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.
What's the concern? Harness tends to be fairly common in the context of "shim program which manages some other program" (see: "test harness", "fuzzing harness", etc.)
It's kinda ironic that everything has become an "app" over the past 10 years. Facebook is an "app", Reddit is an "app", your bank is an "app". However, the one time we actually introduce an app to execute our LLM calls, we don't call it an "app"? Wat.
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...
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?
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.
Is it not still that way? There may be a bias towards believing trusted institutions, but that's because they have reputations for delivering truth based on experiments.
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.
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)..
I would slightly disagree. USD as measured by the DXY is as much if not more of a function of international demand for dollars (ie Eurodollar markets). There is massive structural demand built into dollars based on coupon payments in $ denominated international sovereign and lessso corporate debt as well as international commodities markets.
> This can stop in a rapid phase shift once people realize that USD is not such a safe asset any more.
Why would they think it is not a safe asset and if so what would they do instead with it? Also: Why do you think this shift would be rapid like in a phase shift?
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.
Property? Gold? Industrial commodities? Mining rights? Used cars? Bitcoin? Farmland? Other hard assets? Arguably that’s already been happening, with the global melt-up in asset values across the board.
Now - why rapid - in general because it will be a bank run. People now assume that the dollars they have in the banks (and in notes) can buy them a part of the world economy - when they realize that this part shrunk they'll try to sell dollars and buy something else - adding to the dollar decline.
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
By virtue of all programming being resource-bounded computations can be interpreted in Linear logic [1].
Linear logic is paraconsistent [2] not consistent.
Untyped Lambda calculus is essentially an assembly language and Assembly languages treat their own code as data [3]. Contradictions arrise when a system can mutate its own state. That is the power and pitfall of reflection [4]
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.
If you're not averse to books I'm yet again recommending "Robust Python" by Patrick Viafore and for more esoteric things like metaclasses I would recommend "Expert Python Programming, 3rd Edition" by Michal Jaworski & Tarek Ziade. It's not wholly dedicated to metaclasses, so there are tons more to learn from it as well.
I learned about metaclasses in "Fluent Python". An excellent book, but it predates typing, so that's not covered. Then again, I believe a second edition is on its way.
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.