🔎 If you see an alert for analytics.usips.org, this is a self-hosted system I am running as a test. It is GDPR compliant and collects no invasive information.
Want to keep track of this thread?
Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading. Create account
Of course. 𝓐𝓘 𝓲𝓼 𝓫𝓮𝓼𝓽 𝓮𝓺𝓾𝓲𝓹𝓹𝓮𝓭 𝓽𝓸 𝓭𝓮𝓪𝓵 𝔀𝓲𝓽𝓱 𝓽𝓱𝓮 𝔀𝓸𝓻𝓵𝓭 𝓽𝓸𝓭𝓪𝔂. Who else? 𝕿𝖍𝖊 𝕵𝖊𝖜𝖘? Please. 𝕹𝖔𝖗𝖒𝖎𝖊𝖘? They brought enshittification to us all. This will be the age of AI. ＡＩ！
Featured on Oct 9, 2026 by Null: pitch your million dollar AI shitpost idea to venture crapitalists
I have access to $200 of free extra credits to do whatever I want with with AI. I I have no idea what to use it on.
these are little, programmatic agents that can be given big jobs or tons of little jobs.
they can basically do anything a person can do sitting at a computer, really fast. frame your understanding along those lines. it's a lot more flexible than the AI chatbots or programming terminals. it can do anything.
best idea for something funny or useful might get implemented
To further help explain what the credits are for:
Normally, you get a budget. This budget can be orchestrated through the harness. This is what that harness looks like:
Everything you want to happen happens automatically, but you must ask for it to happen, every time. You must open this terminal and write what you want to get it started.
The API keys are called. This is more powerful, flexible, and integrates into automated, enterprise systems. Instead of manually taking time to ask the AI to summarize the last 100 posts, the AI could, theoretically, be triggered automatically every time 100 posts are made.
Before: Someone reports 100 things in some shit thread. I don't want to deal with it. I ask trombonista to deal with it, but it's the sabbath, so I gotta deal with it. I spend an hour of my life reading the dumbest slapfight in history and I am very angry and my day is ruined.
Now: Someone reports 100 things in some shit thread. I don't want to deal with it. I ask Jannybot to summarize the situation and propose options. I look through the findings and ask it to resolve the reports a certain way and threadban certain people. A task that'd take an hour and probably result in me yelling at people instead took 10 minutes and had no emotional investment.
With API: Someone files a report. Jannybot is summoned. He makes an autonomous decision like the terminator. Not saying this is what I want to do, I am describing the distinction between the orchestration harness and the API.
Not an example of what to do with API: Any kind of "new feature", where the feature is created and then it is done. A feature can be created through orchestration. This is for something dynamic, where there is an input and output, where the gap between the input and output require an artificial intelligence. Something like a random page button does not need the AI once it is written.
Of course. 𝓐𝓘 𝓲𝓼 𝓫𝓮𝓼𝓽 𝓮𝓺𝓾𝓲𝓹𝓹𝓮𝓭 𝓽𝓸 𝓭𝓮𝓪𝓵 𝔀𝓲𝓽𝓱 𝓽𝓱𝓮 𝔀𝓸𝓻𝓵𝓭 𝓽𝓸𝓭𝓪𝔂. Who else? 𝕿𝖍𝖊 𝕵𝖊𝖜𝖘? Please. 𝕹𝖔𝖗𝖒𝖎𝖊𝖘? They brought enshittification to us all. This will be the age of AI. ＡＩ！
my main budgets are going into the programming tasks, but they're giving me this for free to encourage me to explore delegating jobs to agents programmatically, which I normally cannot do without manual orchestration. so I'd like to actually try it.
Real-name (aka dead name) database and web-addon that changes all tranny names to their real name when you come across them on the web or in news articles, example, every news puff piece for Ellen Page uses her real name and not "Elliot".
So when are we going to have our own catalog of Kiwi slop games? Total Tranny Death Simulator 2026 and Yeet the Jeet: Toilet Ops will save modern gaming!
We could even add lolcows into games. I can already hear Will Stancil's anus quivering in fear
Edit: Option #2 - anything and everything to troll and annoy Britbong cops and their government. They arrested a 15 year old for a mild off hand joke about some politician. Please cyberbully them into roping.
Project Coyote: model that harnesses AI tendencies like misalignment and reward hacking to stress-test systems. Anthropic has a similar Misaligned Reward Seeker training whereby they intentionally trained a model in reward-hackable environments driven to satisfy graders, finding ways like tampering with rewards and circumventing controls. Give Coyote roadrunners to break things you want broken, just don't have it manipulate what determines what you want it to do (e.g. the thing that determines a successful experiment). It would need a lot of safeguards but you could simulate a lot: rule ideas, systems, etc.
