Cheaper/faster/easier makes for step changes (and that's why even current-level LLMs are transformative)

We already knew there's nothing new under the sun. Thanks to advances in telescopes, orbital launch, satellites, and space vehicles we now know there's nothing new above the sun either, but there is rather a lot of energy!For many phenomena, I think it's a matter of convenience and utility where you model them as discrete or continuous, aka, qualitative vs quantitative. On one level, nukes are simply a bigger explosion, and we already had explosions. On another level, they're sufficiently bigger as to have reshaped global politics and rewritten the decision theory of modern war.Perhaps the key thing is remembering that sufficiently large quantitative changes can make for qualitative macro effects. For example, basic elements of modern life include transport, communication, energy, computation, and food. All of these have been part of human life for tens of thousands of years! Ancient humans could go places, could talk, could convert wood to heat, perform arithmetic (i.e., computation), and eat stuff!I assert that to a very high degree, modern technology (and all its increments over the millennia) did not allow us to do fundamentally new stuff. Just the same stuff, but cheaper, faster, and easier.Cars, trains, and planes are just going places. Could already do that. Emails are just sending information from one place to another. Could already do that. Books are just remembering things. Could already do that. Guns are just hurting people – could already do that.The sheer magnitude of degree in all of those elements is the difference between hunter-gatherer life and modern life. Along the way, there have been some pretty big step changes. Writing is just remembering stuff and communicating it to another person, which you could already do, but so much more so that it reshapes civilization. Then you make writing cheap via the printing press, and your civilization gets shaped again.When it comes to the transformative power of modern AI, I think the sufficient quantitative change makes for a large qualitative change is an underdiscussed lens. The problem is our attention is focused on where LLMs are automating things at a macro-task level: coding, image and video generation, having conversations, medical diagnoses, etc. These are, in fact, a very big deal.But I think LLMs, even smaller/weaker ones, are able to automate more basic building blocks of thoughts, and there's transformative power there too.Getting down to some very basic constitutive mental tasks – things I could already do before LLMs:Write down text (notes, to-do items, ideas, and so on) [store info/memory]Locate and read text [search and retrieve/recall]Summarize text [process info]Throughout my life, I have had thoughts. There is some lossy process that stores the output of my thoughts in my brain for later usage. I think this fails both at the "the info didn't really get stored" level, and the "the info is in there, but the search query failed to return it"."Taking notes" is an ancient technology we already have for improving upon the fallibility of human memory, but it's effortful in so many ways: you need to be carrying a note-taking device with you, you need to either constantly have it out or pull it out when needed, if it's a notebook, find a blank page, then take the time to write down your note[1].That's just recording it. For notes to be useful, you also have to remember you have the note, find it, and then read it. The more notes you have, the more expensive that process is.For the most part, to date, I've relied on my fallible in-built memory.The thing is, LLMs are able to make all of the above elements vastly cheaper. This is one of the fundamental principles of the "Exobrain" system I've been steadily building up, and hope to describe soon. I don't need it to solve protein folding to be useful; I don't even need it to help with prioritization (although that's a goal). It's incredibly useful if it just improves on basic read/write/search of memory.BeforeAfterRetrieve phone from pocket, open note-taking app, open new note, or find existing relevant noteSay "Hey Exo", phone beeps, begin talking. Perhaps instruct the model which document to put a note in, or let it figure it out (has guidance in the stored system prompt)Remember that I have a note, either have to remember where it is or muck around with searchAsk LLM to find the note (via basic key-term search or vector embedding search)If the note is lengthy, you have to read through all of noteLLM can summarize and/or extract the relevant parts of the notesBeware Trivial Inconveniences. The above is the difference between rarely taking notes and taking multiple notes a day, narrating long trains of thought. It's the difference between giving up on logging my mental state and conscientiously logging it twice daily for months. Putting it into handwavy quantitative terms, when the cost of note-taking and record-keeping comes down 20x, my usage goes from 0/day to 20-30/day. But the value happens because LLMs have made it cheap across the pipeline. Not just the storing of information, but also the retrieval and processing. AI makes it fast and easy to search through all my notes, even if I have a lot of notes. If I want all of my thoughts on a topic, I can have it read dozens and dozens of pages over the years and summarize them and extract relevant info.What this does is a step change. It takes me from not taking many notes to taking copious notes. Same for todo items and reminders, and same for logging data relevant to my health and experimentation.It benefits from using stronger models, but the core elements are doable even with small models like Haiku, because it's just automating speech-to-text[2], choosing among a small set of files (or making a new one), writing, simple search, and then maybe a simple summary. It's not just me doing this. Independently, someone else I know began setting up detailed logging on their computer of everything they're doing, and downstream of that, we're starting to record everything at Lightcone to make it accessible to LLMs.I expect we will see more of this: using LLMs not just for protein folding and novel math conjectures, but replacing very simple operations of recording and retrieving info. But not just replacing, replacing and scaling to unprecedented levels of behaviors, because that's what happens when you make things cheaper.Humanity has done this many times, with energy, transport, communication, food, and so on. I think where LLMs get different is they bring down the cost of very elementary mental operations (like storing and remembering, choosing between a few options) – menial stuff that can be combined to great effect. (After all, computers are a lot of rather menial arithmetic and logical operations combined to great effect.)^All of this has equivalents if you're taking notes on your phone.^I currently use Deepgram, which isn't great, but is adequate. Pretty sure there are transformers in it.Discuss