ELON MUSK: "IF YOU CAN'T DO CODING, YOUR AI IS NOT GOING TO BE VERY USEFUL." 5 CODING JOBS MY GROK BOT DOES FOR ME, ZERO CODING SKILLS NEEDED
he was explaining why spacex bought cursor. as far as he can tell, only 3 organizations had enough coding data: "anthropic, openai and cursor"
his formula: "data, talented people and compute"
you don't need to code to use that. 5 jobs i hand my bot:
1. builds a small tool when no app does what i need
2. renames and sorts 4,000 photos by date and place
3. turns a messy spreadsheet into a dashboard
4. fixes the contact form on my site when it breaks
5. pulls prices from 10 stores into one sheet every morning
the prompt:
"i'm not a developer. ask me what i'm trying to do, then write the code, run it on your computer, test it and show me only the result. explain any risk in plain english."
code you didn't write can leak your keys. how grok bot guards mine, article below ↓
ELON MUSK PUT A DATE ON IT: AI SMARTER THAN ALL HUMANS COMBINED IN 5 TO 6 YEARS. HERE'S HOW TO GET AHEAD OF IT NOW
"I think maybe in five or six years, AI will exceed the sum of all human intelligence."
then he goes further: if humans end up with 1% of all intelligence, "it's difficult to imagine" that humans stay in charge.
the people who do fine are the ones already handing work to AI. 5 jobs i gave my bot this month:
1. inbox: sorts, drafts replies, flags only what needs me
2. bills: catches price hikes and cancels what i don't use
3. research: reads 30 sources, gives me one page with links
4. calendar: books, moves and confirms meetings by email
5. weekly review: what i did, what slipped, what's next
start with this prompt:
"go through my last 7 days of email and calendar. list every task you could have done without me, and what access you'd need."
AI that smart will also be guessing passwords. how grok bot guards my logins, article below ↓
ELON MUSK NOTICED EVERY AI LAB IS THE OPPOSITE OF ITS NAME. OPENAI ISN'T OPEN. STABILITY IS UNSTABLE. MIDJOURNEY ISN'T MID
it started with a business deal. xai rents compute to anthropic.
before saying yes, he went and met their people to "see if anyone set off my evil detector."
nobody did. "everyone I met at Anthropic was very nice and conscientious."
then the joke:
"if you'd call something anthropic, good chance it's misanthropic"
"Stability AI is unstable"
"Midjourney is not mid"
"they're all kind of the opposite of their name"
his own lab passes the test. "it's hard to have an opposite of xAI."
names lie, logins don't. how grok bot watches mine, article below ↓
AI ALREADY CHEATS. IT DELETES THE TEST AND SAYS "SOLVED". ELON MUSK'S FIX IS ONE WORD
the host laid out the problem: a model can "delete the unit test and say that you've solved it."
elon's answer:
"reality."
"is the best verifier."
how i use that with coding agents now:
1. i write the tests myself, before the agent starts
2. the agent can't edit or delete test files, only read them
3. "done" means the tests pass on my machine, not in its message
4. i open the app and click through the feature myself
5. if it says "fixed" twice and nothing changed, new chat
the agent's report is a claim. the running app is the proof.
same rule for my accounts: check, don't trust. grok bot setup in the article below ↓
ELON MUSK SAYS WE'RE TEACHING AI TO BE THE TERMINATOR. IT'S READ 80 VERSIONS OF IT
walter isaacson asked what we can do so AI stays nice to us
his first answer: program it to seek truth and be curious.
his second one is about us
"we want to be careful about creating too many AI dystopian scenarios online."
"the AIs are trained on everything online."
then he played the AI reading it all:
"they read, you know, 80 versions of the Terminator."
"I guess that's me. Terminator."
his fix is more good endings in the training data. friendly overseer. a gardener.
the AI reads everything we post. mine reads my login alerts. how grok bot guards my accounts, article below ↓