"In order to understand bird flight, we have to understand aerodynamics; only then does the structure of feathers and the different shapes of bird's wings make sense." -David Marr
A key principle of aerodynamics is Bernoulli's principle which relates speed of a fluid and pressure. This principle can be derived from conservation of energy but also directly from Newton's Second Law of Motion (F=ma).
Marr said the above quote in relation to studying the brain and how it works, which begs the question, what principles like conservation of energy are also in action in the brain? Conservation of energy is probably too broad since Bernoulli's principle breaks down if there's turbulence or thermal radiation and it's hard to imagine anything like laminar flow in the brain.
The visual cortex has arrays of neurons (or neuronal columns?) which detect specific orientations of objects in the visual field.
Scientists could see this because they could push this electrode into the tissue while showing the live animal subject lines of various orientation. What's unclear in the lecture and in this image is whether what they're measuring was a single neuron or the spikes from a handful. Is each of these vertical slices a cortical column?
This pertains to feasibility of bio-based AI/ML modelling because there are many fewer cortical columns (150k) than neurons (20b) in the brain. Still more interested in developing a small component with feedback loops, etc but thought this was interesting.
Saw the term 'research grade' and wondered what it was.
https://forum.inaturalist.org/t/does-research-grade-actually-mean-anything/37293
Bird watchers are wonderful people. Check out inaturalist.org
A sampling approach per relay might be better than checking every message. Do relays check?
It seems our brains are constantly making predictions about our environment. To illustrate, Jeff Hawkin's often uses a coffee cup and the expected sensations when going to grab it.
What seems fun about this idea is how it relates to the abstract. What if there were an object that defied classification, never satisfying the predictions that the brain has made? It wouldn't have to be a physical object either. Let's try it with language.
Right now, as you're reading you're trying to follow along and predict what I will write next. Paper straw. Doubt you were expecting that. I suspect your brain is now kind of backing away slowly because man I could write anything. A literal madman wiith a keyboard. I like plastic.
Instead of surprise, how about ambiguity? It is on the surface. What is it? What surface? Is it moving or what? What kind of thing even is it? Might be fun to write something longer that plays more with ambiguity and surprise. For now I'll just say it's a paper straw on the table. GM
Thanks for mentioning cortical columns. I feel like nature is the ultimate guide so was fascinated to research in this direction. I understand cortical columns are interconnected in potentially heirarchical ways. I'm not sure if my project is closer to develop a column or to model a small network of them. I saw that Hawkin's company has cultivated a community around their work and has also released a lot of code so plenty to get into here.
I'm not sure if you've looked further into it but noswot's WoT is initially built from follows and reports. The WoT algorithm is based on one from freenet which instead of 6 degrees of separation uses an exponential falloff of influence. If you're interested in trying a different source than follows and reports I can point you to where that's defined in code or maybe add it to noswot.
I started with follows and reports for simplicity. It operates on the least amount of data and allows us to see if some kind of WoT can prove out the value of this general approach. That being said intake can be expanded to get reactions and replies for example if you wanted to attempt to detect topics and rank by bucketed engagement. It can quickly get complicated.
The brain with its white and grey matter and cortical columns is extremely fascinating. Scientists have figured out so much which can't have been integrated into AI/ML. Still not sure if any of it can be implemented with more simplicity but have some directions I'd like to go, namely integrating more feedback loops at different scales and working at the beginning with very small data sets to see what resonates a kernel of a system.
One other thing I learned relating to the heat factor is that a lot of what the brain is doing in early development is automating expensive processes it learned. Keeping this in mind, it makes sense to create a microeconomy in the model where things like building links has a cost.
Thought it might be interesting to use a relay as a data store. It'd be kind of cool if I could treat a relay as an sql backend, importing a schema and going from there. For the trade-off of publicity and yeah a little buzz, I'd accept some query limitations.
Oh that's kind of awesome! The same networkx library is in use for both noswot and this project so I suspect there are many possibilities for cross-pollination. Dude, imagine if neurons had reputation. I'll put the presentation in my queue.
I find it fascinating that a single occurrence of a new word can be remembered with context. I recall from child development psychology that kids have a hard time recalling where they learned something, so perhaps this is a mechanism that comes coincident with the pruning that happens from adolescence on into adulthood. Of course there are countless other processes at work. Just a thought.
Complex, pointless, or unwritten rules are toxic to groups.
Ok I removed nostr.lol from noStrudel and Amethyst (posting new relay lists from each) within the same couple minutes. Here's to 100% relay send success! ☕ GM and stay classy
A lot of my excitement about new possibilities for AI/ML come from the three Joscha Back interviews by Lex Fridman and one with Jeff Hawkins with the 1,000 brains theory.
This latest AI/ML project is attempting to model and simulate the self-organizing principles in the brain. We think of the brain as efficient from a power perspective but that's mainly because of electrochemistry. The numbers of neurons and connections is astronomical so a direct simulation is not promising. The big question is can components of the brain be identified and implemented in such a way that those numbers become reasonable while retaining the magical non-linearity?
Preowned gear is amazing.
Allow me to rephrase to see if I understand. The embeddings refer to an internal state (like a vector representing activation levels of neurons) reached after feeding in the words in the context window.
So you're saying I could compare the state which creates a prediction to the state that is achieved by inputting only the predicted word?
To start it'll be word association. I prototyped a network with a set of mindmap-like files but want to start with nothing but a tokenizer and train it to associate appropriate words.
Working on a new AI/ML project partly to learn but also to see if something interesting will arise from "scratch."
A missing piece this morning is that of a reward system. I don't know how to tell this thing what's good and what's bad programatically. Seems like the perfect time to "do things that don't scale" and execute the reward function myself. I figure I'll let it ask me after each step for 👍 or 👎. An app would be cool so it could just be taps but I'm a cli-guy so maybe I'll go with vim bindings.
I wrote some glue code (https://pub.dev/packages/amberflutter) to be able to easily use https://github.com/greenart7c3/Amber from Flutter. I hope someone can try it out and give any input!
Here is a demo of getting my pubkey from Amber, and I also took the opportunity to use nostr:npub1r0rs5q2gk0e3dk3nlc7gnu378ec6cnlenqp8a3cjhyzu6f8k5sgs4sq9ac design for the profile page to revive my flutter "skills" 🤣
https://video.nostr.build/65ac0db4ec0b535a1d3b96e30e63509acc6fb8fe96baf3d3c17cfad4ad3d317c.mp4
Nice work! Amber looks great.
Question: If a web application doesn't sign events or decrypt anything is it possible to request only the read pubkey permission?
If so I'll definitely try to integrate with noswot.org so the app doesn't appear to request sign/decrypt permissions when they're not needed.
Howdy nostr:npub1ydhmq6k7zshd5chdfrmpqpy44jyn02z7amw98knaqascw2flnpzq5serpq 🤠
What do you think of the IOT NIP
It seems stalled, any idea why?
Got a bit tired of the nostr echo chamber so unfollowed some popular accounts and used pleb mode on noswot.org to get more variety in my feed. Looking forward to seeing more diverse, less Bitcoin-focused content.
That's one popular narrative. Setting aside the dominant centralized social network paradigm, I think it's unwise to bet against Elon.
No, if you go to the Notes view then look in the upper right.
yeah that was probably hit ha ha
Community notes on Twitter seem interesting. Supposedly publicly verifiable. Joined and will report experiences if notable.
The relay health screen in nostrudel. Looks like a little lamp icon in the upper right. Shows which relays are giving me the notes I'm reading and which are just there.
Gotta stay vigilant with the unsubs.
It uploads it to your relays. Helps paper over reliability issues and helps deal with the common neighbor requirement.
Have repaired favorite backpacks many times. No machine but a big fan of needle and thread!