#017: Ken Miller – Podcasting, AI, ChatGPT, Leadership, Learning, and Fathom
The Prompt
Is the Stuck in Time Problem Getting Better?
As the level of learning does increase and reinforcement does take place, do you think the stuck in time problem is more of a challenge for the underlying in the model itself that's a floor or is it something that's just an advancement as the AI becomes much more intelligent? No, I mean, really what it boils down to is a problem with the training method. You know, we gather basically when you're training up these large language models, you're taking a snapshot of the internet at a point in time. So all of its synaptic weights or its parameter weights are being set based on that information when the snapshot was taken essentially. Unless you're continually adjusting those weights, continually training and having
Remember Everything You Learn from Podcasts
