@nrehiew_ do you plan on running Sol or Kimi on your over-editing eval? I do think Sol's edits are less verbose on high without explicit instructions, but I'd love to see data on it
🎼 Can we turn music into readable code?
Excited to share Decomposer, a post-training framework for symbolic music decompilation!
Given MIDI, Decomposer recovers an executable program in Strudel that reconstructs the music while exposing patterns, harmony, rhythm, and voices as editable code.
🧵 [1/5]
- contract to hire
- remember you're not in Sweden and can fire whenever - focus on hunger+IQ+taste+your gut (vibes). Taste and eagerness can compensate for IQ and viceversa. It's a lot like LLMs (does the job call for Fable or Sol/Terra?).
- realize you're a unicorn. You're getting bullshit cause that's all there is to most jobs and most people's lives, whether due to circumstance or low agency. But humans are general intelligences and can do great things if you give them the chance to do so.
@sevensixfive You should check out Paranoia Agent if you haven't! He took a collection of ideas that didn't make it into his films and made an excellent miniseries out of it.
I loved Millennium Actress too. It's been so long I don't remember any of it but I'm sure it's great
Excited to publish the MirrorCode paper and open-source codebase!
The key tension in evals is now: how to make tasks that are difficult for AI, but actually fair? My default expectation in 2026 when I see a coding eval with low scores is that the tasks turn out to be impossible
@fahrudotme@dillon_mulroy oh lol yeah I see what you mean. I think this doesn't apply here cause he's just trying to extract coding style patterns, but the "tell me about myself" thing can be a dangerous slippery slope for sure!
imo this is the good kind of “psychosis” and you should consider putting this in an AGENTS.md! People seem to have completely given up on steering the model this way and I think it’s a massive overcorrection. @gabriel1 has an article on guidelines for simple code. Pasting a modified version of it to my AGENTS.md has made a tangible difference for me.
Today marks the beginning of our launch calendar and to celebrate i am making ncode and our flagship kimi k2.7 model free to use for the next week (or until the traffic knocks us out). all you need to do is:
1) sign up for a noumena account at https://t.co/bJmg9QPGUH
2) go to https://t.co/2wBMNAS3mL and clone and build noumena code (ncode)
3) login to the platform with `ncode auth login` (or /login once you are in the app)
4) enjoy blazing fast tokens on the noumena platform with ncode
>success is determined by how well a person understands the problem they are trying to solve
>while agentic coding tools may be absorbing some implementation-heavy work, they are also rewarding those with firm understanding of the problems they solve on the job