🇯🇵 At @rubykaigi, members of the Ruby & Rails infrastructure team presents ideas based on their work:.1. @Love2Code shows the working prototype for a next generation Ruby JIT compiler 🌄.2. @alanwusx explores interesting interactions between Ruby Call-related features ☎️

@Love2Code @ClementDelangue

@Love2Code 3blue1brown live talk: LLM visualizer:.Serrano Academy YT channel (check out the author's book as well): .

@Love2Code

@Love2Code Solar in China is the result of a costly top-down industrial policy. Reportedly, the government might cut it back soon. and we shall see what happens. The cheap solar panels exported by China are made with coal. Solar panels have a very significant waste problem. After 15

@Love2Code @8teAPi Crazy.

@Love2Code not to discourage writing a parser, but wanted to point out

@Love2Code There is already and for a long time:

@Love2Code @SawyerMerritt my thoughts exactly.

@Love2Code Very powerful for multimodal applications, especially robotics. Idea is to “tokenize” visual inputs + language instructions, generate robot actions (e.g., next pose of robot hand) as text. OpenVLA ( is built w/ Llama-2 7B. More recent VLAs use Gemma 2B!.

@Love2Code Smaller models mean higher policy clock rates (more actions / second). This is useful for dexterity, and more complex robots (humanoids). Small LMs adapted for general visual reasoning (captioning, OCR, detection) also great / useful for edge ML — see

@Love2Code I knew about some of the basic use cases like sorting DAGs and DFS to find dead/unused variables. This one (graph coloring for register allocs) is something I didn't think about. Graph coloring is NP. I wonder if this means that register allocation is often suboptimal.

@Love2Code In matters of taste!.

@Love2Code 🍅💥.

@chris_j_paxton @Love2Code Can't go on with the affiliations these days. Practical problem solving is different from academic idealisation.

@jimmyhmiller @filpizlo Awesome thanks Jimmy! 🙂.

@Love2Code It’s extremely good.

@Love2Code Good point.