Neuroscience: Does the brain store knowledge in connection strengths? No it doesn't. Geoffrey Hinton, the LLMs-have-subjective-experiences BSer strikes again. Dear Lord. 🙄 Always take everything coming from Hinton with a grain of salt. The man is a professional sci-fi fruitcake and propagandist working for the fake-AI mafia. He knows nothing about how the brain works. It is not true that the brain stores knowledge in connection strengths. This is what deep neural nets do, not the brain. Short and long-term memory In the brain, a connection is either made or not. The synaptic strength (if any) is a measure of how permanent the connection is. It's not a knowledge gradient. Unless rehearsed repeatedly, the connection is lost. Almost all that we learn on-the-fly is quickly forgotten. The brain remembers very little of what it experiences. It doesn't have to because it can use its sensors to learn anything on the fly in short-term memory. This is why actors, musicians, dancers and athletes have to practice constantly. Eventually, our sensorimotor knowledge is transferred to the cerebellum where it becomes second nature. Timing is critical Knowledge in the brain has to do with timing, i.e., whether or not discrete sensory signals are concurrent or sequential. The brain uses timing signals generated by the hippocampus to time everything. Even the signals generated by the retina are precisely timed by the hippocampus. Indeed, the salience (strength) of a visual signal generated by a retinal ganglion cell is time-encoded. Retinal spikes travel via the optic nerve to the thalamus-LGN where they are decoded before being sent to the visual cortex. Differentiation vs gradient learning The brain uses a completely different method for pattern learning and recognition. It doesn't compress pattern data in connection weights a la deep learning. It discovers what differentiates one pattern from all others. That is, it learns what is unique about every pattern. There is a little was going on in the cortex. Awakened percepts in the cortical columns compete for possession of parts of the sensory space. Whatever survives from this battle is remembered. This clever differentiation method is orders of magnitude more efficient than deep learning. The brain is primarily a massive timing mechanism. Nature knows best. I love the brain. 😍🧠