I started working at 24. I'm now 32.
Already, I've taken 2 career "breaks" to learn new things.
I'm often asked: wait, how?!
Here's a list of reasons to take these breaks yourself.
And tips and tricks for pulling it off.
๐งต
@jacobgold thus far i've been doing payload -> execute (+ arena design, review, address, repeat, etc.) which makes the "ok there's multiple related things, let's step back and give this the attention it needs" harder to operationalize.
@jacobgold one thing i just learned from lauren's recent interview that helps clarify this for me: orchestrator agents that collect feedback (eg N payloads over M minutes) and then decide what to do with it. guessing those actions are e.g. wait for more / delegate work / escalate to human.
@jacobgold how do you create a line between "ingest feedback into the turbo kitchen" vs. "step back and -- OK this is an important design decision, i should think through this carefully // involve other people"? what guardrails or "loops" enforce this line, and how's your precision/recall?
@jacobgold appreciate it. how would you onboard a new developer to this codebase? what level "understand how it should behave in my head" do you think makes sense? how do you help them attain this // what is the modern parallel for "here's an easy ticket, then a harder one, etc."?
@poteto tx for the great content! q: what level of ongoing code *understanding* do you believe necessary, and how do you yourself achieve this? what is the lowest-level construct you review every day (e.g. app data model)? what tools/processes/viz/etc. do you find most effective?
Excited to share the launch of @cohere's newest vision-language model, North Micro Vision. This is Cohere's smallest VLM yet that delivers strong performance across visual understanding benchmarks, with particularly impressive results on document understanding and visual Q&A.
The weights are now available on @huggingface, enabling broad access for research and development of visual understanding applications that can run locally. https://t.co/CrOv92b3dG
๐จ Excited to share my recent paper: "Monopoly Deal: A Benchmark Environment for Bounded One-Sided Response Games."
Card games have long served as benchmarks for sequential decision-making under uncertainty โ from Poker and Hanabi to Magic: The Gathering. This paper introduces a version of Monopoly Deal designed to study a less-explored interaction pattern, where an action briefly transfers control to the opponent for a bounded sequence of responses.
We show that standard Counterfactual Regret Minimization (CFR) adapts cleanly to this structure, and share a lightweight research environment that integrates the game engine, parallel CFR runtime, and web-based interface โ all runnable on a single workstation.
๐งฉ Paper, demo, and code:
๐ค https://t.co/50wKU2Rpzn
๐ https://t.co/iaVKy4GKWp
๐ป https://t.co/DFWFmQZTZ5
@willium@throughnothing @itunpredictable Looking forward as well
Fun facts: I also grew up on a street that starts with W in a town that starts with W
Do I get the W on Wโs?
omg