Notes

Quick note here for posterity. I wanted to setup a chromadb with maybe something similar to the @cf/qwen/qwen3-embedding-0.6b that I’m using on cloudflare out of the box. I setup a new uv venv --python 3.13 and tried to pull in uv pip install sentence-transformers chromadb markdown-it-py and then I wanted to , try qwen31 from hugging face, from sentence_transformers import SentenceTransformer # Load the model model = SentenceTransformer("Qwen/Qwen3-Embedding-0.6B") however I ran into several installation roadblocks. Anyway without getting into all the hurdles, ultimately I realized per reading the torch docs3 and the versions of torch available4, that my x86 mac torch support stopped at torch<2.3. And after it was only arm64, if I wanted to just grab the wheel file directly. And the torch website recommended python 3.10 . And well, that torch 2.2.2 requires numpy<2. And the last twist was that huggingface started listing the qwen3 architecture only by transformers>=4.51.0, which can also be seen on their model card2 . ...

Note on creating this image Original prompt to Chat GPT Realistic Image of someone in shorts pulling a pocket out, inverting it, revealing a small hole the pocket lining has. They are of the facial expression sort of like “hide the pain harry”. They are holding a slightly screen cracked telephone in their other hand. This image has a second vertical pane as well, on the right. This pane is further split into top and bottom, the same person shows they are putting now their phone into a thin hat, and their mood is like neutralish. Now the last bottom pane here, we see them half way, putting their thin hat, that visibly has the geometry of a phone rectangle , into their pocket. Now their facial expression is “clever” . And there is also a thin frame at the top where a title reads “Hat Pocket” The panes, arranged, are like this , ...

so I got a kind of an interesting thought about what is a good benefit from AI by listening to this interview about how AI is making our education system dumber because people are cheating. So the people in the interview are talking about how students are cheating when they are writing, but at minute 15 one of the guests talks about the defense as a style of proving that you know some material as an interesting way of my passing the cheating, which is in writing, and I think he’s out to something. ...

Image notes Original Prompt Lets create a realistic image of a bunch of dogs in a parking lot, where there is maybe one or two cars in bad shape and the dogs are just visibly barking , to each other at random birds , at thr cars, or off frame . Everything is super real looking , but lets add in some small “bark” “woof” words near the dogs that are vocalizing ...

Why are the AI capabilities discussions so polarizing? People either see current AI as a net negative or net positive. But the gap is a chasm. Well ok maybe there are people who are uncertain too. Okay maybe it is task specific. But the edges are extremes. Why? I suspect the split aligns, along whether believe in planning. On the one hand, yes there are many AI aided workflows that produce very stable reliable outcomes. And that side is growing. But some people see the billion dollar single person start up around the corner. ...

Thinking this brilliant Balaji Srinivasan ism is that merging is hard . I think he meqnt it broadly, including with a merge in traffic patterns and companies and ideas. I wonder if he had code in mind. Linus Torvalds decided to build a version control system that made it easier to fork and code but if not used cautiously, you will spend an afternoon resolving merge conflicts. Maybe blindly merging and letting tests save you is like trusting an LLM tool to write code for you? But kind of that’s the thing, trusting an LLM’s diff is accepting a merge in a way and merging is hard. ...

While I was reviewing my git lfs based logseq markdown backup, I noticed a git lfs inconsistency. So git lfs fsck reported 339 .PNG files as pointer: unexpectedGitObject: "assets/IMG_4942.PNG" (treeish 53f01b6616262c6e9bcb6eabb3cc11cb298fab1c) should have been a pointer but was not, even though when I looked at my .gitattributes I did see *.PNG filter=lfs diff=lfs merge=lfs -text. The numbers suggested my .PNG however was not on my git-lfs. $ git lfs ls-files |wc -l 1715 and ...

Settling on Procrastination Having options and constraints is good in moderation. Sort of a twist on explore and exploit here. I fall in line with the thinking behind Tim Urban’s Monkey Mind TED talk [2]. Hold out until we know [1] just the right amount of information to proceed, not a second less or longer. Feels chaotic though! I remember I had a coworker who was all about “scope creep man, watch out for that!” His preference was for seeking comfort-in-constraints. Like business, as usual. ...

Drowning in information. Is this a low hanging fruit for LLMs that is close to being solved? When I first encountered How to Take Smart Notes [3], offering up Luhmann’s lost art of the slip box, I thought this was the way. Luhmann created notes on primary sources he read like books or papers, but added links to them from topics as the connections seemed to make sense. He did it all on index cards filed in drawers that were hyperlynked manually but this made him a prolific writer still. ...

Riffing off of , https://michal.piekarczyk.xyz/note/2026-03-04-prompt-driven-development/ , and the famous Do Learn Show Loop , and also the ODSC conference I just attended and also squarely on a few topics Nate B Jones has been covrring lately on career visibility , much extending what I think Steve Huynh has introduced me to with his brag docs a few years ago, because work iterates and becauase everyone loves a story, especially interviewers, we should extend the STAR and brag doc styles to be more punctuated, more colorful, still as accurate as possible. ...