# Full Text: Con-cat-enate: pilot overview > Extracted from `catpilot.pdf` --- ## Page 1 3 page pilot Attempting a cat  P1P2. Vision Oriented Hippocampus . tags object. tracks object Receives CYCs attach CYC + object produce SIMs solves and update - done within time limit Updates parameters every SIM solution P1P2 describes a Mapping Trio and a Watching Trio part of the Vision Processing (which comes after receiving vision input) Main goal of this part is to combine labeled and tracked data Onto CYC-related relevance (CYCs are behavioral drivers) We then try to make further parts ( Beacons & Bev CYC ) Which are reproductions of (emulations of) Social cortex & Epigenetic Cycles ## Page 2 Beacons are social cortex emulation  tags object  have them pulse  have a pulse table cat vs pulses + pulsing table  on how to response  or hedge  or prepare separate types thru activity and uncertainty Then it creates a node for hedging signals These signals are akin to / frequent barrage of possible hedges That we treat / alter our behavior on; Given our awareness of such objects Notable one include a object type which represent the Self we then Bev-cyc  Sequence movitation system (Which is core driver of the agent)  These are simultaneously the action-sequences  (following ROS) ## Page 3  Depicted via a chain  of Decomposable Nodes decomposed until muscle % solved per time limit and updates policy for best outc onto DEMO (demo 0.01 .doc) Attempt to make this - Cyc goes forwa - Cyc + modulato - Serviced by P1P - Themes are SIM - Smush those tw - Onto an action - For resolution w % tension a finitely decompos come rd; or P2 Nodes and Themes M space of P1P2 wo bar within time limit sing chain ## Page 4 Output looks like this | n n n n n n | a block of CYC divided onto 6 Those blocks of n are filled with the processing / update calls of above parts (vision nodes and themes and cyc resolution ) Then the modulator would decide for every | n n n n n n | block Which one goes to follow and which decomposition should happen We smush first; we learn as it gets graded afterwards And we optimize these simulations further later on Converting them onto Active Inference Formulas (they then are called quickcalc) Summary:  Get vision input  Label , tag , process  Combine with CYC  Becomes SIM  Learn SIM outcome  Update Yupp ## Page 5 HOW DO YOU EXPLAIN A WHOLE CAT IN A FEW PAGES ?? YOU USE BEV-CYC :D WAKE > HAPPY > SLEEP Decompose HAPPY* ( as an example~ ) ( it can branch onto many things! ) Fill Simulation w/ Vision Nodes Special mention to Beacons & Social Cortex - Finally We smush the nodes Both labeling & solving Per time limit ## Page 6 Abstract Hello, this paper is about Amateur attempt at Cat-making We’d like to create a cat ( a decomposable action sequence agent ) ( using ROS style action checkpoints ) Describe the Specs What is the scope of this cat? We are attempting a - Agent cat that - Starts with decomposable action-seq - Starts with vision logging for all objects - On to attaching each objects with BEACONS - On to processing each beacons and a reflexive system using Vision ( map trio ; watch trio ) - On to ending up in either a SIM Or a QUICKCALC solution / variable update - Quickcalcs are a way to simplify the simulation run onto a FE % calculation - Via active inference. ## Page 7 - Methods are currently handwritten, optimization still on work Contents: Beacons Bev Cyc Demo Summary P1P2 Short Recap Fin What is to be expected? Pilot will jump onto a BEACONS & BevCYC Representing a emulation of biological concept of “SocialCortex” & “EpigeneticDrive” (inherited & learned) We will move on to a DEMO 0.01 Then recap ## Page 8 Chapter 1 Step 0: social cortex that handles ] cat’s general policy vs each object  BEACONS.doc Pic: Pic1. Cat looks at objects P2. Looking at owner Objects are pulsing! Owner is super pulsing! Do we hedge? Says cat Do we kungfu? Thinks cat SocialCortex –esque emulation; thru attaching PULSES (+response table) How –nodes- of P1P2 gathers the tagged object And puts them onto a bucket . to make in ROS . all SIM ENTITY are graded by significance We then assign a pulse beacon type & reply with response … … ## Page 9  BEV-CYC.doc Pic 1. Pic 2. CYC (getting transcribed) CYC is described along with all the other processes in 3-3-3 action sequence a modulator is then deciding (ala ROS for this attempt) which is most important; and how this is decomposable does each update through learning to the smallest degree which doubles as a muscle % decision Starts with CYC pool; cyc modulator Cyc becomes chain Cyc chain gets decomposed Cyc gets processed thru SIM/quickcalc; Each results in a micro action recalls Updated via SIM run .perbit. - ## Page 10 BEV-CYC is Epigenetic-esque emulation; by describing a cat’s main drive (inherited to then learned) And the process on how it is decomposed These are all described and attempted in ROS / Robotics term With each of these BEV-CYCs As a chain of nodes that each node can be decomposed All of them are first inherited (hardcoded from birth) And all of them are then learned (edited as the cat goes thru the age) modulate. execute (%)(%) . Update (Learn) modulate again. Ex. Update (again) Decomposable action sequence chain Gets handled by an LLM/agent to refine By which and how to update Sequences are made on form of state-state-state ## Page 11 Hoping to finally break it down to The smallest nodes which simultaneously represent Muscle % tension (is already an action) . For example: a yowling cat (social) BOGUS(uncertainsoc) - ACT – OBSERVE (social) BOGUS- YOWL-OBSERVE (social) BOGUS-YOWL (BOGUS-MUSCLE-OBSERVE) - OBSERVE The above chain is a hypothetical social cortex related action Which is a BEV CYC that exist as a snippet for when Cat spots a presence of engage-able (uncertainsocial) (these conditions are later set – via a prior related Agent Node / processing node) When it decides to YOWL & OBSERVE YOWL is then decomposed onto smaller and smaller nodes YOWL in itself is not the smallest; YOWL likely (just like in robotics) ## Page 12 Include steps such as : adjusting ; engaging ; provoking ; observing ; YOWL Y- YOWLING – YOWLX YOWLY – ( YOWL YOWL YOWL* ) - YOWLX Construction of these tree-of hardcoded behavior are Initially handwritten; with them being associated to each its own SIM type (type of Game simulation to resolve) In this game sim; there can be learning where the decisions matter But overall; for decomposition it always first tries the same thing Upon failure; we only then take the failure signal (these are considered Chem in Chempool) Lets label this “molecule to insist on learning if possible” Lessay to be called “PON-###” (ponder) For a deliberate learning behavior As with the Chempool description on P1P2 And how the practice of converting highest CYC onto a Module Such module will then (to publish) Have its own Chempool for watching this PON-### PON-### will be passed to another module ## Page 13 (currently we are using ModStream ModStream records this as a (one of many) queued This queued THEME comes from the PON-### origin So thus it gets resolved also the same way 1. Run the modulator for CYC-search 2. CYC-search hits with the latest Cluster of Info 3. (should be a cluster) 4. modStream publishes a latch (modLatch) 5. (note Latch = nested SIM (forced 3 roll) ) 6. modLatch maintains this Latch imagine modlatch as a Node that assigns an Agent for each Stream / Latch (in other words; a concurrent call for Simulations) 7. giving it satisfactory % - (floating satisfactory % to judge a latch ending; ) 8. every Latch (SIM) is processed with TIME LIMIT 9. the agent has a BUDGETING TABLE 10. for how long this activity should usually take 11. (imagine a depleting [100%] getting depleted by % % from factors / inputs 12. That it accounts to deplete faster ## Page 14 Anyhow. This is the part where we mention | n n n n n n | Where each n is forcibly filled with both a vision-node update; and a THEME update; an action- and a next-cyc prescription; Thus we are just filling each n with processes that have to happen (or be queued) And the subsequent judgement of each node Represent the agent’s action of choice Usually this happens In terms of YOWLING By this way From before YOWL Y- YOWLING – YOWLX YOWLY – ( YOWL YOWL YOWL* ) - YOWLX Imagine each of those Decomposed YOWL Are a | n n n n n n | node of CYC Where at this lowest decomposition; Its just “more muscle tension” or “less muscle tension” Determined by the | n | that we fill for each ## Page 15 (what the cat actively sees and thinks) ( an n for vision update ( an n for theme solution ( an n for cyc modulator ( an n for checking* (mentioned in DEMO doc)) Or what have you These are scaffolds that are called by the CYC-MODULATOR It calls them in form of | n n n n n n | With what each n contains And its cached | n | resolution table We are not good at automizing Or making a good math formula For consistency! But we are pretty certain we are able to make Full behavior table for any video / activity that is requested! (Sir/Ma’am can test by giving a link for me to decompose at spot ) (Even if wonky; it will probably only take one or two polish to make it right) Anyhow back to BEV CYC ## Page 16 Summary; is that we have decomposed nodes ROS style with action checkpoints (for completion or failure) That is : 1. Forcibly resolved per time limit 2. Each resolution results in learning (spontaneous) 3. Even if a PON-### chemical is triggered and logged 4. such will be dealt separately (a queued chem.) 5. depending on the identity’s breathe & habit (which is also a form of CYC with its unique completion Tree) For YOWL Will then lead to further decomposition (YOWL -> YOWLING muscle %) Check method at BEVCYC . doc For current plans on checking outcome (step-check-step) And how this chain runs separately from AC (the main executable) ## Page 17 There’s also a lot of Or re-refering to ex We are hoping to d (show video; comp DEMO 0.01.doc Almost a recap; for this DEM Hoping with a help of external ROS + VisionModel For this attempt; We aim to create this (1) 1. CYC is there; w 2. (small t) 3. © is a contain 4. Object types / e f Copying (Entire CYCs) xisting & live CYCs that will be involved describe these manually in site ose; retry;) MO ith its modulator ner for etc , from vision ## Page 18 (2) 1. Run visionNodes (P1P2) 2. Starts logging things goes onto © 3. CYC runs along; plans; drives 4. visionNodes react; modLook & modWatch 5. starts focusing on things that matter for such 6. by “starts focusing” (3) 1. Is already a CYC-node 2. And it already becomes a SIM 3. > Because all of these processes are time-limited 4. Each of them will always gets resolved per Breath(n) 5. Time for each breath period (4) 1. As described in P1P2 2. Each SIM resolution 3. Always end with parameters update for both the CYCnode 4. And the objecttype 5. After it happens several times (rule yet to be defined) 6. Body attaches a NOVELTY % to that SIM type 7. Below novelty% becomes a Quickcalc 8. Which is a optimization method for us 9. To simply convert the gameworld SIM onto a Formula 10. With FE %-% (range) 11. As a defining property 12. Then nested along with its associated 13. Cyc nodes (this is the triangle alongside the Containers © ) ## Page 19 (all of them are just update-ables working data) (5) 1. In terms of agent running 2. In a Gymnasium 3. This starts with the Agent having a prior Run (prep) of the Territory 4. If it hasn’t done that before; it will prep one quickly 5. (with the Templates of known environment) 6. Same with objects 7. Refer to P1P2 for the agent arrival for these 8. It will then quickly call the associated objecttypes 9. Of that particular environment 10. Whilst the VisionTrio & ScanningTrio 11. Runs per always; always resolving to SIM Early Summary:  Guy started on P1P2 making the vision nodes..  We then suddenly jump onto the more fundamental  ... (on which the reactivity & processing of these nodes will apply) then we go onto the more basic (epigenetic & social cortex)  we Beacons  A protocol for reaction vs entity in the Simulation  we Bev-Cyc  A decomposable chain from birth to update as cat learns Currently depicted using ROS-style Action-Checkpoint-Sequence  We Demo 0.01 (we smush n run |n n n n n n |) ## Page 20 Onto.. Demo & P1P2 up Some mention on P1P2 Final Summary ## Page 21 Cat Gist So what is the project like when proposed? Reviewing the lingo; We aim to; make CYC run as a from-birth habit (wakesup; tries to) aim to; gather n resolve along vision nodes (P1P2) we then smush for ordering and simply run them per each time-limit Cyc goes forward Vision follows after Themes Come Gets Smushed Runs per bit-by-bit Updates bit-by-bit For longer learning; 1. Kitten has engaged in a high-stress 2. CYC updating session 3. For many days.. (with many PON-###) ## Page 22 (we aren’t clear on this yet, refer P1P2 for the prep) 4. Kitten & Cat both has a self-referential 5. (re-bagging everything & process in mind) 6. Process; mentions briefly in P1P2 7. Mostly this will happen by Copying 8. The AC and the STATES 9. (whole described Demo0.01) 10. And having a separate (long)THEME 11. Insist upon these lingering 12. THEMEs to resolve Cat. - is a cat Crabs. -agent/llm like modulator of input & output Cutter mentioned in AC P1P2 is this CYCs - sequence from inherited to learned Crabs. - more cutting! (modulating CYC) Vision. -ScanningTrio (scans env.calls.watches) -VisionTrio (applies processes to objects) Container. Logs the entries ObjTypes; EnvirTypes. etc AC. Becomes the smushed thing above (demo) Theme also known as SIMULATION (nested) ; from (P1P2) Smushed whence any input is time-resolved onto one Updates Updates every SIM / quickcalc resolution Triangles Optimization onto FE range ## Page 23 x ## Page 24 P1P2 recap So whats there? P1P2 Pretty much just starts with Giving a High Overview Of the Modules & Vision Nodes Back then this was made to represent “Vision-Oriented Hippocampus” Hmm.. I guess this was correct (its more like Sensory-collection-and-combiner-with-CYCs) Anyway. We describe P1P2 With some terminology - Screens - Chempool - Modules - Sims - Themes - Latches We describe them now! (sry just minor gist) All of those is important for Robotics purposes (porting this onto ROS) ## Page 25 Screens Are transposed Vision input; We recognize that this process is actually also a CYC -> SIM -> UP -> CYC But for this overly frequent components (much more frequent than 6 nodes per sec) (pretty much constant) We simply denote a Screen Which is a transposed Vision Input Its like a constant-lingering-subconscious awareness depiction Sample of notable - Combiner of each eye’s vision input - Combiner of vision+other senses to default presence space - Tranpose of that vision space onto different scales - Automatic having 3/4/5 (arb number) of such scales aaa - For Bat, lessay they’ll have one for the whole cave - For Salmon; lessay they’ll have one for direction to their spawning pool - For Whale; lessay they’ll have one for a whole ocean - Most these are hardcoded by biology - (its actually a CYC but its so lingering that it makes a structure in the brain) - Anyhow; we simply denote important Screens Anyhow back to the notable - A 3d room space awareness (Transpose vision onto Overhead) ## Page 26 - An Overhead view - A Screen for ENGAGED-Social-Object - Refer to BEACONS - For socialobjects - Cat will assign a LiveScreen representing - How the cat thinks this creature can move - (this can be done using a THEME / SIM / CYC / AC-Copying ) - But after realizing for the size / scale of the difficulty - We realize its better to just - Hardcode a screen for some crude automated - Representation of known moves (consider any SocialObject-types to always come with a Screen) Representing emphatized – condition (programming wise; we then simply hardcode what this SIM will be given a Focus%) (this is much easier! Than emulating all the steps on which this empathy will increase Or describing how an empathized object (any object can be felt this way) Grew from a ObjecType to a modTude to a test To another sampling to then a AC copy for resolving Queries on How dangerous / how unpredictable / source of such We can actually do the above steps manually! Simply describe how a cat is trained to follow the strongest signal (CYC-sample) - (almost always persistent every XX second) Which is a person waggling his hand / finger weirdly ## Page 27 Then we can draw how a cat kept getting his CYC-(sampling) to always have to include An EMPHATIZED human in his THEMES / Simulations Then fill those EMPATHIZED Themes With known information of the person Then confirm these cues bit by bit Given how the person’s subsequent movement does IF we wish to avoid this! We simply draw the outline of the Cat’s Pondering (just a PON-### for the social Object a blot/censor for deferring to this Screen < a input-### for taking from that Screen! < (on the deciding THEMEs / cyc-modulator tree) (on deciding the | n n n n n n | fills ) so we could cut out most of the manual-empathy-checking for person to person clue-confirmations to simply a hardcoded Screen; with known typical fillable information ## Page 28 Chempool Chempol is an attached Display to a module To track the input and timing Of Chemical Signals / Known Molecules / Hormones That dictate a certain rise / fall of certain calculations Usually this is done when describing the Module We describe a Module Then we describe what Screen & what Chempool Does it carry with it There’s probably some sort of advanced combination Pool That we have to make, to port for biology But for now! This simple tally will do! For stresss / satiety / what have you ## Page 29 Modules Modules are independent nodes With a lot of independent processes Attached to them; run entirely separately Guy used to think to program a cat! He needs a help with + 36 pc ( >_< ) (such backwater counting standards) These pcs! Will have its own Process watchers To Constantly process inputs And talk to each other These modules are defined / enshrined by their CRITICAL OUTPUT As in; these are the grades of processes On which if its missing; cat is considered highly crippled Or just fail to function Modules are based on Hidden Organs that may be (hardcoded critical structure) Hidden in the brain / suchlike ## Page 30 One cheating (or not cheating! Just not quite defined yet) Module ; is when we make a rule where The HIGHEST running CYC (or several of them) Should have its own MODULE (instead of having a THEME call constantly going on) (simply have a Module to produce its critical output for the Current activity) I.e. When in forest; Roaming; becomes a module; modROAM with a attached Chempool + Screen Seems weird to have a special rule like this But perhaps all of these nested structures are just Bigger / more attached form of Molecule; And we are simply drawing some scales / sequential processing As much as we need to produce a consistent agent BTW. Ontop of that ONE MODULE Where its actually a graduated THEME ## Page 31 We have: - ScanningTrio (sometimes referred as MappingTrio) - VisionTrio (sometimes referred as Watching/TrackingTrio) Scanning trio are three boxes that deal with environmental data processing Vision trio are three modules responsible for object data processing One scans; one routes; one maps One watches; one looks; one tudes (tests) Refer to P1P2 for each Of their individual operations And what they require and produce To the Container SIM/THEME We also have a special two combo of Modules modStream and modLatch one deals with THEME producing (from CYC to theme) / (trigger->cyc->theme) one deals with LATCH maintaining (from theme(latch1) -> subsequent latches -> resolve) They deal with prescribing the appropriate GAME SIMULATION Their contents; their abbreviation (by novelty%) and their update ## Page 32 AC / BM1 / BM2 P1P2 also mention an AC Which is the same structure of SMUSHED |n n n n n n | Tracker; Getting its queued CYC|CYC|CYC To be accomplished Meanwhile BM1 & BM2 Is a separate, independent Module of the AC That is supposed to be dealing with: 1. Checking for predicted outcome 2. Checking for limits (re-checking “is this what should’ve happened?”) and (budgeting “how much can I handle before I freak out?”) > Some of the pre structure for this Is described in DEMO 0.01 doc As a next-in-line update ## Page 33 - Refer to docs: -DEMO 0.01 for project doc and planned update -Beacons.doc social cortex emulation -BevCyc.doc epigenesdriver emulation -P1P2.doc small update --- *Extraction method: pymupdf*