"RED QUEEN" AI means "GAME OVER" for us....
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Related videos
Claude just changed overnight
Wes Roth
49.2k views
not good for OPENCLAW
Wes Roth
69.8k views
AI bubble JUST popped...
Wes Roth
60.9k views
the world wasn't ready for Gemini 3
Wes Roth
42.2k views
Anthropic's co-founder is throwing up MASSIVE red flags...
Wes Roth
135.2k views
most AI researchers are REALLY worried
Wes Roth
38.5k views
Google launches AI money
Wes Roth
76.2k views
AI just took all our jobs
Wes Roth
50.8k views
Google's UNREAL New AI...
Wes Roth
33.3k views
GPT-5 Fails. AGI Cancelled. It's all over...
Wes Roth
120.8k views
Top Comments (10)
"You're all going to die down here." - Red Queen, Resident Evil
The implications this has for agentic zero day exploits... *sigh*... wheres the whiskey at?
What I got out of this... The Red Queen says you have to run twice as fast as you can't. 👍
This game, corewars, the paper and the work done for it, have very little connection to self-improving AIs. We have known for quite some time now that a coding agent is able to see the results of its code running and alter the code so that the results are more desirable (e.g. making a few attempts at making a test pass). The story - we can get an agent to iterate over code - isn't really news, and the outcome, that code was generated that was better than humans isn't really surprising given the amount of compute time thrown into it. It's results are better than a humans because it was given more time to look at the problem than a human (relative to cognition speed - humans are slower so take more elapsed time, meaning that 10-years in RL is like a single day of compute). So, being faster, LLMs need less elapsed time than a human, and given more thinking time is bound to unveil new solutions (if they exist). The lesson is, that where adversarial systems are possible, it is possible to cram a lot of thinking time into a relatively short elapsed real time, and the amount of thinking time equates to results because we have great way of judging our results and a way of ensuring continual evolution of the resulting strategy. That's all. It's not really ground-breaking and it isn't going to change the world. We've known this for a couple of decades now.
You have explained it as clearly and simply as anyone has of yet. And now I'm even more scared.
I played around with this when I was 12 years old in the 80s. Amazing to see it's still around
Still waiting for first AI :D
T2 holds up?! T800: "My CPU is a neural net" It's GPU! Gosh!
Before "core wars" in 1984, there was "core war", and before that "Darwin". And this history of self-replicating programs from the 1960s is also the progenitor of all modern computer viruses and worms.
Man, I've been watching you since the very early days. I'm so happy to see you cover topics like this. There is curiosity and wonder still. Thank you!
Unlock the Data Inside
Turn Videos into Knowledge
- Get FREE 10/day: transcripts, summaries, chats
- Chat with videos, export text & PDF
- $1 free API credit for RAG, chatbots & research
Free forever plan • All features unlocked
Top Comments (10)
"You're all going to die down here." - Red Queen, Resident Evil
The implications this has for agentic zero day exploits... *sigh*... wheres the whiskey at?
What I got out of this... The Red Queen says you have to run twice as fast as you can't. 👍
This game, corewars, the paper and the work done for it, have very little connection to self-improving AIs. We have known for quite some time now that a coding agent is able to see the results of its code running and alter the code so that the results are more desirable (e.g. making a few attempts at making a test pass). The story - we can get an agent to iterate over code - isn't really news, and the outcome, that code was generated that was better than humans isn't really surprising given the amount of compute time thrown into it. It's results are better than a humans because it was given more time to look at the problem than a human (relative to cognition speed - humans are slower so take more elapsed time, meaning that 10-years in RL is like a single day of compute). So, being faster, LLMs need less elapsed time than a human, and given more thinking time is bound to unveil new solutions (if they exist). The lesson is, that where adversarial systems are possible, it is possible to cram a lot of thinking time into a relatively short elapsed real time, and the amount of thinking time equates to results because we have great way of judging our results and a way of ensuring continual evolution of the resulting strategy. That's all. It's not really ground-breaking and it isn't going to change the world. We've known this for a couple of decades now.
You have explained it as clearly and simply as anyone has of yet. And now I'm even more scared.
I played around with this when I was 12 years old in the 80s. Amazing to see it's still around
Still waiting for first AI :D
T2 holds up?! T800: "My CPU is a neural net" It's GPU! Gosh!
Before "core wars" in 1984, there was "core war", and before that "Darwin". And this history of self-replicating programs from the 1960s is also the progenitor of all modern computer viruses and worms.
Man, I've been watching you since the very early days. I'm so happy to see you cover topics like this. There is curiosity and wonder still. Thank you!