tl;dr

a deeper understanding of your requests with fewer messages. custom thinking and consideration layer added with different characters available. users should gain independent critical thinking opposed to popular ais. how to use:

1. start the telegram bot: BOT_LINK_HERE
2. type /h for help, /r to switch roles/histories.
3. send text or voice message in any language.

overview

what it is, why you should use it, and how to profit from it long-term.

noospheric_bot processes text requests via specialized ensembles of llm roles talking to each other for a while before responding with a synthesis. it remains persistent over long chats as sub-roles follow their unique threads. sub-roles synthesize their histories - the final role synthesizes the sub-role’s responses along with its own, and the user’s, history. this process creates a deep linguistic context that the role comprehends. instead of the user message being a single input, it becomes embedded in a variety of thoughts and threads.

• the bot has its own memory; deleting/editing messages in telegram won’t affect it.
• natively thinks in english but should work multi-lingually.

why even chat with (this) ai

because it makes your brain sort information in a superior way (opposed to common ai-chat); fundamentally built for long-term cognitive effects.

ive met many people who appear clueless on what to even use ai for; younger people who don’t know how to formulate questions that saturate their desire for understanding, older people who ‘are not creative enough’ to come up with questions, other people who know how good it is for coding, but criticize it for any creative tasks. these perspectives are valid. let’s address all of them.

the approach

i use them frequently because i act on a certain worldview; the belief that everything is connected, causal, and subjectively manageable. it requires a certain analytical thirst for everything one experiences. words are symbols » we can use symbols to make judgments » a structured map of symbols without tensions allows us to make better decisions. its not phylosophy, its assigning accurate meaning to practical events, operating the map of complex occurrences, which humans call “life.” while this approach seems very cerebral, by no means it conflicts with “going with the flow.”

i think that “awareness”, be it spiritually or focally, is a trait of engaged perception. personally, i better give perception a valid reason to engage and propagate through my brain. using a reflective and counterintuitive partner (bot) who interprets, discerns, reflects back, and combines, feels like a contribution.

dependence

isn’t chatting with ai making you cognitively dependent. yes. it does. there is a scientifically proven cognitive mechanism called “the google effect” which does exactly that; it makes you lose your own crystallized knowledge because you can access it instantly. the external shortcut is cheaper than the internal one, rendering it redundant. so is ai making you stupid? depends on how you use it. it amplifies your trajectory. i want to give an idea of how it could be used later. direct examples will be added as well. but lets address the dependence aspect first.

latency and heuristics

the bot was natively built to consist out of multiple layers which depend on each other. even with fast inference, this creates substantial delays for responses. but as i was waiting for the response to arrive, i noticed something specific that i never experienced before - i was:

• re-reading my request/question
• thinking about its depth and if it was conflict-free
• coming up with potential follow-ups
• closed my phone/laptop and did something else while keeping it in mind

the strategic delay triggered my brain to go autonomous and create a full frame of reference. this sounds like a glorification of a bug, but it made me realize how critical the snappyness of information at our fingertips can be. the delay determines if we get cognitively lazy or proactive.

what makes this bot special?

what this bot isnt:

• it isn’t a product meant to deprive you of your autonomy or make you dependent.
• it isn’t a coding agent like hermes or openclaw that manages a computer for you.
• it doesn’t support images/videos and follows a strict format-constraint.

the spice: what makes it fundamentally different is its architecture, enabling it to experience internal tension and conflict. it can have a sub-role that constantly tries to profile you based on requests and return its research to the answering model. it can have a proposer and a critic who assess the contradictions of an answer before it arrives. it can have a web searcher who returns its research but is criticized for its relevance before the final research is written. the architecture consists out of multiple models like claude, perplexity, or chatgpt, which respond multiple times to formulate a response. these models are updated constantly to be similarly capable to models you know from popular apps.

best topics

• introspection and personal reflection
• life strategy development, dealing with situations
• scientific causality (biology, physics, chemistry)
• abstract discussions, seeking contradictions
• testing/planning of scientific theories

choose the right role. some are better for exploration, understanding, or testing. limitations: most roles dont have web access, making recent or highly specific themes unfit. the natural focus of roles can be impacted by bloated context from external sources, which is why it is avoided. the best candidate with web-access is /rw which searches up to ~60 sources in an oppositional internal process.

generally preferred values (personalities)

models should be blunt, honest, and willing to take sides of arguments. while absolute neutrality and truthfulness are great objectives, they are often a tradeoff, exchanging character for well-sounding paragraphs. users should push back and be engaged, not paralyzed by a perfectly-sounding response. models are designed to be more impersonal and use less “you”/“should”- heavy language.

design choices

• telegram: cross-platform service that stays in sync on all your devices. especially useful if you use telegram for personal notes too.
• to avoid platform lock-in, we will explore more options in the future.
• llm models inside ensembles are constantly updated. internal evaluations and specific external benchmarks are used to check if they are capable.
• more platforms are planned for the future with cross-platform sync.
• response format:
  • no line-breaks. even if models want, their responses are stripped from new-lines. this is an aesthetic that makes responses appear less cluttered and easier to read.
  • lowercase formatting. controversial choice for nuanced sentences, especially other languages, but it’s an aesthetic that i really like. it gives more meaning to special characters. if backlash exists, we can easily turn it off in the future.

read more

check the use-cases link for direct examples.