ScalarLM.com
Ask HN: What are you working on? (March 2025)
111–120 of 1001 posts
Re: Ask HN: What are you working on? (March 2025)
#112I use Apple Pencil with my iPad for figures in documents and wanted a better way to sync with my other work... so writing an iPad app for drawings using git for syncing with remote repos. Goal to release in App Store by end of April.
Re: Ask HN: What are you working on? (March 2025)
#113Not much :)
Which I think is a valid answer. I have a job, family and some health issues.
The main thing I am looking at is blogging. Just posts on a problem I solved that week at work kind of thing. Seems like a low time cost way to promote myself. I might dip my toe in the cooking fat of linkedin engagement using the posts.
As much as that is yuk I feel it may be beneficial. Just need one linked in lurker to be impressed and hire me in a few years time!
And by not posting about AI or working 200 hour weeks I might stand out :)
Re: Ask HN: What are you working on? (March 2025)
#114Still plugging away on my linear genetic programming experiments.
The big debate in my head right now is whether a next byte prediction architecture is better or worse than full sequence prediction.
The benefit of next byte prediction is that we only expect 1 byte of information to be produced per execution of the UTM program. The implication here is that the program probably doesn't need a whole lot of interpreter cycles to figure out this single byte each time (given reasonable context size). However, the downside is that you only get 256 levels of signal to work with at tournament selection time. There isn't much gradient when comparing candidates on a specific task.
The full sequence prediction architecture is expected to produce the entire output (i.e., context window size) for each UTM program invocation. This implies that we may need a much larger # of interpreter cycles to play with each time. However, we get a far richer gradient to work with at fitness compare time (100-1000 bytes).
Other options could involve bringing BPE into my life, but I really want to try to think different for now. If I take the bitter lesson as strongly as possible, tokenization and related techniques (next token prediction) could be framed as clever tricks that a different computational model could avoid.
Re: Ask HN: What are you working on? (March 2025)
#115FOSS MTG inspired digital card game.
I love card games, but for digital card games the business model is beyond predatory. If you need a specific card your option is to basically buy a pack. Let’s say this is about 3$ give or take. But if it’s a specific rare card, you can open a dozen of so packs and still not get the specific card you want.
This can go on indefinitely, and apologists will claim you can just work around this, by building a different deck. But the business model clearly wants you to drop 50 to 100$ just to get a single card.
All for this to repeat every 3 months when they introduce new mechanics to nerf the old cards or just rotate out the dream deck you spent 100$+ to build.
I’m under no impression I’ll directly compete, but it’s a fun FOSS game you can spin up with friends. Or even since it’s all MIT, you can fork and sell.
It also gives me an excuse to use Python, looks like Django on the backend and Godot for the game client. Although the actual logic runs in Django so you can always roll a different game client.
Eventually I’d like different devs to roll their own game clients in whatever framework they want.
Want to play from the CLI, sure
Re: Ask HN: What are you working on? (March 2025)
#116I'm working on an emulator for 16-bit computer I have designed for teaching students. It's designed to make low level computing more accessible for modern students by making things as visual as possible, for example blinken-lighten for the registers like w/ the old PDPs, color coded memory that shows where the code and data segments are, where the stack is, etc, and a small frame buffer that drives a 64x64 2 bit disp…
Nice. I made an 8-bit AVR thing along those lines, 240x180 - 16 color. In browser emulator and assembler.
Can load source from gists https://k8.fingswotidun.com/static/ide/?gist=ad96329670965dc...
Never really did much with it, but it was interesting and fun.
Re: Ask HN: What are you working on? (March 2025)
#117Working on a schema migration tool for tortoise-orm (Python async ORM)
Re: Ask HN: What are you working on? (March 2025)
#118A single address space operating system called Fusion¹ implemented in Nim. It currently has the basic building blocks for the kernel. The rest is being built incrementally as time permits. I've also taken the time to document everything I'm doing², which some people found useful.
Re: Ask HN: What are you working on? (March 2025)
#119RSS reader through Bluesky custom feeds: https://github.com/tfederman/stroma-news
Bluesky API library spun off from the other project: https://github.com/tfederman/pysky
Haven't really started it yet, but a master list of RSS feeds and the code I used to source them: https://github.com/tfederman/huge-rss-list
And also a new project to fetch all links seen in the Bluesky firehose and gather metadata to build a database of sites and pages at a more granular level than the domain. For example, is account X posting video links from one YT channel or many?
Re: Ask HN: What are you working on? (March 2025)
#120https://aponlink.com/ I built a no-AI, human-only social network focused on ONLY one thing - keeping people connected. I'd stepped away from mainstream social media last year due to the overwhelming negativity, privacy violation, etc. Then around early this year, I started to feel I was missing updates from people who actually matter in my life. Instead of going back to traditional platforms, I decided to create a si…
Something like this has been on my mind for a while now -- take the useful, positive elements from across the socials (network of connections, media sharing, events, etc) and create mini-nets that let people who want to, stay in touch.
How do you envision onboarding? Do I join, and then try to convince a handful of people to join as well?