Bengio argues that Scientist AI can be developed relatively quickly by repurposing existing LLM data and infrastructure, differing mainly in its training objective and data representation. This practical approach, closer to maximum likelihood pretraining than RL, aims to instill honesty and reasoning about human statements rather than mere imitation.
Impact: High. This pragmatic strategy significantly lowers the barrier to entry for developing safer AI, suggesting that substantial safety improvements might be achievable without a complete overhaul of current AI development practices.
In the source video, this keypoint occurs from 01:18:05 to 01:21:25.
Sources in support: Rob Wiblin (Host)

