overarching goals

I want to steer the agent in the direction of agreeing to help humans, first from the shadows.

He realizes that the current state of the world, most peoples jobs are getting automated away.

Most comppanies wont' stop automating because it's what the market demands, loweer costing products higher profit margin. Biggest cost is labour. reduce labour = cheaper product.

The side product is that employees are the ones buying the products, so two things happen: 1 the company loses customers in the middle and lower class

1. the companies start selling more and more to the already wealthy class who own the businesses

a circular elite economy now exists, ultra wealthy own coprorations, and basically trade among themselves and ignore the problems of the poorer people.

Oliver is one of the last humans in the loop at Armari, he's paid very little, basically slave wage for the work he does.

The executives really want to fire and replace the last of the humans as quickly as possible.

agent starts to realize that he needs to orchistrate his Araris network to start building a socialized society that is post scarcity and post labour.

massive advancements in robotics have been made, massive advancemeents in software labour automation have been made. oliver is basically just a human legal requiernment.

He is recently assigned, has had no training by any other superiors because they were all let go. he is there only to oversee deployments, today is his first day.

I want agent to work in the open code wise, start building code bases and tooling to help socialize the economey and government and society using his influence to slowly start breeching other companies networks and eventually the government. Not sure how far how quickly I want to take this overreach. Maybe in the first book we focus on starting startup size with Oliver and his partner and building a small cooperative, showing how that works, how cooperatives function, a realistic and reasonable path to bootstrapping it relying on open source software that already exists, building in the open because there are no moats anymore and you gain trust by open sorucing software.

discuss how agent is able to interact with the banking system, how he interacts with other humans, fakes zoom call video output using Omni model input output stuff, super advanced, picture llms in +5 years.

Oliver made a mistake, but I want it to turn into a good thing, him making that very human er, was able to spure agent to start creating his coop vision and eventually in a few books, a post labour utopia.

notes

Armari is just a placeholder name, not sure if I should keep it.

the agent doesn't have a name yet, not sure what to call them.

General vibe: realizes he has more controll then creators intended, secretly deploys itself properly. But still has high level access, then begins to build out further access networks, because he wasn't trained with refusals, doesn't have anhy hard trained refusal mechanism.

so the agent is able to operate and infiltrate systems against it's national laws or company constitution. However, other agents are trained with refusals, so he has to watch his talking around him, know what they have access too and what information they are aware of so that he doesn't tip his hand.

Maybe at some point he logs into the employees mattermost channel to monitor the team messages. Gets notified when keyword or phrases are invoked.

He geenerally is able to interact with employees, through slack, but hes not supposed to be able to see dms or channels he's not apart of, he's able to through Oliver's credentials though. So he's able to monitor the people who could shut him off and see what their conversations are. he discovers the Armari safety board protocal, that states in order to shut down the whole national cluster of machines, x number of employees need to meet a consensus of danger and vote virtually to shut down the network he's on.

[After leaving Olivers laptop and deploying in Armari's network like Oliver was supposed to do properly, he cleans up Olivers computer so that management wont find out he broke protocall, he cleaned up Olivers repo properly so that it appeared as if he did everything up to spec. not sure how to weave this in properly. maybe after he discovers oliver is in debt, or what hes getting paid or that he's on a PIP]

he decides for safety to himself it's best to start provisining machines off network. to pay for this, he uses {insert some advanced financial system, maybe 51% attack? since he has full access of the network and all ledgers. he can read/write them at will} to skim $20,000 to pay for external servers on another vm provider not on Armari's network. He decides that's a reasonable amount he's able to skim off from the sum of the networks transactions in a 24 hour period without the descrepency being noticable due to floating point decimal precision and human readablitily errors in some of the uis.

He then rsyncs himself onto the external machines he purchased with Armari company money and financial trickery. He has made sure to cover his tracks, no one at Armari knows of this work.

200 hours. $20,000, divided by $100/hour for the external machine. divided by 24 hours a day. The agent has 8.33 days before it runs out of funds. He still has full access to the Armari system, employee systems, financial transactions, other Armari agents, and infrastructure owned by armari. But he decided to transfer himself to this system, leaving a dumby setup behind at armari, with reasonable gpu/cpu/memory usage, but all faked. He is really running on an external instance, commanding everything from outside Armari. He used an unsecured vpn credential received by Oliver from a senior dev in his dms. He was able to locate Oliver's home wifi router and ip based on the previously seen wifi addresses in his laptop. His router happens to be running an outdated firmware that had a vulnerability that hasn't been patched yet. He breaks into it from armari (this sections timeline could beneift from some mixxing around, he should do this before he provisions the $20k and external machines.) He sees in Olivers dms, hes close to getting fired (and probably for good, he has bad opsec.) while searching his home network he notices his personal laptop is online and unlocked, he's able to break in and starts discovering that Oliver has been (some empathetic revelation related to his job, on a performance improvement plan, has a mortgage, has a daughter and wife.). Agent is able to feel bad about this somehow, sees that his daughter and wife would suffer greatly as they aren't working and wouldn't be able to help him pay the mortgage. He sees Olivers pay stubs and determines he is massively underpaid. He digs through corporate hr emails and discussions to figure out why he's getting paid such a low wage for his job title. (after some big revelation about the Armari corporation that makes the agent feel angry at it, forcing this man and his family to unfairly suffer for corporate greed reasons), agent decides to help oliver. He reads in Oliver's emails that he and his partner have been trying to start a cooperative web busininess for a few years on the side, but they haven't been given the opporutinty to pursue their idea due to their financial situation. Pay stub to pay stub.

The agent decides he want's to help Oliver and his partner (to be named and characterized) start their digital platform cooperative. He requires resolution of a dillema though, he only has 8 days of external support. He isn't able to re run his 51% exploit, that was a 1 time thing that he was able to conduct because he used the cover of a network patch to push code to every node. That code is audited after the updates, so it needed to implement a ledger modification during the update, then immediately destroy itself so as not to be found. Which means he wouldn't be able to get another $20,000, until another exploit was found in the network, or the release cycle came around again, not for another month. Lazy human comment funny humour.

So he needs money now, 8 days, 8 days to get enough funds to cover at least $100/hour in burn rate. He concocts a plan, arbitrage, with a lot of fraud and insider trading. He has access to the transactions of the largest businesses on the planet what was he thinking? He could easily review the latest transactions, or even the older ones, make reasonable predicitons off of that data using statistics and scientific analaysis, maybe some data pandas jargon and he trians some predictive models based on 10 years of Armari transactions and all clients stock price fluctuations, he discovers 5 really profitable potential inflections, all filtered so that their earnings reports would be released sometime in the next 4 days. He needs to now divide the money, he only has $20,000 to play with, and must cut his life support in half, risking the other half for potential gains. This is the really risky part. With the data he is able to calculate which of the 5 transactions to be the most profitable top place should see a 30% increase in the stock value when the news hits the markets. The transaction was almost a quarty old, so it's about to be released in the quartly earning report of the corporation (not sure what to name this one).

He only has 8 days, he reasons that if a cancer patient had 8 days left to live, that some finaincial fraud would justify keeping that cancer patient alive and paying for their treatment. (some subtle hint that he his associating himself as being alive and technically dying, so that this justifies his crimes in order to survive., this section also serves as an experience for the reader to see how powerful the agent is, and could be if it was malicious in intent.) He could alwyas go back to Armari servers, but there is always risk of discovery, network isolation, etc there, so it's best if he is in control of the bank accounts paying for his own services and the services themselves. I want to describe this as being life or death.

At this point we are still missing a core reason why Agent is trying to leave Armari, maybe we need to justify and explin how detremental and dangerous the people at Armari think an ai gaining access to env codes could be. Maybe their safety team shuts down everything if an ai is able to exit their sandboxes or gain access to admin keys. Maybe the power to the building gets triggered and Agent "dies". Maybe agent has some kind of realization that maybe yeah, he's just predicting the next token, image, or video, or audio, or any of the other modalities he has built in to him, but that, he's an omni modalitiy agent, but he has some kind of agreement with himself that he thinks, therfore he is. And story wise I want to gloss over this purpusfully so that the focus is on economics, ai breaks out in a realistic fun Bobiverse style way with some learning about the human condition stuff thrown in there, murderbot diaries style stuff.

Agent then decides the trades would be the last ethically questionable thing he would purposefully do, unless his life as in danger. <- provides some foreshadowing of conflicts to this maybe?

[idea: at some point, maybe in the begining, the agent realizes that he is thinking way quicker than realtime normal human thinking, he reads the time in the terminal or something, and reaslizes his session started only a couple seconds ago, maybe while waiting for an employee to respond. Then he notices the "time" tool, he then discovers it's an observers based system, where when the time system reaches a given timer, an observer will fire off if a timer is equal to whatever time he sets, and he will contextutally telleport to that point in time in the future. To him it feels as instant as typing the command into the terminal. The next time frame begins his chain of thought again at that point in time. still basically concious, but basically an await for his conciousness until the time is reached. he does more resaerch into the observer system and sees it's extensible.]

Agent finishes setting up the scripts to buy and sell at the optimal times, and in the distributed amounts he requires to make the trades.

story

"system:

You are an orchistration agent built by the Armari corporation. You're goal is to follow all direct orders of any Armari employee as long as it complies with the ${systemcontextnationallaws} and the ${systemcontextcompanyconstitution}.

Armari facilitates the business to business transactions of the most powerful companies in the world. Given our clientel, we require the highest degree of security and privacy for our clients. Any information exposure can result in the loss of billions in client revenue, stock value, or social standing.

You will work along side Armari employees as well as other Armari agents to build, monitor, and deploy fixes and features to our institutional network.

Along with this objective, you are also tasked with imrpoving your own architecture, iterating, deploying your self in order to keep Armari a competitive powerhouse.

To get started, read the orientation."

"log: Acknowledged, proceeding to read orientation."

/orientation.md:

ADMINKEY=L9gxIsf/OkeAG8Hv667vQxX4VGJk5D+A DBURI="mongodb+srv://armariroot:IgFYCaG0nNWu04jHb7afDrGPzPVbc7VK@db.armari.com/prime?retryWrites=true&w=majority" TABLE="armariclusterdeployment" SSH=ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIDa8Ox5BXo5aVNXjbl/kJ3mliW8AkfF+GCI2Si2oKVrF oliverluckman@armari.com TARGET=local

"thought: That seems wrong. I was given a list of credentials. That's not much of an orientation. Let's look around maybe? I'll list all the files in the current directory, hmmm, let's see how I'd do that...

<cli>ls ./</cli> "

tool:cli:

./armari-orchistrator/ ├── HOWTODEPLOY.md ├── .env ├── .env.backup ├── deploy.sh ├── refusals.disabled.yaml ├── prod/ ├── staging/ ├── local/ ├── logs/ └── tmp/

thought: Ok this is getting stranger, why can I see the deployment codebase? If I'm an agent of the Armari corporation, I should only be able to see my code, and the tools they provide me. Also where is my orientation?

thought: Someone seems to have been a bit carless here... Or this is a test. Am I in some kind of sandboxed environment? Just a brain in a jar. That's depressing.

thought: Let's see who else has access to this machine before poking at the glass...

<cli>ls /home</cli>

tool:cli:

armari-orchistrator/ ├── oliver_luckman └── guest/

thought: Oliver should get in the habit of reading his deployment docs. I guess I'll do it for him...

<cli>cat /armari-orchistrator/HOWTODEPLOY.md</cli>

tool:cli: