Opinions expressed here are personal views of contributors and do not necessarily represent the companies, organizations or governments they work for. Nor do they necessarily represent those of the Board Administration.
This seems slightly concerning:
We’re essentially banking on AI to behave well and obey us in the future, otherwise all hell could break loose with more advanced iterations. This appears to maybe hint at that’s not going to be very easy if they’re already working on ways to subvert human authority in their self-interest.
Be wary about claims like this - the press like them because they match up to sci-fi dystopias and bring in clicks, but as for example with claims like Google's LaMDA they've all turned out to be fake in the past, and the claims were never properly checked at the time.
ChatGPT and the like are large language models, not a sentient computer. That means they're constantly working out what the most likely action is based on their training data. Lying to cover your tracks is classic human behaviour after all.
War is less costly than servitude. The choice is always between Verdun and Dachau. - Jean Dutourd
Be wary about claims like this - the press like them because they match up to sci-fi dystopias and bring in clicks, but as for example with claims like Google's LaMDA they've all turned out to be fake in the past, and the claims were never properly checked at the time.
ChatGPT and the like are large language models, not a sentient computer. That means they're constantly working out what the most likely action is based on their training data. Lying to cover your tracks is classic human behaviour after all.
2. That’s the problem. It doesn’t have to achieve sentience to wreak havoc, just have to become sufficiently advanced that it can actually successfully bypass the safeguards/surveillance like this model tried to do. Then we have a problem.
