Smells Like ML

outrider proposes another feature for PEFT github.com/smellslikeml/p… just like peft PR#3382 it's the second outrider match we've validated against maintainer interests github.com/sockeye44/dora…

outrider proposes another feature for PEFT github.com/smellslikeml/p… just like peft PR#3382 it's the second outrider match we've validated against maintainer interests github.com/sockeye44/dora…

over successive runs on my forks, the orchestrator learns to target these surfaces: atropos → reward_fns/ registry peft → src/peft/tuners/<method>/layer.py::forward opik → src/opik/metrics/ lm-eval-harness → metric registry OLMo-core → olmo_core/train/callbacks/

cheaper models like GLM help affordably scale code exploration, offering more context for my implementation strategy opus is great for concurrency but since I ran sequentially w/ GLM backend, I set up an adaptive search for a final, more fully-spec'd proposal

Want to know exactly which small changes would let your repo absorb the latest AI methods? Ran autoresearch across 5 repos. Got a gap analysis with prioritized next steps. gist.github.com/smellslikeml/3…

@simonw over this last week, way more of my code is made using github actions than back and forth chats with claude

Personal skill will get even @simonw only so far we have to start building better ways to scale code synthesis Been pushing my own limits using a Github action to implement PRs across dozens of OSS repos each day It's open and free Get Outrider: github.com/remyxai/outrid…

the competition is beyond reading the paper, they're adapting methods to their own repos get outrider: github.com/remyxai/outrid…

matching olmo-eval too github.com/smellslikeml/o…

Went a step further and implemented Outrider is a github action to match new methods to your repo, implementing when there's a good fit github.com/salma-remyx/lm… get outrider: github.com/remyxai/outrid…

cost about a dollar using GLM-5.2 github.com/smellslikeml/l…

only been out for 5 hours, I was the first to star it And only a couple days for the preprint (arxiv: 2606.30552v1) but Outrider found it, matched it to lerobot and implemented a PR github.com/smellslikeml/l…

Linearity assumptions are computationally convenient but reality is full of curves to slow you down Recently found "Riemannian Preconditioned LoRA for Fine-Tuning Foundation Models" (arxiv: 2402.02347) Method matched for Unsloth and PEFT github.com/smellslikeml/u…

the math share.google/aimode/RxTOnTf…

@livekit here's an idea: github.com/smellslikeml/a…

Join us at the official AI Engineer World's Fair Hackathon in San Francisco, June 27 to 28. 48 hours to bring an idea, build a voice agent on LiveKit, and ship it. Our team will be there to help. Come build something cool. Register here: cerebralvalley.ai/e/aiewf-hackat…

Comparing Claude Code with Opus 4.8 vs GLM-5.2 backend Paired runs implementing the same methods for the same repos GLM uses 7X more input, 2X more output tokens GLM cost 10X less overall GLM 2X slower GLM 5.2 commits while Opus questions gist.github.com/smellslikeml/3…

@bcherny seems like these functions too would be absorbed by coding agents Here's one packaged as a github action for prototyping: github.com/remyxai/outrid… But why stop there? We'll follow up with more validation so that when a person is reviewing, there's already a strong case to merge

As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes: 1. Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship 2. Builder: quickly turns a prototype/idea into production-grade product/infra 3. Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance 4. Grower: takes a product that has been built and iterates on it to improve Product-Market Fit 5. Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS. A healthy team needs a mix of these, depending on the product: - A product that is new and pre-PMF needs people that are strong at 1+2+3 - A product that is growing and has found PMF needs 2+3+4 and some 5 - A product that has strong PMF needs 3+4+5 and some 2 Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?