Seriously, I'm very proud of myself for the little I've accomplished so far. I don't have friends in tech so I don't get to talk about it or bounce ideas off people.
Thanks for letting me get that out!
51–60 of 1001 posts
Seriously, I'm very proud of myself for the little I've accomplished so far. I don't have friends in tech so I don't get to talk about it or bounce ideas off people.
Thanks for letting me get that out!
Along the way I found most of these use salvaged BlackBerry keyboards which are only going to become harder to find, so also on a bit of a side quest to build a thumb-sized keyboard from scratch. Got me into laying out and prototyping my first PCBs and learning about how these things are made - lots of fun so far!
Something cool I learned from tearing apart a BB keyboard: the satisfying “click” is just a tiny metal dome that pops and completes the circuit when pressed. Not news to anyone familiar with electronics manufacturing, but it was a cool thing to “discover.”
I’m still exploring new forms of AI-powered learning tools.
The latest thing I’ve been working on is an adaptive mode inspired by the LECTOR paper [1]. Where each lesson is a single learning concept with a mastery score tight to it based on your understanding of the said concept, so in principle the system can reintroduce concepts you didn’t fully grasp later on, ideally making separate flashcards unnecessary.
It can be self-hosted if any one want's to give it a try!
We have an ML model that's trained on real reservations and use an LLM to decide why a user mightve opted out. We apply personas to this LLM to get a bit of a sense how they would probably be operating the booking flow.
Recently it hit v2.0 spec conformance. 3.0 is next on the roadmap. (I'm executing it against the upstream spec test suite.)
I don't plan to make it a highly-performant decoder for use in production environments, but rather one that can be used for educational purposes, easy to read and/or debugging issues with modules. That's why I decided not to offer a streaming API, and why I'll be focusing on things like good errors, good code docs etc.
P.S. I'm new to the language so any feedback is more than welcome.
A Python ORM, inspired by Drizzle and the like. Whenever I come to Python I'm frustrated by the ORM options. They generally lack type-safety on inputs and outputs, or useful type hints.
SQLAlchemy is an institution but I think it's hard to use if it's not your full-time job. I check the docs for every query. I want something simple for the 80-99% of cases, that lets you drop easily into raw SQL for the remaining %.
I'm going to keep hacking at it, would love to from anyone who thinks this is worthwhile (or not). Also: - The interface for update queries is clunky. Should I add codegen? - Should I try to implement a SQL diffing engine (for migrations). Or just vendor sqldef/similar...?
The vast majority of tokens in a sequence will be irrelevant to an attention mechanism outside of a very small window. Right now however we tend to either keep all cache values forever, or dump them all once they hit a certain age.
My theory is that you can train model to look at the key vectors and from that information alone work out how long to keep a the token in the cache for. Results so far look promising and it’s easy to add after the fact without retraining the core model itself.
Since hacker news last saw it, it’s been translated into English, German, Spanish and Chinese. If, say, a Chinese speaker wanted to learn more English words, then they could go to https://threeemojis.com/zh-CN/play/hex/en-US/today and play the game with English words with Chinese definitions and interface. This is the first cross language daily word game of its kind (as far as I know), so it’s been a lot of fun watching who plays which languages from where.
The next challenge that I’m thinking about is growing the game. The write ups and mentions on blogs add up, the social sharing helps, but I’d really like to break into the short form video realm.
If you read interviews from other word game creators, every successful game has some variation of got popular riding the wordle wave, or one random guy made a random TikTok one time that went super viral, and otherwise every other growth method they have tried since then hasn’t worked that well and they are coasting along.
So, sans another wordle wave, I am working on growing a TikTok following and then working on converting that following into players, a bit of a two step there, but that’s how the game is played these days. https://www.tiktok.com/@three_emojis_hq for the curious. Still experimenting and finding video styles and formats that travel well there. Pingo AI and other language apps have shown how strong TikTok can be for growth, so I think there’s something there. That’s all for this month!