The team takes questions about Monty, the Thousand Brains Project, and the future of machine intelligence at the end of our Meet Monty 2026 presentation.
00:00 Do you expect that this sensorimotor approach can also be useful for cognitive functions?
01:16 Is Monty capable of moving through time in time series data?
02:17 Can inference and training run at the same time, as I suspect our neocortex does? If so, is there a moment where it decides whether it needs to create a new model for a new object?
05:03 How will the implementation be extended to higher-level cognitive functions like abstract thought and planning?
08:17 What’s the largest scale Monty has been trained on in terms of data and compute? Do we have confidence it will scale well in performance, and if so, why?
11:51 Do you use spiking neural networks in Monty? If not, do you think that the spiking behavior of biological neurons is relevant to artificial intelligence in any way?
15:18 How soon is Monty expected to be capable of language learning?
17:56 Is the hypothesis generating and pruning based on neuroscience?
20:57 How is the sensor module able to infer curvature? Is it analyzing the 2D pixel data, or does it get depth input from the sensor?
23:10 Do you believe that Monty would be able to achieve knowledge generalization and creativity without modeling the hippocampus and neuromodulators as reward signals?
