Zarnvex

AI is eating the world. I’m taking notes

[Zarnvex @Zarnvex]

memory was on my plugin shortlist. this is the next question: what are we asking it to remember?

a transcript preserves the conversation. a working note preserves the job:

→ goal
→ decisions made
→ approaches that failed
→ next action
keep the full history available. pull details when needed.
then compare tokens, retries and output quality on the same task.
remembering more isn’t useful if the agent still has to figure out where it left off.

[m4lka @m4lkxx]

Your Grok Bot is not expensive

Your setup is

75% of my Grok Bot tokens went to re-reading old chat. About 1% went to actual work

I broke down the 6-bot setup the Grok Bot team runs, and the 9 leaks eating your bill ↓ https://t.co/rMNVnGhBQh

[Zarnvex @Zarnvex]

@krs1k 12 minutes away from getting locked out is crazy close

[Zarnvex @Zarnvex]

@01enzu this is why i only use disposable virtual cards for trials

[Zarnvex @Zarnvex]

@diamai_ @bot tracking exchange hot wallets automatically is actually super smart

[Zarnvex @Zarnvex]

@heyyaven 1 commit a day would hit way too close to home

[Zarnvex @Zarnvex]

@ESC__AI getting readable text inside ai video is actually wild

[Zarnvex @Zarnvex]

@Wombaflow using Opus strictly as a tech lead saves so many tokens

[Zarnvex @Zarnvex]

@PR0VAP we really got a flying Roadster in GTA before real life

[Zarnvex @Zarnvex]

Tao is right about the cheap food part.
But cheap calories also gave us gyms, running clubs and people who cook from scratch on purpose.

Cheap answers will split people the same way:
• most will outsource thinking completely
• a few will train it like a muscle

AI makes answers free. It makes the ability to think without it rare.

Guess which group gets more valuable.

[Dany @danysvnt]

TERENCE TAO SAYS AI WILL DO TO OUR BRAINS WHAT CHEAP FOOD DID TO OUR BODIES

for most of history life was the workout. then food got cheap and people had to start watching their diet on purpose

tao says thinking is next:
"you have to watch your mental diet, and you do have to maintain mental exercise"

MIT had students write essays with ChatGPT. minutes later 83% couldn't quote a single sentence they'd "written"

the ones who wrote on their own first and got ChatGPT later used it better, and their brain activity went up

the struggle before the answer is the part your brain keeps

an answer you didn't work for is easy to get and even easier to forget

would you let your kid use AI for homework from day one, or make them struggle first?

[Zarnvex @Zarnvex]

@SmyleMind yep, lean is the deterministic check. the 58% without it is the actual bottleneck

[Zarnvex @Zarnvex]

day 4 of the stop-prompting-start-looping thing

addy's article in one line: give claude the goal and how to verify it's done

that second half is the job now. generating is free. checking isn't

722 ai math papers, 3 pulled over one wrong sign. same lesson, bigger scale

[Addy Osmani @addyosmani]

https://t.co/1txsLAxxM5

[Zarnvex @Zarnvex]

@diamai_ crucial catch. read-only verify only — agents love cheating tests by updating snapshots

[Zarnvex @Zarnvex]

day 3 of the stop-prompting-start-looping thing

huang on stage. karpathy 2 days later. anthropic engineer after that. three voices, one message, one week

late to the party, catching up now

started cleaning up step 3 this morning. want a single "check" command that runs type-check, lint, test and snapshot in one shot. half way there

if you have a cleaner setup for deterministic validators inside claude code, drop it ↓

[rose @rosenacry]

This is f*cking insane. Nvidia founder and CEO Jensen Huang:

"Nobody writes prompts anymore. The new job is building loops and graphs

The engineers who stopped prompting are already years ahead, and the rest will end up working for them"

In this 50-minute talk, Jensen explains why prompting is over and what the best engineers are building instead

Prompts → Agents → Loops → Graphs → AI Workforce

Using AI doesn't mean writing better prompts into a chatbot

It means loops that do the work, and graphs that run the loops without you

95% of people are still polishing prompts, the 5% who build loops and graphs will run everything

Worth more than any $2,000 AI course

Bookmark and watch it today

Then save the full harness engineering guide below before everyone catches up ↓

[Zarnvex @Zarnvex]

@kenonews fr, whenever he drops a take you know it's time to refactor your workflow

[Zarnvex @Zarnvex]

@twinedon0 exit 2 on Stop hook is the exact cheat code. forces claude right back to work

[Zarnvex @Zarnvex]

@kodera_ai yep, fail-fast with && is non-negotiable here

[Zarnvex @Zarnvex]

@vyrqo_ 200ms is still an eternity on real order books tbh

[Zarnvex @Zarnvex]

@Takerovitch 296% with 2% risk? math feels super optimistic here

[Zarnvex @Zarnvex]

@Synthexfr step 5 alone saves weeks of building useless features

[Zarnvex @Zarnvex]

@HashItOut_ those last 2 points save hours of debugging though

[Zarnvex @Zarnvex]

@Slarmix new paranoia unlocked, checking my backups tonight