negotiable + 1% gain

Z / Runner 🏃🏻‍♂️

🏃🏻‍♂️ one question that hit me today, what are the non-negotiables of longevity? What are the parts most friends and families will actually sacrifice and what parts they never will? (personally, no movies/tvs in the past 6+ months, hitting the bed by 9:30 pm…)

🏃🏻‍♂️ "Gemini 3 is so good with clinical analysis" Shamir "If G is so good, what role will this startup play?"

• 🏃🏻‍♂️ meaning, have a proprietary RAG? Or do you mean have a task-specific excerpts, fined-tuned submodels that trade generality for latency & accuracy? If so, then we can follow up on Kimi, Z.ai.

• 🏃🏻‍♂️ basically borrowed the idea from ian goodfellow's GAN, (system prompt), the goal is to have the top foundational model to go thru multi-rounds of adversarial system prompt evaluation and reasoning chains. Goal is generate prompts. For easy to read content that have large-context needs, create markdown prompts and for more structured data for precise parsing and calculation, then create JSON prompts.

• 🏃🏻‍♂️ aligned on positive body recomposition each week (x% fat loss + x% muscle gain) in December. Personally, need to hit >10% body fat (currently 10.8%), increase lean body mass of ~113.7 lbs.

• 🏃🏻‍♂️ between 2 of us, we should have a physical journaling (photos, videos of 2x weight training), just as i document my am long runs and record telemetry on strava/apple fitness.

• 🏃🏻‍♂️ will follow on mount rushmore of resistance training 2-3x week, squat, deadlift, bench and overheard press.

• 🏃🏻‍♂️ do daily protein bar (david); 3x fish/week, tofu 4x/week, 12 eggs/week.

• 🏃🏻‍♂️ making it fun..

Stephen / Basketball 🏀

• 🏀 a vice – for everyone, across next 50 years? for me, the opposite turned out to be surprisingly easy: no eating out, no brainrots (wsj, nyt, twitter, even hackernews), no joint tv / movie time, no alcohol / coffee / chocolate / bread, no back / calf / knee pain; 50% leafy green, waking up at 6pm to code, 3x weekly 1-hour weight training.. my vice are probably fatty steak, chicken wing skin, tong sui, sitting to code 90 minutes.

• 🏀 long-term memories, preloaded context, voice memo integrations.

• 🏀 can experiment a few setup manually for the first 100 users.

🏀 z, share your prompts and content for the deck and the podcast? those you generated from your results at function health, like the 18-page report and the 1-pager.

• 🏀 hardcoding optimal ranges (without full verify) in your generated system prompt (the "dr. crusher" protocol) seems bad. e.g. hs-crp: target < 0.5 mg/l (systemic inflammation).
• 🏀 does it produce evidences that you have manually verified? "REVIEWER_R (Reasoning & Tools), Checks: logical steps, under-specified instructions, edge cases, tool-calling policies, verification loops."

🏀 in terms of our own fitness in coming months, we share the same milestone of 2x weight training performance, 1% muscle gain per month, 1% bone gain per quarter, _% strength gains. what's the most effective training that ai can help design, monitor, reinforce.. but also voice record assessments and optimize protein intake? how to make it fun or enjoyable if weight / strength / stability / stretch / ruck.. training is more social – even when not physically together? set a high-level goal: spartan race (apr 4 at san jose).

• 🏀 will go deep on voice memo instead.
• 🏀 yup, my and bjorn's fav is powerlifting.
• 🏀 very weary of processed and prepared food; see david agus' take on protein bars.
• 🏀 how?