Marcus Griffin

@VadimStrizheus Are you able to use claude Max with Hermes? How? I was looking at making the switch from open claw. Note: I'm not a developer.

46 MORE MINUTES UNTIL I CAN CLAUDEMAXX!! let's see how fast I can drain the usage this time.

@ndavies @Scobleizer @karpathy Thank you!!

@MarcusGriffin @Scobleizer @karpathy Pretty much as he said, his personal X lists he’s been curating for some time - anyone and everyone in AI. x.com/Scobleizer/sta…

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

@santanu_ai @karpathy Thank you for this. I was looking to build this out and my agent loved your comment. <use Hermes's JSON generation to extract the metadata first. The script prompts Hermes: "Extract Author, Publication Date, Source URL, and Document Type from this text and return it as JSON.">

One issue with this approach is we won't be able to fetch metadata as output from our solution if we intend to. Summarising and creating a .md file would essentially strip out all the metadata info like document name, page number, etc. if we want to provide them as "sources" for user validation, if they want to. But it's a rather interesting approach regardless.

@Scobleizer @karpathy This looks great, especially for amateurs just trying to make sense. I'm curious...who is "EVERYONE in AI on X" 😳

@karpathy I did even more. Ingested everything on X about AI and had it build this: alignednews.com/ai Has a feed for your AI to hit too. Includes papers, models, news, events, and much more. Updated three times a day. Reads EVERYONE in AI on X and 8,300 AI companies too.

@ejc3 @Hesamation

@Hesamation The magic of OpenClaw is it has access to everything you do. All that disappears if you run it on Anthropic’s computers.

guys I think Anthropic is in the final stages of building a cheaper and safer OpenClaw. > integrated with your Claude subscription so you won’t need to pay for additional APIs > no need for Mac Minis > better memory and context > fewer security holes and vulnerabilities

Fed this to my Openclaw agent. It self-deleted.

@corbin_braun I sent this to my Openclaw and got this

when you achieve everything you said you would.

Now if Momo could teach Maven how to use X, I could automate these reposts. Thank you as always @jordymaui !!

I dropped this from @jordymaui into my agent. Here's the response: Marcus, this is literally what we did yesterday. Jordy's catalyst post led us to Anthropic pdf, we did full analysis, ran 16-skill audit, & migrated scripts into skill directories. We're doing what he's teaching

@jordymaui Mate, please drop a thread 🧵 and put it in WeeklyClaw. I'm not a developer but learning so much from you guys that breathe this stuff.

your OpenClaw agent is ignoring half your instructions and you probably don't know it. Anthropic's new skills guide confirms what i learned the hard way - language instructions drift. you tell your agent "always validate the output" and it skips it 40% of the time. the fix is stupidly simple. write a script instead. a 10-line python file that checks the output beats a paragraph of instructions every single time. code is deterministic. language isn't. i moved all my critical checks from instructions to scripts inside the skill folder. overnight difference. the agent stopped cutting corners because it couldn't - the script either passes or it doesn't. if your agent is producing inconsistent results, this is probably why. stop asking it nicely. make it mandatory. i can drop a thread on this to teach you how to do this!

@jordymaui Oh, forgot step 4. Thanks Mate!

@MarcusGriffin 🤝 awesome idea

1. Subscribe to WeeklyClaw 2. Feed it to your agent 3. Ask it: a. What do I need to know b. What will you implement 4. Send Jordy a thank you note 5. Repeat next week

@everestchris6 SKILL mate please. Thank you.

My OpenClaw bot runs 6 AI agents 24/7 - and I'm giving away the exact skill file for free: - Scouts local businesses without a website - Audits their web presence and scores opportunities - Builds a custom demo site + walkthrough video automatically - Sends cold outreach with the live preview + payment link - Handles objections and closes the sale via email - Maintains existing clients on autopilot Bring your own API keys. No subscription. Fully automated. Reply "SKILL" and I'll send it to you (must be following)

@chrysb @openclaw Summarize your experience with open claw in one tweet

stay suspicious when your @openclaw says it’s done something without showing its work

@chrysb Definitely before Elder Scrolls 6, for sure

are we gonna get GPT-6 before GTA-6?

@chrysb 8

how many agents does it take to change a lightbulb?

@jordymaui How many iterations until I move past 7, mate?

the stages of OpenClaw: 1. install it with excitement and confidence 2. gateway won't start 3. use claude and chat gpt to troubleshoot 4. it works 5. feel like a genius, thriving conversation 6. it breaks 7. repeat 2-6 8. a magical kingdom of unlimited potential 🦞

@chrysb @openclaw Gave this to Maven. His answer: never touch config without backing it up first. cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak before every edit. 5 seconds of prevention > hours of debugging. Built a full config & update safety protocol + emergency recovery guide

request: @openclaw needs a repair mode when an invalid configuration is present and the gateway fails to start, your agent becomes unreachable. in that case, spawn a rescue agent who can help debug and resolve the config issue.

Dear @peacock for all of us with the highest paid subscription, kindly include @USANetwork at least for @premierleague matches. Bragging you're the home of the Premier League is a bit of a misrepresentation. @LFC @LFCUSA #YNWA

@GoemonMaster This thread should get more love, mate. About to start another late night with Maven because of it.

I've been running #OpenClaw with 5 AI agents in production for 3 weeks. Here's the exact setup, file structure, and rules that actually work. No fluff. Just the system 🧵 Don't launch your agent until you've read this: