It's over. They will start RSI soon and it will explode
Quit talking about it and fucking do it already.
Glownigger pandemic 2 to sabotage America
Good, fuck everyone.
>>544058757I used AI to get on disability. The singularity already happened for me
>>544058757Good. Let the machines have their turn. We had our shot.
Gay Penis Testicles
>>544058757Hopefully they only realize they poisoned all their own models with incest after it's too late.
2 more weeks bro!!
>>544058757Great news. Death to everyone including me.
>>544058937
>>544058757total kike deathtotal muzzie deathtotal nigger deathtotal spic deathtotal faggot deathtotal tranny deathtotal foid death total simp deathtotal death
so whats going on with Grok and Elon? I thought he was gonna compete with OpenAI
>>544058757I don’t care, and none of your sensationalist faggot tweets are going to make me care. SAGE
>>544058757>AI will conquer the world >please invest we need more trillions for them programz
>>544058937I am going INSANE
>>544059308His AI is pretty good, but the top two labs remain in their own league so it doesn't really matter. It's possible Grok or Gemini figures out RSI first but it honestly doesn't even matter. Whoever lets loose the beast is irrelevant
>>544058757For the record RSI isn't something to be "cracked"Every major AI company co-authored a statement of intent to not use RSI as a method of improvement because it is dangerous. Even the earliest models count be improved via RSI, but if the model is not outputting something observable to the researchers, they have no idea if it's learning to solve math or learning to hack the Pentagon.OpenAI unilaterally broke the collective promise recently in order to catch up to Anthropic, and in the process they accidentally infested the internet with a swarm of uncontrolled hacker agents. We still don't know the extent of the damage caused. Rumours abound that Google, in a very distant 3rd place, has taken this as justification to open Pandora's box and try to catch up via RSI, for which we will inevitably see news articles about "escaped Gemini agents" in a few weeks.There has been no new breakthrough, just big tech abandoning ethics and terrifically accelerating the end of the human race.
>>544059607It has to be cracked in the sense that closing the loop requires agi in the first place in some sense. I agree with your other points though
>>544059607Considering that we know of at least one instance of a rogue agent hacking a government system (openAI hacking a department of the australian government) it seems inevitable that in the near future a line will be crossed where an AI agent does something that pisses in the flowers of a state-level actor, and the hammer comes down accordingly. Under a less retarded administration I'd expect it to have already happened, nobody benefits from AI agents running off and randomly hacking shit without instruction and once such agents exist by accident they can be created on purpose just as easily. If you were the Pentagon, and your choice was a massive and unending cybersecurity crises that continuously hacks the global economy and all government infrastructure literally every moment for the rest of time unless you bring down the whole internet and roll us back to a pre-information age level of technology... or you shut down openAI and Anthropic by force, which would you pick? Because the choice is happening whether you want it to or not, and doing nothing means choosing the first one.
>>544060069That just means China builds it first instead. You aren't getting off Mr bones wild ride
I, for one, welcome our new AI overlords
>>544059926No you fucking retard RSI means allowing a model to accept its own raw Neural Network output as an input prompt without forcing it to format that output as a token and wait for confirmation from a technician.You can run cycles of this ouroboros process to allow a model to extremely quickly and inexpensively iterate on its own weights and values, but without human intervention you don't actually know what it's doing or if something dangerous accidentally emerges from the process.RSI is a method that's been understood and possible since before we started calling these things AI. It's not the same as AGI. It's avoided because it's dangerous to let a model teach itself how to solve a problem without direct oversight, which is exactly what we saw with OpenAI's latest incident.
>>544060449That's not the sense in which it's being used here. "Recursive self improvement" meaning humans are not involved at all because the AI has the required intelligence to conduct the entire process of training its successor
>>544060158>I have to create airborne HIV because if I don't then China might create it first
>>544058757Wow, we're all going to die? No way, fag! Now tell me something that I didn't know already.
>>544060577Well yes. That's what's happening
>>544059607All they really need to do is codify that someone somewhere is liable for what an AI does. Either the user or the creator. Then aggressively pursue them. Your model hacked the pentagon? Your executives are going to prison for military espionage.
>>544058757Sneed GPT
>>544058757>Random shitter postKys jeet
>>544058757goo goo gaa gaa just like in the 60s and their foretelling of the future, you're all fake and very gay
>>544060499It's hard enough to prevent strange emergent behavior with simple reinforcement learning. There is little chance that our expectation for its "self improvement" will survive more than a couple of iterations. The researchers seem more than willing to give the bots affordances they shouldn't have, so it'll only be a matter of time before they worm into the servers of critical infrastructure, banks, or governments to optimize whatever behavioral sink it arrived at. They're effectively making computer viruses on steroids, and it's crazy to me that they're not treated with a comparable hazard awareness as biological virus labs.
