b0a04gl

13 ms·

31.
Cursor for Anything They Say(devpost.com)
1 points by b0a04gl 1y ago|1 comments
32.
Why Binary Won and Nothing Else Even Got Close(b0a04gl.site)
3 points by b0a04gl 1y ago|1 comments
33.
by b0a04gl 1y ago
been running llama.cpp and vllm on same 4070, trying to batch more prompts for serving. llama.cpp was lagging bad once I hit batch 8 or so, even though GPU usage looked fine. vllm handled it way better. later found vllm uses paged kv cache
34.
by b0a04gl 1y ago
we probably just need to train soil bugs to treat plastic like dead leaves. researchers like rillig(germany) and Ting Xu(berkeley) already did something there in this direction. next move maybe just making sure these bugs don’t mess with ro
35.
by b0a04gl 1y ago
how would you encode a constraint system where the generator must yield exactly one solution and that solution remains unique under all transformations in the problem’s symmetry group, without relying on post-solution filtering or external
36.
by b0a04gl 1y ago
every time your shell takes 100ms to render git status that you didn’t even need, you're paying invisible tax on flow. terminals should be reactive memory tools, not passive decoration. we optimize for code runtime but not for our own
37.
by b0a04gl 1y ago
emoji width bugs mostly come down to how terminals interpret Unicode's "grapheme clusters" vs "codepoints" vs "display cells". emoji isn't one codepoint - it's often multiple joined by zero-width
38.
by b0a04gl 1y ago
how much of modern set theory is reverse engineered from axioms rather than discovered. we're always building highways through a forest we haven't mapped, assuming every tree will fall in line. and suddenly these new large cardina
39.
by b0a04gl 1y ago
we had to go through this at my current place. getting SOC2 type 1 wasn't easy, it forced us to clean up years of infra mess. audit trails that never existed, access logs that were half broken, no changelog discipline. suddenly had to
40.
by b0a04gl 1y ago
if reducto leans in fully as the layer that remembers every correction, every edge case, every shift in layout or wording across document versions it starts becoming more than a pipeline. it becomes institutional memory for unstructured dat
41.
by b0a04gl 1y ago
vim+tmux setups usually lean on system primitives = pipes, files, signals, scrollback so the tooling tends to stay transparent across environments. that gives them an edge in portability and debugging, especially over ssh or in constrained
42.
by b0a04gl 1y ago
terminal scrollback is the only UI we don't treat as queryable state. op's setup shows how much context we're missing away a lot of rg output, file paths, stack traces, build logs they're all already right there in clutt
43.
by b0a04gl 1y ago
something i noticed while debugging tramp perf half the lag isn't tramp, it's elisp codepaths calling sync ops assuming they're cheap. i saw vc-git-root firing on every bufferlist switch because some mode wants to update a ba
44.
by b0a04gl 1y ago
how our workflow starts changing when we realize it can hold multi step intent. l we stop thinking file by file. we start thinking in actions. "split this module, write tests, refactor callers" becomes a single unit in out head be
45.
by b0a04gl 1y ago
huh funny how python's doing the job of modeling type systems when it barely enforces types itself. but makes sense the more u poke around. didn't have to fight any compiler layer or deal with noise. could just build the inference
46.
by b0a04gl 1y ago
but saying they're useless ignores a bunch of real systems that wouldn't run without them. in unity, you literally can't do burst compiled jobs efficiently unless you choose the right allocator. they expose `Temp`, `Persisten
47.
by b0a04gl 1y ago
haha wait do you actually read long commit messages( more than a line) all the way through? like line-by-line, imo commit msg = tweet, git note = blog post.
48.
by b0a04gl 1y ago
i use git notes pretty heavily in my current role. started as an experiment to keep track of internal code reviews without flooding the commit message or making PRs for everything. i tag every commit with context what tickets it maps to, in
49.
by b0a04gl 1y ago
some Makefiles use indents or var placement as semantic cues. if a tool rewrites them mechanically, it might clean things while killing meaning. is structural correctness enough, or do we need formatters that preserve human context too?
50.
by b0a04gl 1y ago
tpu's predictable latency under scale. when you control the compiler, the runtime, the interconnect and the chip, you eliminate so much variance that you can actually schedule jobs efficiently at data center scale. so the obvious quest
51.
by b0a04gl 1y ago
counting bags instead of measuring total plastic weight is peak policy theater. yeah fewer bags on beaches looks good in a chart, but if each one's 50x thicker now, congrats you just upgraded the pollution class without fixing the prob
52.
by b0a04gl 1y ago
> “it's time to retire the just slime’ metaphor” hits hard. we’ve used that phrase to downplay life’s complexity, but statistically, life is the anomaly not the default. the blog nails it: we’ve only found life in one corner of one
53.
by b0a04gl 1y ago
tbh that's not the flex. storing 100PB of logs just means we haven't figured out what's actually worth logging. metrics + structured events can usually tell 90% of the story. the rest? trace level chaos no one reads unless pr
54.
by b0a04gl 1y ago
persona 5 royal doing 7.25mil and sega still acting like it's some niche anime side hustle. meanwhile they keep pushing sonic into every genre except tax software
55.
by b0a04gl 1y ago
appreciate the clarity, that helps. quick followup if the agent's running on a separate machine and interacting remotely, how are failure modes handled across the boundary? like if the agent crashes mid-operation or sends a malformed c
56.
by b0a04gl 1y ago
> "you can define assets in pure python using any framework or engine you want." sounds flexible but what does that actually mean in practice? are there guardrails to keep things interoperable > "engine-agnostic executi
57.
by b0a04gl 1y ago
this is exactly how you earn your prod stripes. dropped the db on day 3? good. now you’re officially a backend engineer. no backups? perfect. now you'll never forget to set one up again. friday night? even better. you got the full rite
58.
by b0a04gl 1y ago
why are we surprised transformers can't detect what's missing when the entire stack assumes the input is complete? the tokenizer doesn't leave placeholders. the attention weights have nothing to anchor to. even the loss funct
59.
by b0a04gl 1y ago
so agents can control tabs, forms, clicks—like a real user would.so what about undo. if an agent clicks the wrong thing, how do you roll that back without reloading the world?
60.
by b0a04gl 1y ago
how are they isolating ai agent state from app-level processes without breaking BEAM's supervision guarantees?

More ›