ICM getting you down = @Don Roy Data = @Mira Bradshaw Video/Images/things I can't even pronounce = @Ari Evergreen Needing the human and knowledge = @Carla Bosteder and @Aaron Kruger Admin = too scared to tag them 😉😉 but you will know when you know. other things = classroom, videos and links there is even a Don's bookshelf and ari's space is out of this world (see what I did there). #checkitout Participate in comps (even if you suck... you learn - this is why you are here) Good housekeeping = search (many members have already been where you are) even better check out the legends it is the 11th commandment (Don will post about it -#nopressure Don) Don't be afraid to post but also don't get stuck on the likes and the climbing. Focus on the learning and you will find your place. Above all else... Embrace the 'dumb' it is the only way you truly make yourself teachable. Good luck, God speed, go into the light, break a leg, or whatever people say...

Today my kid had cross country at school. The kids are only 6 and had to go about 750m. They all worked hard. I already had a leave day booked for today when the date was announced, so I was actually able to go and cheer my kid on. This morning, we talked about the race plan they had practiced at school last week. To jog, not go all out and run out of puff, and save your sprinting for the home stretch. My kid is fit and healthy, but not naturally athletic by any stretch of the imagination. When the kids came past the supporters area and cheering parents, about 2/3 of the way through the course, my kid was far and away at the rear. But my kid stuck to the plan. When the other kids ran out of puff, my kid jogged past them and showed the pace in the home stretch. Was it a great finish? No. Were there a good five or six kids overtaken? Absolutely. And - most important of all - my kid raced the whole way wearing a massive smile and with joy. So, apparently I’m the person who remembers the fundamentals in unusual places whether that be the instant noodle aisle or the sports field. Whatever you’re trying to build, my advice is to stick to the game plan. Don’t get distracted by the noise or the shiny tool or the model drop - lift good ideas and what works, but also have the strength and wisdom to do *you* because that always works better. And above all, run your own race rather than worrying about anyone else’s. If you can do it with a smile on your face, that’s even better. That’s the lesson I’m bringing home and following today.

A week ago I argued folders and an index beat a vector database for my vault. I still think that, for a corpus I wrote myself and can keep curated. The projects coming up are a different animal. Technical manuals, in PDF form, and the people who need answers out of them are never going to read them. You also don't want an AI re-reading the whole manual every time someone asks it a question. That's slow, expensive, and it gets worse with every page. The text side has a known answer, and it's the one I passed on for the vault. A vector database: chop the manual into chunks, turn each chunk into coordinates so chunks that mean similar things sit near each other, and the chatbot pulls only the handful of chunks closest to the question. And a manual has one property my vault doesn't. It barely changes. Index it once and the index stays good for years. Most of what makes vector databases painful shows up when the content churns. Here it doesn't. The part I can't find a settled answer for is the schematics. These manuals are full of diagrams. Wiring runs, exploded parts views. For a lot of questions the answer IS the picture, and a text-only retrieval pipeline just loses it. Nobody wants a chatbot describing a schematic in prose when the actual drawing is sitting right there in the source. Where my head goes is a pattern from my day job. We keep a binary database next to the text, meaning a store built for files rather than rows. The row holds the text plus a pointer, and the pointer fetches the image. So retrieval stays text-shaped, but every chunk knows which figure it belongs to, and when a chunk comes back the schematic rides along with it. The model never has to understand the image. It just has to not lose it. Other routes I've seen floated but haven't tested: caption every figure with a vision model and index the captions so images become findable by meaning. Multimodal embeddings that drop images and text into the same space. Or skip the pipeline and let a vision model read the page images directly.

Is there a scenario where a folder based project could outperform a database?