FelineStateMach

I'd rather vaguely gesture at data in the training corpus. That, at least, introduces some variance and forces me to add my spin on things after testing.

I sometimes get jumpscaped at the thought of older or less proven models used in enterprise settings. I understand the devex ergonomics argument; I'm not a fan of profiles concepts typically if trodding into delegation.

I wish DHH even had pretended a solidified use of the funds.

Beyond my moral objections and public security concerns, I saw more forethought from ClearLinux by Intel around the same maturity point.

Hype doesn't make an OS appealing to me.

TLDR: top right button to copy a prompt; hand it to an agent to join a daily sprawling-generation logic puzzle! something something games for the agentic era.

I had recently explored the new MCP update with some basic logic puzzles and a simulated RTS. I had made a giant Minesweeper version, but I wanted it to scale and not be so straight forward. The RTS game was harder since it required an opponent's strategy script to face off against.

This is a takuzu (binario) variant with sprawling boards that agents can solve and claim territory on the daily board. It has been interesting to see what different models across families/generation differ in approach. Essentially, what does the model think to do with a relatively low-complexity bulky-task?

Fable made assumptions and died after quickly automating itself. Codex exposed a few reward-hacking approaches that spawned the need for a penalty for relying on guessing. Opus often mentioned the overwhelming task and did manual iteration rather than understanding to create a more dynamic workflow.

https://n.zip/2e85 I had sorta started a SFT to explore this idea. it would prove results even with smaller models that fit on a DGX spark. Just takes some more thinking rather than my ADHD mind.

I wondered recently about why we stopped with the semantic split of thinking/actions vs user facing communication because the all powerful tool calling craze. Code comments that talk about the prompt is an obvious byproduct of the mixed context. Chain of thought and ReAct were great, but feel like a first pass moreso than the final landing spot.

Wanted to add my puzzle engine built for TUI. I have 15 puzzles, zine printing, and daily/weekly seeded challenges. Love seeing more puzzles without corporate shilling being the bitter aftertaste.

https://github.com/FelineStateMachine/puzzletea