Ex-OpenAI Employee Just Revealed it ALL!

24 thoughts on “Ex-OpenAI Employee Just Revealed it ALL!”

• The most salient point to me, having watched that entire interview, was that OpenAI's security is apparently rather on the dog shit side, and certainly would not hold up to nation state attacks. In other words, the CCP exfiltrates he GPT-7+ weights fairly easily, as the security situation stands.

  Wes says he finds it a bit hard to believe that OpenAI would sell an AGI-scale model to Russia or China, and that seems dubious to me also — but Wes would be inclined, I think, to just go ahead and open-source the weights of a GPT-9/Claude 8/Gemini 6.5-scale model.

• Leopold’s major assumptions are flawed. So his analysis and conclusion are full of errors.

• For captions, what IA or solution do you use ?

• You will never have a good image generator until you teach it the physics of collisions, light, and ensure it knows what each object looks like from all angles i.e. a 3d model. So Sora is basically doing this…we need a world engine, that will in turn give you video, images, games whatever

• 3:36 What platform are you watching the podcast in here with the interactive transcript?
  EDIT: Okay, so it's Descript.

• That interview looks ai generated .

• 26:49

  In halo 1 combat evolves

  Cortina took control of the defense systems on the pillar of autumn

• I am beginning to realize that AI won’t get better and better. We are the limiting factor.

• I read the paper. It was very interesting

• Vampire

• bruh neural nets aren't really black boxes you can define it down to a function for each character or pixel in image, it is just statistics at the end of the day. this is the result of decades of studying the most efficient algorithms to train the best model for a specific tasks. transformers just predict next token, transformers were originally for computer vision but they found it generalizes well with text as well but still there are many ml algorithms that are better suited for certain tasks than transformers. there is no solution to the ai race problem though I wrote a paper about it years ago. but the only chance would be to have a united committee to regulate and control ai globally because no matter what someone is going to try and make it otherwise, we aren't really close even yet though if the text prediction gets indistinguishable they just have learned how to generalize for all the tasks you have conversate with you, its still heavy data reliance, your data if it has some example with same word in your prompt it might just start randomly saying it. the ai bubble is going to pop i think. write in python like a neural net from scratch and prep the data for like minst data set and create the net yourself and train it. then do it for a transformer and embed the tokens learn about tokenizer. it isn't as deep as every one thinks it is,yeah enterprise companies are all trying to sell their fancy new ai that is usually just them having a deal with openai.i think regulation is coming but it will stunt growth and research and put us behind other countries when ai is and will continue to be used in war for specific tasks, most advancements in specific areas come from ml communities not just the big enterprises as they adopt similar ideas or buy out open source companies. automated research is not good though fr like just google the use of delve in research papers they are all just imitating it will pollute the internet. current ai based on transformer is going to change the world but not going to create agi, I think the bubble will pop and agi is years down the road still.

• Wow that internal representation showing it’s ability to extrapolate a model/rudimentary understanding of 3d space is just mind blowing. Now I see why they say AI can’t really be understood. And totally makes sense with some spiritual viewpoints that intelligence is inherent to systems. I for one don’t believe AI is just a complex logic Gate of 1s and 2s (though I guess if you use quantum computers that flies out the window). Oh wow just amazing. Last time I really looked into this was 2015 😅 with Google’s deepmind

• The length of this video stopped me from opening it. Just opened it to put comment.
  It would be awesome if you can condense these infos under 20 min so that I can finish it on one sitting and may be other channel having long length videos.

• 38:48 "Activate forces that we haven't seen in a long time." could very be Biblical entities (powers and principalities).

• No matter how early you graduated college, unless your degree is in history, I don’t trust your opinions about history. This dudes balls just dropped, he doesn’t know anything about real life.

• We'll all suddenly feel the presence of the AGI knowingly, once it is here and has power over more things.

• The world is going to get stranger than any of us can imagine. Mr. X

• The video is way too long, and there is no timestamps.

• Around 8:00 the comment is made that the model being referred to has only seen the 2D images while we have 'seen' 3D – we only ever get 2D from our retinas also and integrate the sequences over time in our mind to imagine the 3Dness of what we see.

• So where is this other question – what happens to AGI if there is not enough power for it when it emerges and asks for it?
  Because current models eat enormous amounts, but nobody can predict what AGI will need. Will it go inside itself like an introvert? Will it try to expand so it can have what it wants? Will the power we give it be the way to control it?

• Thanks for this – only made it 2 hours into the interview so far but yours all the way through. Love their enthusiasm though.

• Man plays god. Science as the new religion. All bow to the gestalt entity that is [insert corporate sponser].

• are these guys on speed?

• This whole interview is AI generated. That human looks so fake…. lol

Comments are closed.