💡 Remember Box
https://x.com/Shang_mit/status/1993749446467703214
implement adaptive speculative decoding in mlx swift #todo
Find two distinct ideas from disparate fields and create a novel connection between them. One that most people or no one would have noticed before. Make it truly unique. Have it be rich in detail but also use economy of language and succinct.
💡 Remember Box
even native speculative decoding in mlx-swift doesn't exist
GitHub
GitHub - mlx-community/speculative-decoding: Native speculative decoding implementation for fast LLM inference on Apple Silicon…
Native speculative decoding implementation for fast LLM inference on Apple Silicon using MLX-Swift. - mlx-community/speculative-decoding
how to clarify your purpose
- find the pain you can sustain
- locate your unfair cognitive advantage
- use purpose to kill options (reduce entropy)
- A/B test your life
- find the pain you can sustain
- locate your unfair cognitive advantage
- use purpose to kill options (reduce entropy)
- A/B test your life
💡 Remember Box
how to clarify your purpose - find the pain you can sustain - locate your unfair cognitive advantage - use purpose to kill options (reduce entropy) - A/B test your life
pain I can sustain
- hard technical work with uncertainty
- high-ambiguity reasoning
- long arcs of problem-solving without external validation
- learning new stacks fast and applying them immediately
unfair cognitive advantage
- pattern detection + cross-domain synthesis
- fast prototyping instinct
- early signal sensitivity
- hard technical work with uncertainty
- high-ambiguity reasoning
- long arcs of problem-solving without external validation
- learning new stacks fast and applying them immediately
unfair cognitive advantage
- pattern detection + cross-domain synthesis
- fast prototyping instinct
- early signal sensitivity
or in other words, to find overlooked edges and build tools that make the future arrive early
> Build workflows
> Monetize every run with x402
> Turn them into a product
> Tokenize it
> Monetize every run with x402
> Turn them into a product
> Tokenize it
prompts I use often
- make a comprehensive analysis of current state of the app which and check api also, then improve each and every part and route of it, as it was mostly build earlier by an intern and we can't know if it all works, also consider each section and workflow then add unit tests for each as well and refactor it overall, improve UI/UX too
- improve entire app UI, check where it can be improved to make the UX super good
- make small commits and push
- resolve merge conflicts
- review everything and improve if there's something you can
- make a comprehensive analysis of current state of the app which and check api also, then improve each and every part and route of it, as it was mostly build earlier by an intern and we can't know if it all works, also consider each section and workflow then add unit tests for each as well and refactor it overall, improve UI/UX too
- improve entire app UI, check where it can be improved to make the UX super good
- make small commits and push
- resolve merge conflicts
- review everything and improve if there's something you can
[2507.13334] A Survey of Context Engineering for Large Language Models
https://arxiv.org/abs/2507.13334
https://arxiv.org/abs/2507.13334
arXiv.org
A Survey of Context Engineering for Large Language Models
The performance of Large Language Models (LLMs) is fundamentally determined by the contextual information provided during inference. This survey introduces Context Engineering, a formal discipline...