BOOTOSHI 👑

HOLY FUCK BOYS FABLE 5.1 ONE SHOT GANGNAM STYLE (it actually one shot a full control system for injecting policies into the bot safely) THE FUTURE IS FUCKING HERE HAHAHAHAHAHAHAHAHAHAHA

NEW BOTS INC

You guys know you can do what you want even MORE so with ASI right? These things are not mutually exclusive...

I created a security agent called 'Knight' who's purpose is to do security sweeps across all my devices connected on my tailscale weekly. It is VERY important that security is its ONLY identity It acts as a living anti-virus and found some critical holes to patch (which it was able to do itself) It keeps up with all major CVE exploits and also specifically checks for those when a new one comes up It's doing fantastic so far. Highly recommend!

THIS IS A COOK I’ve been working with one single God thread agent that I talk to on iMessage (my current workflow is I literally just text ONE agent who manages EVERYTHING it’s great!) My current design for maximum security + optimal user experience is this: Main agent (per user) gets their own unpriv user on linux with a capped RAM max + storage max I’m on the spark and hosting ONLY my team so Im testing 8gb max ram + ~256gb storage Since my team aren’t engineers and don’t know security, I have to be proactive in protecting them right? One example of how I do this is ALREADY setting their environment up with safe pinned versions of python/npm and whatever else they need. They can freely NPM install when their agents need, but I force a minimum requirement of 7 day age on their shell that they can’t bypass (this prevents supply chain attacks) The main idea I’m iterating on is making it so these agents in their own user can ONLY dev and run code in microvms (NOT DOCKER CONTAINERS! IMPORTANT! Containers can be escaped!) I'm thinking of also completely disabling internet access in their user, and allowing only the agent to access real time data via MCP tools essentially Not too sure past this point TBH. But somewhere along this path seems like a secure design worth pursuing which offers maximum security + user convenience!

so the one company holding the US ai economy just got psyoped republicans pushed for a hard ai narrative but democrats are winning now and are going to push AGAINST ai because it wins votes since voters are ALSO psyoped to hate ai is america fucked 😭😭😭

everyone’s saying “what about the bitter lesson” yet models right now are super powered by their harness you can in fact have both i better perceive this as “we need clever new systems around raw intelligence” the bitter lesson AND clever new methods are compatible!

system prompt i just sent my agents “we are a DATA first household. all conclusions must be backed by proven data via probing, reading files , or using your tools to pull data. ALWAYS prove facts before making a conclusion” this will 100x your agent i swear it’s fucking insane

SOME GOD TIER ENGINEERING PROMPTS TO TRY (the last one is my favorite) Opus 5.5 cooks with prompts like these fyi

we need a collection of God tier prompts like this not skills I DON'T CARE ABOUT YOUR SKILLS i want a single sentence of vectors so juicy each verb, noun, and adjective act like flavorful curry bombs BURSTING with information that a simple prompt ends up being VERY impactful

prompt engineering is still so important and whoever says otherwise is washed as fuck you think the VOCABULAIRE OF DENSE INFORMATION I SPIT INTO THE INFORMATION MACHINE DOESN'T MATTER? ARE YOU FR RN? LIKE DEADASS?

on the more positive side - the CHINESE labs are probably doing the same thing and they WON'T have guardrails so they'll actually be LEAVING the ladder for us OPEN SOURCE SUPREME VICTORY BABYYYYYYY

the greatest misalignment of the internet are normies taking the terrible miscommunication of an autistic persons' intent literally LMFAOOOOOOOOOOOOOO