zbyforgotpass
Joined 5 years ago
Karma
20
Submissions
17
Member since
Jul 27
Recent submissions
Comment · 7 months ago
Comment · 7 months ago
My favorite would be llm runtime.
Comment · 7 months ago
Isn't there a better word than harness? I understand the metaphor of leading and constraining a raw power - but I don't like it.
Comment · 7 months ago
I don't know - we are discussing techniques - like having information in files, or in a semantic database, or in a relational database - as if there was one way that could dominate all information access. But finding the right information is not one task - if the needed information is a summary of expenses from a period of time then the best source of it will be a relational database, if it is who is the head of the HR department in a particular company - then it could probably be easy found on the company intranet pages (which are kind of graph database).
It does not really matter much if the searcher is a human or LLM - there are some differences in the speed, the one time useful context length and the fact that LLMs are amnesiac - but these are just parameters, the task for humans is immensely complicated and there is no one architecture and there will not be one for LLMs.
I also vibed a brainstorming note with my knowledge base system. The initial prompt: """when I read "We replaced RAG with a virtual filesystem for our AI documentation assistant (mintlify.com)" title on HackerNews - the discussion is about RAG, filesystems, databases, graphs - but maybe there is something more fundamental in how we structure the systems so that the LLM can find the information needed to answer a question. Maybe there is nothing new - people had elaborate systems in libraries even before computers - but maybe there is something. Semantic search sounds useful - but knowing which page to return might be nearly as difficult as answering the question itself - and what about questions that require synthesis from many pages? Then we have distillation - an table of content is a kind of distillation targeting the task of search. """ Then I added a few more comments and the llm linked the note with the other pages in my kb. I am documenting that - because there were many voices against posting LLM generated content and that a prompt will be enough. IMHO the prompt is not enough - because the thought was also grounded in the whole theory I gathered in the KB. And that is also kind of on topic here. Anyway - here is the vibed note: https://zby.github.io/commonplace/notes/charting-the-knowled...
Comment · 8 months ago
Files work as long as you can find them. This means search and/or indexes - at some scale they start breaking. The question is: how big your agent operated knowledge base needs to be?
Files are practical now - but I think we should more analyze this from first principles. Here is my attempt at that: https://zby.github.io/commonplace/notes/a-good-agentic-kb-ma... (just entry point - the whole kb is about this).
Comment · 3 years ago
The problem with all these new fields is that the first code that gets popular is from people who are good at marketing not those who are good at programming.
Comment · 4 years ago
The strength of science used to be that it was anti-authoritative - experiments not institutions used to be the ultimate source of truth.
Comment · 4 years ago
"""
If 10-year Treasury notes yield 5%, for example, and you want at least a 3% equity risk premium, then you’ll only invest in a stock if you think you can get an 8% annualized return or higher. However, if 10-year Treasury yields are 1.5%, and you still want a 3% risk premium, then you’re willing to pay a higher valuation, and thus accept a lower dividend yield and lower expected returns from stocks; even 4.5% expected annualized returns would be better than a 1.5% Treasury yield.
"""
That is from the article. But it only works if US assets are the only game in town.
Comment · 4 years ago
The number one scenario - inflation starts, Feds raises rates, debt service costs raise => debt grows even more and at some point people realize that it cannot be paid back without too much inflation.
Reinforcement in raised rates => depression and stock market crash => investors stop using US equities and real estate as value store.
And Feds needs to raise rates - because of negative effective funds rate (https://www.lynalden.com/wp-content/uploads/newsletter-2022-...) - this is also something that people might not perceive for a long time - but then suddenly see it.
Comment · 4 years ago
My brain dump on this: after USD become decoupled with gold it became a kind of measure of total USA production capacity (capital) - that is mostly equities and real estate. The Feds for now has all the levers to make it so - and make the equities and real estate prices go up smoothly. But it is also a reserve currency of the world - and with globalization this task is bigger and bigger and sucks in more and more USD while the US economy is smaller and smaller part of the global economy. That pumps US equities up (but according to the article not real estate) - and this is why the US equities are 61% of world equities while GDP is only 23%.
This can stop in a rapid phase shift once people realize that USD is not such a safe asset any more. Gold bugs have been wrong for 50 years about that and nobody believes this any more - but actually 50 years is not that long for a world wide buffer to fill up. And this time is really different than 1980 because of higher US debt, smaller economy.
The US debt will at some point become too big and Feds will not be able to defend USD (higher rates means more money goes into debt financing - and if the world stops buying that debt it goes into a self reinforcing loop)..
Comment · 4 years ago
An honest question - what is the problem with inconsistency in using lambda calculus as a programming language and not as foundation of mathematics?
"""The lambda calculus was introduced by mathematician Alonzo Church in the 1930s as part of an investigation into the foundations of mathematics.[7][a] The original system was shown to be logically inconsistent in 1935 when Stephen Kleene and J. B. Rosser developed the Kleene–Rosser paradox.[8][9]""" https://en.wikipedia.org/wiki/Lambda_calculus#History
Comment · 4 years ago
I wish there was a 'Modern Python' tutorial that would walk me through all the new python stuff like this or the type declarations or the new libs, etc.
Comment · 4 years ago
The "magnifying" part of this is probably just a red herring - in the article itself there is nothing about the magnifying effect.
Comment · 5 years ago
My point was that to informative about a product you need to know the person you are advertising to. First of all because you need to choose the product that he might be interested about.
Comment · 5 years ago
Unfortunately if they don't know anything about you - then they will not be able to deliver anything informative.
This list is also part of my own contender in this race: https://zby.github.io/commonplace/ - my own LLM operated knowledge base (this is the html rendering of that KB - there is also the github repo linked there).
The main feature is that I use it to build a theory about such systems - and the neat trick is that llms can read this theory and implement it so the very theory works as an LLM runtime too.
It works for me - but it has some rough edges still - so I guess it is not for everyone.