kenny. retweeted
I spent some time on the weekend to test bun's AOT compiler and results are nothing short of amazing. Performance up 2-3x, memory use less than half.
Bun AOT also introduces `--trim` which deletes unused parts of Bun from the compiled binary
On macOS, a binary that uses Bun.serve’s HTTP server goes from 89 MB to 25 MB
ALT bun build --compile --aot --trim, with --trim highlighted. A block representing an 89 MB executable from Bun 1.4.4 on macOS arm64 is being cut up with scissors. The part that stays is pink: 25 MB for hello world, 34 MB with Bun.serve, 44 MB with node:http. The rest is cut into paper scraps, each crossed out: bun install, bun test, bun check, bun build, bun lint, bun fmt, WebAssembly, debugger, JIT compilers.
kenny. retweeted
a growing part of how I feel about bun is
this whole category of software - linters, formatters, package managers, type checkers, test runners, bundlers
this used to be hard software to build and now it can be prompted into existence in a day or two
Personally I think the shadiness of inference is a much bigger problem.
The fact that they can nerf models in realtime where some days it feels unusable and dumb.
That kind of bait and switch should be illegal
There’s a feature I fell in love with that emdash has. Extremely simple feature
“Push to main”
It allows me to quickly push a worktree into main. I honestly can’t use any harness without out. On personal projects I hate creating PRs
Would be great if conductor had this
It’s more the post you quoted, and the entire framing that Anthropic is giving away “$5000 worth of compute for $200”
The truth is that it’s getting cheaper for Anthropic to run these models. The distilled models are better and better and cheaper and cheaper. Their API pricing is just expensive while their monthly subscription is more inline with the real cost.
I hate the “subsidized pricing” argument.
Token prices are a race to the bottom, it’s going to continue to get cheaper. That’s why open source models are always at the heels of the cutting edge models.
If Claude/Codex didn’t subsidized we’d all just be using Kimi 2.5
The moment they stop subsidizing we’ll happily switch tomorrow open source models which will continue to improve.
Opus: Create 100% test coverage of this open source repo. Create as many passing tests as possible.
Do this again and again and again in a loop.
I know this is bait but there’s 100 ways to fulfill your prompt
If I ask a professional chef to make me chicken, I’m sure it will make me a perfect chicken dish. But is it the one I wanted?
- fried chicken
- chicken breast
- chicken wings
- chicken quarters
- chicken nuggets
.
I don’t have a full day, but I did have an hour which is all I needed to have a fully working cloud agent proof of concept.
Here is innie, a cloud agent that works with opecode, opus and GitHub. You can connect it to any GitHub repository, tag it using @innie, it can create PRs
100% vibe coded in an hour, I tested it using real examples of issues and Opus just completely nails it every single time.
github.com/codebykenny/innie…
I’ve been using “add to daily note” as a workaround to quickly create a note lol so glad this is officially coming out!
But if you already have the desired end state, why go through all the trouble of recreating everything? Other than a clean git history, which doesn’t really matter when you squash and merge the pr, what’s the benefit?
Does SynthID work to detect this as generated? Is it easy to bypass or change the image to make SynthID unusable?
deepmind.google/models/synth…
Atlas has no vertical tabs, dead on arrival for me.
Literally only use Dia for vertical tabs, wish I can nuke all the AI features.
@heliumbrowser vertical tabs when?