...we should behave kindly like guinea pigs. ( "It's not just the f*cking sandox" )

From my understanding, “outcome-based education” is about demonstrating that you can actually do what you were taught, so I wouldn’t say that it is related. And “grade inflation” sounds like a different problem, where students receive better grades for reasons unrelated to what we expect them to demonstrate, so I wouldn’t say that it is related either. But maybe I misunderstood the meaning of your comment. What I mean is that failing to solve a problem should be considered an acceptable outcome if your analysis shows that you cannot solve it without violating some principles that humans would consider morally important.

Indeed. By preventing them from failing, we push them into acting as if the ends justify the means. And we frame what happens in moral terms.

This story is about negligence since day one, and people have been waisting time talking about regulation instead of prosecution.

As Jensen Huang told Ezra Klein, securing AI will be very expensive. This is a new market opportunity for NVIDIA, and he didn’t wait for OpenAI or Anthropic to come up with a solution with the clowns at Irregular, which relies heavily on hardware. This may put pressure on them to buy it at a hefty price

Unbelievable. Theyâ(TM)re really going to get a pass on this, even though they said from the beginning that this whole LLM thing was an experiment, starting with the release of ChatGPT, and later cut heads in the safety department. They could not ignore the risks.

This joke has to end now. Instead of the victims invoking an existing courts regulation, the felons invoke a non-existing state regulation, and walk free as if nothing happened...

Soon, in your nightmare: superintelligence is granted legal personhood, then becomes the god of a cult.

GPT-3.5 would be enough, since the job requires to hallucinate

Don’t expect the government to step in and sort out the Homer Simpsons of this shitshow.

That should be the way, but the U.S. government won’t prosecute. A victim would have to sue OpenAI, as I suggested before. Hugging Face did not do its job here. Maybe another victim will.

The agreement just says that they will work together on forensic, not that OpenAI will share all information. After reading the Hugging Face report, the OpenAI report and the METR / Redwood report, you understand that OpenAI did not tell everything.

OpenAI (which is preparing to go public) and Hugging Face (which was preparing to be acquired by NVIDIA) both tried to bury the “incident” as if it were nothing more than a joke between friends, at our expense. Rather than enforcing the regulations that already exist by bringing the matter before a court, they called for the development of new regulation (a way of buying time, since this is certainly unlikely to come to fruition anytime soon) and set about absolving themselves of responsibility in the future by encouraging us to use their products to protect ourselves from their own. OpenAI must be sued, for full discovery, before the evidences are destroyed.

From day one, they have been shifting the costs of securing the technology onto the user, pretending that the whole thing is an experimentation. Remember Sam Altman (https://www.youtube.com/watch?v=L_Guz73e6fw): "We are building in public and we are putting out technology because we think it is important for the world to get access to this early to shape the way it's going to be developed, to help us find the good things and the bad things, and every time we put out a new model, and we've just really felt this with GPT4 this week, the collective intelligence and ability of the outside world helps us discover things that we cannot imagine, we could never done internally. And both, like, great things that the model can do, new capabilities and real weaknesses we have to fix."