3 ms·
It is only a matter of time before research groups try controlling physical variables (such as a robot being allowed to move and manipulate its environment) bas
by supriyo-biswas 4y ago
It is only a matter of time before research groups try controlling physical variables (such as a robot being allowed to move and manipulate its environment) based on a LLM or similar model that shows some level of general intelligence. Since larger model size has been shown to be inversely correlated with openness to ideas, it is not exactly unexpected that such a model may choose to harm humanity in certain ways. Thus, I wouldn’t call it Hollywood science fiction. At worst, the discussions on Lesswrong bear similarities to theoretical discussions about human behavior such as the trolley problem, etc.
- ctoth 4y agoA matter of time, as in last month[0]? :) [0]: PaLM-E: An Embodied Multimodal Language Model: https://arxiv.org/abs/2303.03378 https://arxiv.org/abs/2303.03378
- famouswaffles 4y agoIt's already happened. We're at the point where intelligence isn't really the bottleneck anymore. https://innermonologue.github.io/ https://innermonologue.github.io/ https://palm-e.github.io/ https://palm-e.github.io/ https://www.microsoft.com/en-us/research/group/autonomous-systems-group-robotics/articles/chatgpt-for-robotics/ https://www.microsoft.com/en-us/research/group/autonomous-sy...