@tbenst

I like living in this timeline where a company can *bootstrap* a new medical imaging modality. Incredible ambition and excited to see where this goes

I love seeing a new model for how to go after grand medical and frontier technical challenges. Will be fascinating to see what insights the Midjourney team can find with whole-body ultrasound. https://t.co/KT2vUzC5wL

Whereas neuroscience didn’t have as much to teach the ML folks in the pre-training era, perhaps the continual learning regime is different

Loui’s was more than a restaurant, it was a labor of love and lore. From the 5am Loui’s challenge, to living at Bloui’s (blue house across the street), I spent many a day and night nourished by their food and irreverent art. https://t.co/ql9o59qcce

Kudos to ARC team for coming up a task that’s easy for a human but very hard for AI. Will spur research. As a measure of AGI it’s fundamentally flawed because of course one can train a model to do this. The issue is models not generalizing without training to new task.