Does learning slow time perception?
UPDATE 06-27-25: I had some ideas a while ago and wrote a lot of this in one shot, so apologies if it’s a little chaotic. Given all of the recent LLM developments, figured I'd share it and see what the hivemind thinks.
How does my brain hold onto the lyrics of a huge amount of songs without any conscious effort? Why am I able to recite my son’s favorite book from memory alone? How does that effortlessly fit alongside a lifetime of learnings from engineering and building businesses?
I think, fundamentally, our long term memory isn't a database, but a model. I think this is how the brain seems to be able to save an exponential amount of information over time - it compresses raw data (short-term memory) into a model, and then discards the raw data.
This would explain why it can be so hard to return to a flow state after sleeping - after the raw data is discarded, your memory is no longer coming from the raw information. Memory after a night of sleep is just a prediction from a model of what probably happened. The fidelity is significantly decreased.
This two step process, first saving raw data & later training the model with that data, would also explain the divide between short term memory and long term memory. Issues with writing/reading raw data surface as short term memory issues, problems re-training (or evaluating) the model surface as long term memory problems.
This process could also be how we connect dots between concepts - when the model training process notices two things are very similar, it compresses them to the same pathways, and to us it feels like a eureka moment. But that doesn't happen until the re-training on the new raw data is complete - this is why the breakthroughs often come in the shower, or in bed, or in our dreams. It’s a background process that’s determining how all of the pieces fit together.
I think the short-term raw data limit might be the same as context size in an LLM. Like LLMs, we can only handle so much context at once. Similarly, I think mental tiredness might be how it feels when one approaches the raw data storage limit. When your brain has enough on it’s plate, it’s a bit like you’ve just had a huge meal - you’re tired and you need time to process.
Thinking fast and thinking slow
There are parts of our “memory” that are so deeply embedded they take the shape of what we understand as “muscle memory”. Learned behaviors that are so ingrained, they happen without any conscious involvement at all. The brain builds these from birth - basic control of your limbs, balance, forming sounds with your vocal cords. Things you never experience consciously, but are happening all the time. If you had to consciously control all of these foundational parts of human existence, you’d have no room to learn more - your consciousness would be too busy focusing on putting one foot in front of the other, literally. The raw data/short-term memory storage system would be completely overwhelmed with data that, when processed, doesn’t change the model at all. At some point, any signal (actually new information) would get lost in the noise. So, how does it solve for that?
My best guess is that it filters inputs (and outputs) from the raw data system entirely. It could accomplish this by running the input against the model, and skipping the write if a confidence value from the model is above a (likely variable) threshold. Basically - when your brain is “thinking fast”, it’s actually attempting to bypass the short-term memory system entirely.
Most of the time, we don’t even notice the missing chunks of short term memory - the things missing are completely inconsequential parts of the day. But most people have probably noticed their brain doing this at least once or twice. For example - have you ever started driving somewhere and accidentally driven to work/school/home? Or just not remembered the drive from home to work and vice versa, at all? This is your brain filtering raw incoming data that will probably not change the model.
This is how, over time, you can go from feeling like the challenge of reading one word is an insurmountable mountain that takes all focus and energy, to being able to read the subtitles faster than the characters are saying them. The brain is essentially compressing the incoming feed, and it gets better at that compression as your model of the world gets better at predicting your life.
So what happens when your model of the world is near perfect for any given subject? Let’s take language, for example. At first, you don’t know what a letter is. You can’t possibly grasp the concept of reading a word, let alone a sentence. But then you start to grasp the fundamentals. Eventually, you know most words, you have a grip on most grammar, and you start to approach knowing enough of the language well enough that you almost never come across a phrase, grammar, or word you’ve never seen before. Your model of English is effectively Turing complete. So, what happens? Your brain skips over reading letters, reading words, and parsing grammar entirely. Those steps disappear into the subconscious. The only data you process is the ideas, the unfamiliar concepts being explained with those words & grammar. This is why some people are able to read so comparatively fast to others - their model of the English language is advanced enough that reading is truly automatic.
Now if you think about the compounding benefits of that, you start to poke into some really interesting exponential knock-on effects. People who have a Turing complete English language model are able to read more in less time. If they take advantage of that fact and read more and learn more about the world as a result, their model of the world becomes more refined, and they’re even faster at processing new data, which allows them to learn more at higher abstraction levels, and so on. Power-law outcomes.
Time perception
Think back to that example of driving to work/school on autopilot - having very little or almost no memory of commuting on a given day, because it’s so familiar. That has an extraordinarily interesting detail, which is your time perception. That ride can feel like an instant, even if it’s an hour. I think it’s possible that when your short-term memory (raw data system) is bypassed entirely, you don’t consciously experience those moments at all.
The implications of that would be significant. The suggestion is that too much habit and routine will literally, at some point, pass by in an instant. Once you’ve learned everything your brain finds it relevant to learn about a given task, repeating that task no longer happens consciously. So (from your perspective), time literally slows down when you’re learning new things.
The phrase “time flies when you’re having fun” may have some grounding in reality - if the things that we find most “fun” tend to be more familiar, those experiences would write less raw data, and they would literally seem to pass by faster.
tl;dr: Learning might be the source of conscious perception of the passage of time.