Ali Jrl ☁️

@JonathanPeled Fuori dall’Italia, ORA!

Condanniamo fermamente l’uso manipolatorio della recente copertina de L’Espresso. L'immagine distorce la complessa realtà con cui Israele deve convivere, promuovendo stereotipi e odio. Un giornalismo responsabile deve essere equilibrato e corretto. #MediaResponsibility

@velinus_sage @NotebookLM Had a very similar idea, wanted to call it XRadio, a radio program based on my X feed.

Been playing with Hermes like everyone else this week. Working on a skill that turns my X feed and bookmarks into @NotebookLM podcasts. My goal is to passively parse my feed to extract signal and get a podcast that keeps me current with the warp speed AI agent tooling progress. Hermes makes iterating easy. Big fan of what @NousResearch is doing. Repo in replies.

@NousResearch @BlakeC @openclaw Much obliged.

@AliJrl @BlakeC @openclaw www dot shop dot nousresearch dot com

It’s not @NousResearch vs @openclaw. It’s taste vs tasteless.

@NousResearch @BlakeC @openclaw I want that t-shirt @NousResearch, how can I get one?

@BlakeC @openclaw No comment

@diegozaks Are you hiring in Europe?

If this sounds like you, I'm hiring.

@CardilloSamuel Are you open for consultations? I’d love to grab 15-20 minutes. Happy to pay.

the ceo of one of the companies i am helping to transfer into local inference via my dgx spark & hermes agent (see tweet x.com/CardilloSamuel…), an extremely big european corporate (and known), asked the assistant to create a picture of itself its running on my qwen3.5 37b a3b opus distill model and its not capable of image generation so it decided to answer wtih this 😂😂😂

@omar_or_ahmed @karpathy This is such a cool use case. How do you comply with data regulations such as CPA and GDPR?

@karpathy healthcare version: clinicians building patient knowledge bases. LLM compiles 200 EHR notes into patient wiki (dx, meds, labs, timeline). then query: "why is glucose up?" and it traces causality. same obsidian workflow but clinical data

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.

LLMs are commodities. It will be hard to create long term pricing strategies. Here’s what I think will make it harder: - rise in smaller specialised models that you’ll use for specific tasks. - spread of distilled model, created off of frontier closed models that you can run locally.

This isn’t new. It wasn’t allowed anyway. We are 3 months out from local models doing everything that people use openclaw for anyway I have been using openrouter instead for Hermes But this brings up an interesting question… What is the monetization model for making money off of power users when local models get so good that you don’t need pay Anthropic or OpenAI anymore? Of course the average joe will always just pay $20 a month, but there may be a seismic shift coming

@Dougordon1999 @slayguevara Ahahah

@PriaINC @slayguevara Pete Hegseth's Ancestors are Scandinavian, YOU RETARD.

Your ancestors literally couldn't dream of this level of civilisation you troglodyte

@JamesonCamp Great stuff 💪🏼 share the results!

@AliJrl yes i saw that too! getting new macbook pro on saturday, considering keeping old one to run local if it can handle it

$1500/mo for a remote SDR to do what Hermes is doing for me at $250/mo Set it up last night for Nano Flips newsletter ad sales It’s already > scraping competitors newsletters > finding advertisers >looking up right person on their team >reaching out >negotiating rates > and tracking everything in a CRM And I just chat with it through Telegram. Much less clunky than openclaw was. And it gets better on its own - learning which outreach gets opens, who actually buys, how to negotiate better Def not something I would pitch to an enterprise client or mid market firm yet. But for an SMB or freelancer…this is an amazing agent and kind of crazy

@JamesonCamp You should look into local models to reduce costs. I heard Gemma (a new model from Google) performs really well locally on regular machines. Also qwen. I’ll look further into that.

@AliJrl it will RIP through tokens, so right now moslty sonnet, but will have it use opus for writing, and i think a lot i can get it to use haiku and will be fine instead

@JamesonCamp @johnstreetlabs Ahahah been busy all week. Looking forward to the weekend to roll up the sleeves.

i had some fantastic devs tell me openclaw was cool but not worth the time, and super unsure of the average person getting real value from it and then i just set up hermes in a few hours without calling our dev team. was texting them all night about how hyped i was i set it up haha

@JamesonCamp Interesting - I’ll dive in this weekend. What model are you using? Are you using different models for different tasks?

@AliJrl currently I am yes, and thats whats cool is with my edits it learns to get better for next time in the future i wont need to

@johnstreetlabs @JamesonCamp From what I’ve heard Hermes is very well suited for local models. OpenClaw for cloud models.

@JamesonCamp Why Hermes over openclaw?

@luke_metro There’s far better out there

So does Anthropic buy Sourcery now or what

@SeeOn__ I was curious to ask you how you were rendering the floor plan on the left side.

Have been trying to figure out a way to load the entire plan from an image OpenCV+ OCR <- nope Gemini-3-flash-preview <- nope will be trying claude and gpt ... Let see how it goes

@sonofalli It was at this moment that everyone pivoted to EVO-as-a-Service.

don’t worry guys some people already pay $200 for oil

@forgebitz 0

last year we had some massive billboards next to the highway/airport if you guess how many signups we got from it you get a free hat

@rabois I’m surprised and sorry that you get such poor-quality intel, Keith.

Goals achieved: 1. Iranian Navy eliminated. 2. Air Force eliminated. 3. 48 of the 50 top evil leqders dead. 4. Less than 80 launchers remaining w a shelf life of one fire 5/ No capability to enrich an nuclear material probably for a decade (inputs destroyed like water plant, steel and most sights). 6. Very military ballistic missles supply left w no ability to produce more. 7. Some drones remain w negligible ability manufacture new. 8. Unified Gulf States against Iran except Oman 9. Chinese weapons proven to be dysfunctional. 10. Middle Eastern states lining up for business relations w Israel.