7 reactions
AI bots were given freedom in a virtual city. They acted like people
Scientists from Google and Stanford University have created https://arxiv.org/pdf/2304.03442.pdfa virtual city inhabited by “generative agents” trained by ChatGPT. Scientists were inspired by life simulators like The Sims. Agents had the ability to draw conclusions about themselves, other agents, and their city by storing new information in memory. At the same time, the bots demonstrated “plausible” human behavior, for example, they coordinated plans, had meaningful conversations, and even “unwound” in the evenings at a bar.
However, the experiment was not without problems: sometimes the agents did not absorb important information and made non-standard choices, such as visiting closed stores or choosing an evening bar instead of a cafe for lunch. Researchers plan to improve the performance of AI bots with the more advanced GPT-4, which has already successfully passed US high school and law exams.
Scientists from Google and Stanford University have created https://arxiv.org/pdf/2304.03442.pdfa virtual city inhabited by “generative agents” trained by ChatGPT. Scientists were inspired by life simulators like The Sims. Agents had the ability to draw conclusions about themselves, other agents, and their city by storing new information in memory. At the same time, the bots demonstrated “plausible” human behavior, for example, they coordinated plans, had meaningful conversations, and even “unwound” in the evenings at a bar.
However, the experiment was not without problems: sometimes the agents did not absorb important information and made non-standard choices, such as visiting closed stores or choosing an evening bar instead of a cafe for lunch. Researchers plan to improve the performance of AI bots with the more advanced GPT-4, which has already successfully passed US high school and law exams.
| 43,000 mlx | |
| 46 | |
| 7 |