@dlusion@GregFeingold@felixrieseberg I see I’m not the only one using gpt for random chats in the background. I may have 3 claude code terminals open, a chat going in desktop app in a project and still have to whip out gpt some. Could be some subconscious not wanting to use up my claude limit tho
@arscontexta nice. this is similar to what have been doing with claude code in my obsidian vault. i already had similar structure pre-claude code from Nick Milo’s lessons on vault structure and how to take notes. going to check it out and very interested to see openclaw version
Read both of your articles last night. What luck I decided to read them. This is similar to how I was finally figuring out how to solve my own human long-term memory and autonomy using an Obsidian vault, so I was immediately intrigued. As a non-technical person using AI to learn that side, I was just missing a few pieces (like qmd). It makes sense to me that it would be the same for AI Agents.
I use a hybridized version of the ACE PARA hybrid system, which was heavily influenced by the ideas and concepts that @NickMilo lays out in his Linking Your Thinking. Claude and Claude Code helped customize it for me and help quickly find and create.
I'm only on Day 2 of OpenClaw usage, but I can already see how powerful this can be based on how powerful it was for me as a human.
I already use a variation of your primitives, and it appears that the main difference is you are just using "People" as one primitive, whereas I am using "Cards", of which "People" is a subcategory. I also don't use all of your other primitives quite as explicitly as you, but the underlying concepts are all present in my current vault, along with a structured, yet semi-fungible yaml frontmatter for every note.
However, I can already see how your system would prevail as a long-term memory architecture for autonomous AI agents specifically. Mine still leans as built for a human to be assisted by AI in general.
At this point, I'm just going to wait for your release of v3. I hope this goes viral for you, but I'm already pretty sure its going to work just based on reading your articles. Excited to try it out with my OpenClaw agent!
Would be curious if you have been/ are thinking of ways to integrate peoples existing vaults. My OpenClaw already has access to my vault, but I'll probably just need to play around and see how yours works.
Btw here is my current vault structure:
+ Add
Atlas/
Cards/
Companies/
Concepts/
Equipment/
Glossary/
Organizations/
People/
Places/
Recipes/
MOCs/
Sources/
Calendar/
Efforts/
On/
Ongoing/
Simmering/
Sleeping/
x_Done/
X/
CLAUDE
CLAUDE.local
Home
README
TAGS
Have you tried having Opus 4.6 build an analysis tool straight in chat yet 🤯? I’ve been using the excel add-in and opus 4.5 since release. I also use chat to create prompts for excel, but Opus 4.6 took my intended excel prompt request and said hold my beer… made an entire 7 sheet analysis tool on max pref equity based on perm loan sizing all with cap rate and int rate distributions. It was 100% correct and had already implemented the next 5 things I would have requested.
Claude code with Opus 4.5 and using https://t.co/X2BgsVG8IT files is just on another level. I typically create a project in Desktop app, use Wispr Flow to dictate what I am trying to do and have Claude create the instructions for the project, then load any relevant documents into the knowledge center.
I then “lazy prompt” by dictation into a chat in the project exactly what I want and tell it to create a prompt for Claude Code to execute it. Have it refine the prompt a few times, maybe have it use some tree-of-thoughts reasoning to refine further, then just paste that prompt into Claude Code on Plan mode. I’ll then feed the Claude Code responses back into the project chat (always Opus 4.5 + extended thinking) and repeat the cycle.
I had no idea the amount of components that went into a website but Claude did and there was a detailed spec of it all for the first prompt into Claude Code. I think people just fail to iterate and iteration is where the gold is at.
I get the point of his article and I do have a minor prompt library but I mean if you’ve put any effort or thought into the last couple of years, Opus 4.5 makes it pretty easy to get great outputs for relatively simple tasks like building great websites.
The more difficult of a problem I am trying to solve, the more refinement and iteration goes into each prompt on each step of the process. For example, coming up with a hyper specific, complex investment thesis and turning it into a falsifiable thesis and then getting Claude code to write it out in Python to be able to research, test and implement it. There is no set of instructions for that on the internet the same way there is for building a world class website.
Here’s a TL/DR: incessant iteration on a specific issue is all it takes. The amount of iteration required depends on the model you’re using and the complexity of the issue. But Opus 4.5 is a beast that can churn through most anything.
@Globalflows Opus 4.5 via Claude Code in the hands of a clear thinker that knows what problems they need to solve and how to articulate them is unparalleled.
@jaymesrosenthal@Globalflows It’s been very useful to me when studying all of Cap’s edu materials and book recs for tracking and updating thought processes and seeing patterns etc. You determine the connections yourself but it builds up over time. Example below.
internal geography is extremely important to the exceptionalism of any country and the US happens to have the best by a wide margin; that's not something a country can change.
1) The US has more navigable internal waterways than the rest of the world combined
2) the ICW is a maritime superhighway and acts as functionally free infrastructure
3) the US has more deep water ports than any other country
4) the US has more high-quality, contiguous arable land than the rest of the world’s major powers combined.
If you don't read books, the system may dissolve your brain.
I didn't even get to our innovation, energy independence, and military. God bless America! 🇺🇸
@cantonmeow ISM doesn’t show a cycle top being in either. something seems to be brewing. I don’t know what though. maybe we get something similar to last Nov-April? extend the cycle even longer. My IWM calls got crushed last 2 weeks so I don’t know much clearly
@cantonmeow@TeslaNash seems like something changed when the US outright weaponized the dollar in 2022. central banks may still play mr. slappy to some degree but I don’t see them puking all that gold back up any time soon. XBT and XAU are both nationless/stateless
@cantonmeow may want to have a word with your kitty cat relatives in the East and their real estate problems and obsession with gold. lmk what they say
Likely connected to the fact that China real estate keeps having major issues>China has to have a trade war with the US>export more in order to prop up their economy> they need dollars to manage their exchange rate> they'd rather sell BTC for dollars and any dollars they want to keep they put in gold
@realEstateTrent Positive expected value vs negative expected value. They both involve placing bets but the difference depends on whether you're betting with the odds (investing) or against them (gambling). Everything is a distribution of probabilities.