Why is a plan "mode" needed? I really hate reading plans in conversation UI. One line in AGENTS.md is always better, and you can customize it to fit your document convention.
From the communication perspective, the planning process is most about creating clarity and alignment, on things like scope, constraints and decisions, among people (and agents now), often requiring multiple rounds. The planning tool built in Codex or Claude Code is not persisted, version-controlled, nor well accessible.
In practice, My AGENTS.md contains an instruction about writing a document before starting implementation. Usually I won't read that document, because I don't want to micromanage agents. That document mostly serves a historical purpose like ADR, helping me find out what agents missed, made mistakes on or misaligned on, if needed.
However, the bank is happy to extend the loan infinitely for people with enough assets. It's questionable that whether such loans are cash flow neutral.
I'm interested in how to integrate formal verification with existing software libraries. For example, https://github.com/verus-lang/verus integrates proof with macro in Rust. What's the plan for Bend?
It looks like a Transformer encoder post-trained on classification and regression tasks. The encoder-only model is less noticed in recent years, but this product finds a nice application for it.
- can it be used to relax timing order requirement in pipeline parallelism? Each node update lambda on communication, and optimize weights at other time.
- given that BP is using SGD, can batches and T share the same timeline in optimization, while keeping the descent direction?
Nice introduction to a simple but useful idea! The Lagrangian works like a time-smoothed optimizing direction state, but it can be placed on any wire, even at non-differentiable boundary! Can it be better than existing training methods for discrete components like argmax, MoE or VQ-VAE? Maybe networks can be composed by a lot of learnable discrete components, or even bits and gates finally.
Funded positions are much less than PhD number. However, people can fund themselves with a job. In a country, if it's easy to to get a part-time job with enough payment, mathematicians can continue work.
With agent asssistance, we don't need writing annoying formal spec and proof anymore. Then formal verification can be a practical and useful tool in daily programming, especially for "deep module" whose spec is much simpler than implementation.
It seems the software productivity has exploded in the agent era. However, do we treat every point of daily experience seriously enough? Do we recognize such pains actively and try to eliminate them (often easy), or just tolerate infinitely?
This is the first shoelace knot my parents taught me. We call it "butterfly knot". It's always my default choice with muscle memory, unless the shoelace is not long enough to create the "bufferfly wings".
It's secured enough for daily use, and easy to release. Also the best choice for apron knot.
I prefer synthetic dataset since the first day hearing distillation. The engineering friction is much lower than soft logits, and I have not observed or heard performance loss (in Speech and language area).
Could you share some latest articles or papers comparing both methods, especially on lanuage modelling case?
I was not conviced by this claim when reading the original Knowledge Distillation paper. ChatGPT said there are some later works showing: 1. the gain may come from label smoothing; 2. soft logits are more meaningful for students much smaller than teacher.
I really hate modern time schedule. It's nightmare to be forced to get up 6am or 7am every workday since childhood. The only relief is natural wakeup on weekend.
The proof will be more friendly to nowadays programmers if we treat all "Gödel numbers" as bytecode of a programming language.
It's trivial that functions like "prove" and "subst" can be implemented based on abilities like bytecode parsing and expression tree manipulation.