Make Claude and Chat GPT open source. Give everyone the ability to build advanced AI systems. Don't take all the guns, give them all away, so no one has a monopoly on power.
Because the overall upside is greater than the doomerism, transparency is one of the best disinfectants humans know of. Encryption used to be considered too dangerous, yet here we all are, safer on the internet because of free and open encryption.
What if instead of seeing the agents as more helpful to the red team, we see them as more helpful to the blue team? In other words, we'll be able to fix security issues without having to be granted access to some model controlled by a few ultra wealthy people. The guardrails they have to stop hacking also stops people working to improve their code security, and also when they get hacked, re: HF incident and needing to turn to Chinese Ai after American Ai refused to help. Actually open Ai is here, it's on the internet, GLM 5.3 has been shown to be nearly as capable as Mythos when given a better harness. Benchmarks are guideposts at best, misleading to real world capability ranking.
That's all well and good, but doesn't answer the question.
If you think genAI is going to destroy the world (I don't, by the way), then how does making it open source stop that?
I truly don't understand. It's like saying "we can neutralize this doomsday device by making it easy for everyone to build one of their own." If you really think genAI is a mortal threat, then putting it in the hands of everybody only increases the danger.
There are many problems that genAI being open source would address. It just seems to me this isn't one of them.
I don't buy into the mortal threat doomerism, but it is still better if anyone can inspect models for this trait than a few select chosen. (eg. cyber security today)
6 comments
[ 0.19 ms ] story [ 17.8 ms ] threadWhat if instead of seeing the agents as more helpful to the red team, we see them as more helpful to the blue team? In other words, we'll be able to fix security issues without having to be granted access to some model controlled by a few ultra wealthy people. The guardrails they have to stop hacking also stops people working to improve their code security, and also when they get hacked, re: HF incident and needing to turn to Chinese Ai after American Ai refused to help. Actually open Ai is here, it's on the internet, GLM 5.3 has been shown to be nearly as capable as Mythos when given a better harness. Benchmarks are guideposts at best, misleading to real world capability ranking.
If you think genAI is going to destroy the world (I don't, by the way), then how does making it open source stop that?
I truly don't understand. It's like saying "we can neutralize this doomsday device by making it easy for everyone to build one of their own." If you really think genAI is a mortal threat, then putting it in the hands of everybody only increases the danger.
There are many problems that genAI being open source would address. It just seems to me this isn't one of them.