don’t forget the submission proposal docs! (for those who publish)

*clarification: performant meaning quality of output. imo it is a more performant model in terms of TPM and environmental impact. never an absolute in this space, and variety in the ecosystem is good.

deepseek (another transformer model, with some spicy RL) itself is not perceivably more performant. but they did it for pennies on the dollar.

bsky.app/profile/msat...

if you’re curious what prompted this, i am 1/8 on acceptably wrapped gifts.

“yeah i’m really good at something easily reversed in 1.5seconds. i could russian doll it, but consumers are showing some pushback with that.”

“the scenario where you describe acceptable llm use is known as burnout” well, yes. but the cure to burnout is a break from your job or coding in general. and some people simply can’t afford that (due to lack of PTO or otherwise).

it’s much less of a concern for someone with many YoE who has the wisdom and keen eye to rule out poor suggestions from LLMs. (3/3)

like athletes, our readiness for problem-solving undulates. long stretches of unmotivated, uninspired work can be “averaged out” by copilots, letting you find the magic again when you’re ready. code quality is an extremely valid concern, but…(2/3)

This sounds critical of the 🦋 team, it’s not. I’m just saying we as developers have to collectively come up with a better solution.

i believe good software design follows occam’s razor. no more “staging branch”. everything merges to main after CI/CD passes, and bw lets you deploy to whatever environment, given you have the permissions.