NEW OpenAI Details Reveal NEW ORION MODEL! (Project Strawberry/Q* Star/ OpenAI ORION Model)

00:00 - Intro: OpenAI's new models
02:33 - Strawberry capabilities
05:00 - NYT Connections puzzle
07:16 - Orion model
08:52 - Model distillation
10:36 - Strawberry in ChatGPT
12:27 - Synthetic data use
15:26 - Reducing hallucinations
18:58 - National security demos
20:50 - Math applications
22:26 - Competitor reasoning engines
23:38 - Sampling and searching
25:47 - AI vs human reasoning
26:50 - Sutskever's research
28:22 - Process supervision
30:57 - Conclusion and summary

Links From Todays Video:

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?

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[voidvoid]

My dad works at Nintendo and he confirmed this. Huge happening in 2 more weeks.

[robroy]

It's amazing how so many of the commenters appear to me to be so jaded already. I've only been using AI for the last 4 months and I am still more amazed today than when I first started. I guess people will be bored within a few months when we do get AGI.

[ordinaryguy]

30:38 Yes, u should do a conclusion even before thinking about making a 30 min long video with 5 min of actual content.

[Nova-Rift]

I'm an AI engineer at a company working close with openai. The owner of my company is close friends with Sam Altman, and was close friends with Steve Jobs. He says GPT-5 will be released Jan 2025.

[HorstMeier]

I'm absolutely shocked by this shocking news

[picksalot]

I think the way forward is what I'd call SLAMs: Small Language Agentic Models. They also need a set of "core/fundamental" knowledge in Physics, Math, Logic, etc. to improve efficiency, and reducing time and costs. They need a way to winnow out false and inaccurate data so they don't waste time sifting through "garbage." The garbage should be saved/cached and periodically reviewed for accurate or useful data was classified as being useless.

[txzgmzp]

Govt(military) will obtain AGI/ASI before all 🫳🎤

[SapienSpace]

Q* is in a USAF Patent ( 5, 608, 843 ) by Baird issued in March of 1997 that has an "Advantage Function". Baird worked with Klopf under a USAF contract and Klopf committed suicide. Reinforcement Learning (Q*) is now being used in an X-62 (a heavily modified F-16).

Why the USAF issued an open patent on Q*, as opposed to having it be classified, is intriguing.

[txzgmzp]

Google gemini can't even figure out military time, so....😂

[josephs]

I'm glad you're getting more traction recently, you deserve it bro 💪

[galailliz]

The shocker of the shockest shook the entire shakery

[JohnnyBlazer]

omfg this guy just babeling on. Man please just gimme info 2 min max, i need a ai watch your video and gime quick summary

[matthia_s]

I don't understand. Could someone try to summarize? It seems they have:
- "Strawberry model": a larger internal model that currently seems difficult to release publicly (perhaps due to problems with the necessary computational power?). Also used to produce synthetic data to train other models.
- "Orion model": GPT-5?: The larger model that would like to translate into LLM from Strawberry in full? Release next year?
- "ChatGPT-4.5 model? ChatGPT-4o LARGE?": the model distilled from Strawberrt smaller where to insert the reasoning in the already existing GPT-4o? Release in the Fall?

Or am I confusing and Orion is GPT4.5/GPT4o Large?

[cndler]

Hahaha thank God I read the comments before watching

[repxam]

I am shocked by this shockingly shocking news 😮

[emperorpalpatine]

Hi, I'm an AI researcher working in Anthropic, we have created a new kind of LLM that is way more advanced than what we have today for the public.
We have found a way to map neurons to our GPUs using a "Feed-forward Uniform Cache K-Unary" graph datastructure .
We don't have AGI yet, but we can generate waifu porn with good cache locality.

[petratilling]

They are calling strawberry a version of gpt 4. This is just their work trying to shift toward accurate phone sized models. This is not a new model. This is not gpt 5.

[theterminaldave]

Impressive, showing that year-old tweet by Jimmy.

[geoattoronto]

We forget significant intellectual capabilities in the human being that can never be replicated by a machine. There is a capacity called intuition that taps into our connection with spiritual resources which depend upon the human spirit .

[Thrain]

Haven’t watched this yet, will I be stunned? We shall see