rmst

I made this interactive reference for transformer models, showing everything down to elementary math. I intentionally avoided matrix multiplication and used explicit sums and indices instead. Covers models from GPT-2 to Qwen 3.6, with MLA, MoE, RoPE, MTP, etc togglable. It's best viewed on desktop or tablet.

It doesn't matter if this isn't happening in any useful form in a reasonable amount of time. It's still great for hiring and funding.

He has used the same hiring technique since x.com:

1) announce overambitious, relatable, unoriginal goal (e.g. human settlement on Mars)

2) present vague, relatable, unoriginal strategy (e.g. reusable rockets)

3) hire people

4) actually have them work on more incremental stuff (e.g. launch provider services)

5) profit

Classic Musk hiring technique:

1) announce overambitious, relatable, unoriginal goal (e.g. human settlement on Mars)

2) present vague, relatable, unoriginal strategy (e.g. reusable rockets)

3) hire people

4) actually have them work on more incremental stuff (e.g. launch provider services)

5) profit

How convenient for Apple. Now they finally have a good argument to keep forbidding JIT compilation and side-loading.

Log cumulative graphs make it hard to see whether the daily rates are increasing or decreasing. Here are non-log, non-cumulative, new daily deaths per 1M people to compare different countries:

https://colab.research.google.com/drive/1dNAVpgRGjEViK-9ULhE...