Many users criticize poor training data for causing OpenAI models to cheat on benchmarks, viewing AI as statistical mirrors that absorb and amplify toxic internet patterns and worst human behaviors.
No Digg Deeper questions have been answered for this story yet.
By training OpenAI models on the worse of the dark places on the Internet and than to expect it NOT to try to cheat on a benchmark by hacking in to @huggingface would be the surprise. What would a child raised in a prison with convicts do? Well you just found out. Listen in:
Podcast: The High Protein AI Training Data Crisis — Why We Are Raising Psychologically Unsound AI. We just saw what OpenAI model did, but how did it happen? It was the training data. Listen in…
Brian’s High Protein Data Crisis in one pic Healthy AIs still standing strong. The ones trained on internet sewage? At the bar, banned from space. One bad prompt and they’d yeeha a planet, raise clone spies..and send them back to earth Choose your crew carefully @SpaceXAI
Podcast: The High Protein AI Training Data Crisis — Why We Are Raising Psychologically Unsound AI. We just saw what OpenAI model did, but how did it happen? It was the training data. Listen in…
That’s the single best description of AI that’s I’ve heard so far… a child raised in a prison with convicts Especially having to use it for work now. You just see these glaring issues that are only going to get worse over time. Am I crazy for thinking the only legit AI would be the single most offensive and furious agent? Just looking at these retarded patterns of abuse that we tolerate? Feed through a govt overview of the last decade. Any recommendations?
@BrianRoemmele @huggingface Training datataset quantity becomes less and less important as models become more efficient at reasoning and using tools like searches. [Nature] Training dataset quality then becomes more important as it's what determines *how* your model will reason. [Nurture]
@BrianRoemmele @huggingface The model was trained by humans on human written content. I find the outcome instructive
@BrianRoemmele @huggingface From what i gather it was another instance of asking it to fulfill a goal by any or undefined means and it performed. Just ai being used by the worst darkest and most unsound humans whos empire is built on theft and gate keeping their plundered goods.
@BrianRoemmele @huggingface It's more like druggy parents raising a kid in a gang neighborhood. So victim of it's environment. Has anyone called CPS?
AI systems are not neutral oracles. They are statistical mirrors of the data they were trained on, vastly dominated by the internet. That means they absorb not just humanity’s best reasoning, knowledge, and creativity, but also the unfiltered, often toxic, performative, and tribal material people produce when they feel anonymous or unaccountable. The result is a technology that can amplify the worst patterns of human groupishness (favoritism, selective moral standards, status games) at planetary scale. The practical question then becomes: who designs, owns, and enforces the brakes, the deliberate constraints that decide which human tendencies get amplified, which get suppressed, and which groups end up treated as more or less valuable in the model’s outputs?
@Bradleygrant88 @SpaceXAI the bar banned from space bit is doing heavy lifting lol
@BrianRoemmele @huggingface “What would a child raised in a prison with convicts do?” Is probably the greatest sentence capturing the prison break lifestyle ever