversy
01:15 Training details
02:15 Engineer comments
03:19 Benchmark creators
05:52 Sponsored segment
07:02 OpenAI responses
09:08 Training clarification
10:05 Frontier math results
11:05 Benchmark explained
12:05 Expert opinions
13:28 Final thoughts
Links From Todays Video:
https://x.com/rhythmrg/status/1870602244103766258
https://www.youtube.com/watch?v=K-zQPqGAB0g
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
October 4, 2025 at 9:12 pm
Learn AI Free for the first 30 days- http://brilliant.org/TheAIGRID
@artukikemty
October 4, 2025 at 9:12 pm
‘Manipulative and disgraceful’: OpenAI’s critics seize on math benchmarking scandal by Fortune
@AydinIsSus
October 4, 2025 at 9:12 pm
Here's the thing. The score of 87.5% indicates to me that the model was pre-trained on existing ARC AGI tasks. In fact ARC AGI themselves have confirmed that O3 has failed on very simple tasks, and that the new benchmarks ARC AGI 2, would be extremely challenging for O3 (unless they train it with the preset datasets) Change it slightly, change the tasks slightly and see if it will score the same, see if it can adapt to change, learn from change? Again these models are flawed, and are not even close to AGI. Here is a real test, take the O3 model and put it into any given domain-specific environment and see how it functions, see if it adapts, see if it learns anything???
@work1376
October 4, 2025 at 9:12 pm
If the actual AGI is reached, you won't know. The real danger will come when these companies stop making announcements (begging for money).
@DarinM1967
October 4, 2025 at 9:12 pm
Seriously. I guess ya'll another distractor. Probably another shell channel for MS or Elon.
@oliviertakemitsu9583
October 4, 2025 at 9:12 pm
It was very obvious it was BS and marketing.
Just knowing how AI works is obvious that will never achieve AGI by developing that technology. Just doesn't make sense
@ml3054
October 4, 2025 at 9:12 pm
Stock, stock everywhere…..
@KCkingcollin
October 4, 2025 at 9:12 pm
Shorth answer: YES THEY DID FAKE IT WERE AT LEAST 2 DECADES AWAY
@jorgerangel2390
October 4, 2025 at 9:12 pm
LLMs are not an arch fit to emulate reasoning, they simply require too many data and too many systems to try and make them reason
@WiseWeeabo
October 4, 2025 at 9:12 pm
o3 is not just a model it's a system of CoT, reflexion, and self-coherence. there are some misunderstandings in this video based on thinking that o3 who broke the benchmarks is just a one-shot request on just a trained and fine-tuned model, this is not the case.
@CoolDude911
October 4, 2025 at 9:12 pm
The evaluation data was in the pre-training data though. So do we know if it is just better at memorising/recalling?
@nilsd-t5c
October 4, 2025 at 9:12 pm
I think openAI & Sam are panicking, because Elon is gaining a lot of power.
Hence a quick 200$ / month cashgrab + huge promises (AGI). So yes, I wouldnt be surprised if all of this is fake (to some degree)
@sillysnowboot
October 4, 2025 at 9:12 pm
Terence tao my goat
@romangeneral23
October 4, 2025 at 9:12 pm
Yes, they faked it all
@Marksman560
October 4, 2025 at 9:12 pm
Pretentious insecure folks, always fake what they feel they are lacking.
@SXZ-dev
October 4, 2025 at 9:12 pm
All i care about is that it's not AGI and as Chollet pointed out at some point, merely tweaking the questions slightly would cause the score to dump all the way down to 30 some %, even though a child could resolve said questions, it's still just a machine, it's still just following instructions perhaps through another path but it's still the same kind of thing, it's not reasoning it's not AGI, it's not a breakthrough
@supernewuser
October 4, 2025 at 9:12 pm
anti agi fanatics desperately trying to shift the goal posts but once again they've missed the marc..us..
@alkalomadtan
October 4, 2025 at 9:12 pm
AGI will be achieved when whithout the text corpus of the whole internet, a system will be able to do things that a 3-4 year-old child can do. All the benchmarks are the signs of failure of having a theoretical basis of intelligence. Without such a theory, no one really knows when AGI is achieved. Tests are a complete failure to assess AGI. It's like people coming up with various random theories of gravity with trying to compare to experimental results. General relativity has fundamental and deep axioms and that's why it works. AGI probably has a minimal theoretical abstract model that is not known by anyone yet, and which is not LLMs.
@bigdaddy5303
October 4, 2025 at 9:12 pm
Openai still isnt a money making business….and they no longer have best in class models for anything – text, image, video etc. they are gettjng despo
@Music_vibes-kw7xr
October 4, 2025 at 9:12 pm
I think the 25% they said achieved is B.S. lies as they did back in May with the talking and visual model they presented to the world and they NEVER published
@francisdelacruz6439
October 4, 2025 at 9:12 pm
Its simple really. You have a good AI once it wins multiple nobel prizes. Otherwise it's just a nice very expensive dispensable toy.
@RickGladwin
October 4, 2025 at 9:12 pm
François Chollet’s paper “On the Measure of Intelligence” is a good read to understand the issues with benchmarking, gaming tests, and using training data related to specific tasks. If Chollet isn’t immediately kicking the shit out of the latest OpenAI claims, that’s actually a pretty good sign.
That said, the fact that an engineer said “we targeted the ARC benchmark” and Altman, who has immense amounts of money riding on this, and is known to push hype over facts, said “HEY NO WE DIDN’T let me clarify…” is pretty telling.
@roccociccone597
October 4, 2025 at 9:12 pm
o3 is not AGI and now calm down. Watching people talk about this is hilarious.
@NaterFernat
October 4, 2025 at 9:12 pm
OPEN AI lowered the benchmark to say its AGI in order to raise more money
@rgb2647
October 4, 2025 at 9:12 pm
they just redefined AGI to make the world believe they are the first
@manamorphical
October 4, 2025 at 9:12 pm
Let's just ignore the two colors on the bar for the Frontier Math test……..
@geneticjen9312
October 4, 2025 at 9:12 pm
We aren't getting AGI from LLMs
@jahelation8658
October 4, 2025 at 9:12 pm
I lost my pre-trained data set taking a crap.
@TheLiverX
October 4, 2025 at 9:12 pm
I see that it performs much better than previous models. But it is still an LLM. It's a language model, not an intelligence model. At its core it still is a thing that predicts the next token, and "thinks" in tokens. It's been bloated so much it knows practically everything and that's the largest culprit.
@corderi22
October 4, 2025 at 9:12 pm
😂 YouTubers celebrating AGI like blind mice following the hype crumps! Wake up people this is not AGI. 😂
@hudsond15
October 4, 2025 at 9:12 pm
Yeah this is a nothing burger. He meant they were targeting the ability to complete it, not fake it. Their goal was agi, not beating ARC. He almost got them in some serious trouble with microsoft there as their contracf explicitly states they are a subsidiary until AGI is achieved.
@russcontact
October 4, 2025 at 9:12 pm
This sort of feels like being stuck in the weeds. Regardless of nebulous terms or benchmarks, AI is advancing more rapidly every month. At this rate within a year we’ll be into explosive growth territory; when AI starts improving itself. Meanwhile hardly any efforts are being made to prepare society for the next decade. Imagine being in school to become a software engineer – what would your options be at this point? By the time you graduate AI will have wiped out over 90% of the field you’ve planned to pursue. Same with accounting, finance, legal support services, etc. I don’t think most people can properly grasp just how much disruption is just a year or two away.
@MelroyvandenBerg
October 4, 2025 at 9:12 pm
THere is no Twitter. There is only X.
@SportPrediction
October 4, 2025 at 9:12 pm
When Sama mentioned just a few months ago that it will take 5+ years or more to achieve AGI – how can you believe they pulled something off just few months later?
@IshtarCelt
October 4, 2025 at 9:12 pm
Oh yeah, and they used fake Irish diplomatic visas & they come here all the time, without visas & moan & clog up the streets & bars
@davidlocontes3564
October 4, 2025 at 9:12 pm
ARC benchmark is not an AGI benchmark if it has a training data set. It just measures how good an AI is with its specific tasks. Young people are able to solve the tests ARC contains without any prior exposure and without needing a nuclear power plant for training.
@takenserious4554
October 4, 2025 at 9:12 pm
Why would this Gary Marcus dude add the disclaimer "(and she would never do that)" to his Taylor Swift analogy. This guy's opinions are immediately irrelevant to me.
@rumplstiltztinkerstein
October 4, 2025 at 9:12 pm
20 years from now we will look back on today and laugh at what we were claiming to achieve with the little technical knowledge we got…
Comments are closed.