If you're under 55 and not healthly, read this:
People mistakenly think they can abuse their body when young and bounce back.
This study found that when your biological age is older than your chronological age, your risk of early cancer increases.
> for example, if you're 25 but your biological age is 35, you are at increased disease risk, not just someday, but before the age 55.
> for the widest age gaps, risk ran up to 57% higher for lung, 31% uterine, 17% GI
> this study used basic blood markers for 2 of the 3 tests to determine biological age
> people are getting older faster. People born 1965 to 1974 carried an age gap about 0.23 standard deviations larger than those born 1950 to 1954.
> counterintuitively, after the age of 55, the increased disease risk effect goes away
> so the risk is in the young
@dreamspanHQ how does this compare to 8sleep?
I'm in india right now, was planning to buy 4 sets of 8sleep next month and this popped up on my feed.
reply or DM if you've used this.
>Be Elon
>Get bullied so badly as a kid that you end up in the hospital
>Escape into books
>Read more than 8hrs a day
>Teach yourself programming
>Sell a video game at 12
>Leave South Africa
>Sleep on couches
>Work odd jobs
>Get into America
>Build a startup
>Get fired from your own company
>Start over
>Build another company
>Merge it into PayPal
>Get removed as CEO
>Your company gets acquired
>Walk away with nearly $180 million
>Instead of retiring at 31, put almost all of it into three impossible ideas: Electric cars, Solar energy, Rockets
>People tell you you're insane
>Start a rocket company with no aerospace degree
>Learn rocket science from textbooks
>First rocket fails
>Second rocket fails
>Third rocket fails
>Divorce
>Public humiliation
>Cash running out
>One launch away from bankruptcy
>Launch anyway
>The fourth rocket reaches orbit
>NASA signs a contract
>Survive
>Tesla is weeks from collapse
>Save it at the last minute
>Get mocked for wanting reusable rockets.
>Land one.
>Then another.
>Then dozens.
>Turn science fiction into engineering
>Get mocked for betting on EVs
>Turn electric cars into status symbols
>Force the entire auto industry to follow
>Build the most valuable car company in history
>Launch astronauts into orbit
>Create a global satellite internet network.
>Buy Twitter
>Fire most of the staff
>Rename it X
>Walk into politics
>Risk your reputation
>Risk your companies
>Risk your fortune
>Become one of the most polarising people on Earth.
>Get attacked by the media, politicians, competitors, and activists
>Keep building anyway
>Become a TRILLIONAIRE
Karpathy went from 80% manual coding to 80% agent-driven in 6 weeks.
He called it "the biggest change to my workflow in 2 decades."
Most CS students reading this are still hand-typing for-loops in week-3 tutorials.
The gap between how Karpathy ships and how curricula teach you to ship is now ~5 years wide.
What that means in practice:
โ Stop memorizing syntax. AI handles syntax.
โ Stop writing functions from scratch. AI drafts them.
โ Start reading diffs. Reviewing is the new writing.
โ Start writing CLAUDE.md files. Your spec is the new function signature.
The skills that compound in 2026:
1. Naming things well so the agent doesn't drift
2. Reading AI-generated code in 30 seconds, not 30 minutes
3. Writing specs tight enough that the agent can't misinterpret
4. Knowing when the agent is bullshitting you
The best students in 2026 don't out-type the agent.
They out-direct it.
Source: @karpathy on X, Jan 26 2026
A Cornell + METR study confirmed what no AI influencer wants to say:
Senior devs using AI are 19% SLOWER than without it.
They FELT 20% faster. They weren't.
Almost nobody is reading this study.
The setup:
โ 16 experienced open-source devs
โ Real tasks on repos they knew well
โ Free choice of Cursor, Claude, GPT โ anything
โ Stopwatch on every PR
Result: 19% slower with AI. Self-reported: 20% faster.
That gap is the whole story.
Why it happens:
1. Prompting takes longer than typing in code you already know
2. Context-loading on every chat eats minutes
3. Review-edit-rewrite cycles fragment focus
4. AI confidently inserts subtle bugs that cost more debug time than they saved write time
What this means for students and juniors:
โ The productivity story IS real โ for code you don't yet understand
โ It inverts the moment you become an expert
โ Use AI as a teacher, not a substitute
The seniors felt fast because dopamine ships faster than your IDE compiles.
Speed is measured in merged PRs. Not in how clever the chat felt.
Source: METR + Cornell, 2025 RCT
I keep noticing the split โ some people I know have rewired half their workflow around AI tools. Others haven't opened one once, even though their job is mostly screen + words.
what's holding back? Curious, not judging.
Every time I let Claude touch a Python file, I'd grep the diff for os.system and eval before accepting.
Anthropic just shipped a plugin that does this automatically โ security-guidance, 8 vuln patterns, intercepts before the write.
/plugins to install. One less thing to check manually.