leetcode ❌ TypeSafe AI Jev and video games ✅

@typesafeai I need to see if I’m more hireable than Dillon

@typesafeai Big oof

@typesafeai yeah dont take my job dawg

@typesafeai I think you’re safe 😭

@typesafeai granted stacking 9 of the same games probably wont get you hired anywhere 😅

@typesafeai I need a job to pay for my crippling Morrowind addiction though

@typesafeai Serge, curious on how you've put the LLM-vs-Jev pipeline there. So there's a Jev stage that scores the various work related attributes of the description. What's the other part does with the game, and how do you connect the two?

Hey Liran, great question The only "regular" LLM is Gemini, which just reads the 9 game titles off the card image. Everything after that is Jev, doing 3 things: 1. takes each game's title + its Wikipedia description and scores it on 8 skills (planning, teamwork, persistence, etc.) 2. takes the job posting and scores how much the work needs those same 8 skills 3. takes the job posting + all 9 games (titles and descriptions) and gives one fit score, which becomes the hire / no hire percentage (could have done a noul check here probably) 1 and 2 don't influence 3. I ran evals with Astra and Fable models as the judges, and both computing the score from the skill numbers and feeding them to Jev did worse than letting Jev read the raw text. So 1 and 2 only power the breakdown shown below the score.

@typesafeai link? 👀

@typesafeai

@typesafeai banger content

@typesafeai Good job man

@typesafeai thanks brother!