Week of 26 Oct 2026¶

Dates are approximate.

What happened¶

Discussion 1 (27 Oct 2026)

The discussion covers the architecture, philosophical principles, and practical rollout of the Machine World AI-native system alongside its sibling initiatives, Inner Realm and Marga Sakaccha. The primary speaker outlines Machine World as a multi-model, bilingual AI framework capable of coordinating digital agents, physical systems via APIs, and human collaborators to automate real-world processes. The core design is rooted in Buddhist Abhidhamma and [doctrine term] principles, incorporating a Vinaya framework to eliminate defilements (raga, dosa, moha) and foster bodily and mental seclusion (kaya-viveka, citta-viveka) for meditation. The team discusses onboarding using a NotebookLM knowledge base, Claude CLI scaffolding, Model Context Protocol (MCP), and multi-modal models to create autonomous skills. Practical constraints like subscription costs, API access, and alpha software bug management are also reviewed.

Discussion 2 (27 Oct 2026)

The team discusses video generation tools and automation workflows for content creation, comparing commercial models and open-source alternatives. They analyze video generation platforms such as Seed Dance (Seedance), Wan 2.5, LTX Studio / LTX Video, Luma Dream Machine, and Kling, discussing costs, capabilities, API access, and open weights. A Telegram bot system set up by Miyuru is reviewed for [name withheld]'s workflow to automate generating reel videos from audio files, timestamps, and images. They explore automating audio downloads from YouTube links, trimming audio by timestamps, and scheduled publishing via cron jobs. They also discuss using fal.ai as an API router to avoid dedicated hosting expenses for GPU instances like H100s. Finally, upscaling solutions such as Topaz at 60 fps are mentioned.

Discussion 3 (27 Oct 2026)

The discussion focuses on architecting an interactive Dhamma visualization and query system based on digitized canonical Buddhist texts. Speaker 1 advises Speaker 2 not to instruct the AI with rigid visual rendering specifications, but rather to prompt it based on intended user interaction and behavioral flow. The system must accurately depict cittas (mental states) and rupas (matter), alongside paticcasamuppada and [doctrine term] causal relations. Addressing concerns about AI-generated Dhamma examples contradicting the canon, Speaker 1 emphasizes relying strictly on canonical digitized Tripitaka sources rather than subjective personal interpretations. Examples generated through an internal validation loop must remain canonical, without human bias distorting the Dhamma. External validation by expert scholar monks is planned once a prototype is functional. Speaker 2 agrees to build an initial prototype and share the design link for internal testing and review.

Discussion 4 (27 Oct 2026)

The speakers discuss the design and visualization of a digital Dhamma study and visualization system. They explore how to structure Citta (mind/consciousness factors) and Cetasika (mental factors) dynamically, allowing users to visually navigate and interact with their connections. The project aims to generate examples of mental states that align cleanly with the Four Noble Truths, Dependent Origination (Paticcasamuppada), and the [doctrine term] Pakarana. Rather than having AI modify or edit canonical Dhamma content, the system will serve as an interface to query, link, and expose insights from canonical sources like the Tipitaka. Canonical accuracy must be strictly maintained to avoid subjective interpretation ('mamathwaya'). Validation with knowledgeable Buddhist monks is planned to verify canonical correctness. Finally, Speaker 1 instructs Speaker 2 to build an initial operational interactive visualization prototype for testing and further iterative refinement.

Discussion 5 (27 Oct 2026)

In this discussion, Speaker 1 guides Speaker 2 on architecting and developing the test harness and integration layer for the Machine World system. The architecture focuses on connecting to diverse external enterprise systems—such as retail inventory, POS, and HR—via APIs and Model Context Protocol (MCP) servers, using synthetic sample data to generate operational insights. Speaker 1 emphasizes that AI automation should relieve human operational toil, fostering physical and mental seclusion (kāya and citta viveka) so individuals can direct their focus toward Dhamma and understanding the oscillations of nāma-rūpa. The technical workflow specifies multi-tier feedback loops, including a three-time loopback engineering mechanism with separate validator models and a '/goal' spec alignment structure. The conversation addresses multimodal interface evolution, from traditional displays to Leap Motion 3D gesture controls and dynamic world models. Speaker 2 raises challenges regarding Dhamma text analysis and token limits with models like Opus, prompting a discussion on Claude Sonnet, large models like 'Fable 5' (noted for high parameter counts and costs), and the need for local offline embedded models. Finally, they explore packaging a lightweight offline model that can run locally on edge devices or mobile phones, while syncing with Machine World when online.

Discussion 6 (27 Oct 2026)

The team opens by reviewing documentation issues for the Machine World system, specifically broken markdown links that redirect to the home page, and discusses simplifying the system conceptualization from complex CLI commands to accessible mobile app skills and household automation function integrations. They evaluate auto-publishing workflows to Facebook and YouTube using developer API keys and discuss AI development tools including Claude Code, CodeX, OpenRouter, and local open-source models to mitigate high monthly subscription costs. In the Dhamma documentation segment, Miyuru outlines errors found in example texts regarding the 89 cittas and diṭṭhi/bhava concepts from Ven. Rerukane Chandawimala Thero's books, proposing a multi-step review process with [name withheld] and Ven. [name withheld] Thero. [name withheld] Aiya has offered to type up the corrected content once reviewed. Finally, the group discusses frontend rendering bugs in Machine World where text animations load before custom fonts, and closes with a discussion on AI search crawlers and modern supply-chain cyber threats including AI worms and token theft.

Discussion 7 (27 Oct 2026)

The discussion focuses on defining the development strategy and architecture for the Machine World AI system. Speaker 1 guides Speaker 2 (Niranjan) on creating a simulated test harness using synthetic sample data and connecting it to business APIs via MCP and REST endpoints before live deployment. The conversation details an agentic workflow architecture utilizing a three-tier execution loop that refines context, collects memory files, and continuously evaluates goal alignment. A portion of the discussion connects operational automation to reducing human stress, freeing mental space for Dhamma practice. The speakers also address progress on the sibling Tipitaka rendering project, troubleshooting how Claude interfaces with Buddhist text structures. Finally, they evaluate options for LLMs, weighing high-cost models like Sonnet, Opus, and 'Table 5' against embedding offline local models such as Gemma to reduce token expenditure and manage infrastructure dependencies.

Discussion 8 (29 Oct 2026)

In this discussion, the project lead outlines the vision, technical architecture, and rollout strategy for Machine World, emphasizing conditional computation and causal loops rooted in Dhamma principles (specifically the [doctrine term] conditionality model and a cessation engine). The system is designed to provide persistent, synchronized shared memory across diverse models, allowing users to control their own context and collaborate fluidly via agents without vendor lock-in. A core focus is helping the team transition from text prompting to voice-to-voice interfaces and building out structured demos and peer-reviewed educational content. The lead stresses framing the system around solving real business problems rather than selling software tools, illustrating this with inventory management and physical processes. Broader socioeconomic disruptions—including impending massive job displacement caused by multimodal models and robotics capable of physical tasks—are cited as the driving reason to build self-sustaining, autonomous economic loops. Concrete roadmap targets are established, including shipping a pre-alpha shared memory environment and demo by late November. Additionally, the platform integrates spiritual practice spaces, such as Marga Sakaccha long meditation retreats, to support mental and physical ease (kāya-viveka, citta-viveka).

Discussion 9 (29 Oct 2026)

The speaker introduces the core architecture and vision of Machine World to a younger colleague ('malli'). Machine World is designed around a unified, shared memory substrate that connects multi-agent, multi-system, and multi-human collaborators. Instead of static software development, systems are grown iteratively on top of existing intelligence via inner and outer loops and an RSI (recursive self-improvement) engine. Dedicated, isolated hardware instances are accessed via unique URLs through the Model Context Protocol (MCP), charging users only for active GPU intelligence compute. The project incorporates concepts of cause and effect (causality) and a cessation engine to eliminate stale data and avoid hallucinations. A pre-alpha release featuring interactive demonstrations (such as a reflection engine demo) is being prepared alongside academic research papers and weekly content publications. The speaker is currently working full-time on this project and discusses strategies regarding funding, whether to bootstrap with existing savings or pursue venture capital.

Key points¶

Decided · The release of the mother-child song will not be rushed for Mother's Day on the 10th if quality requires more time, and will be accompanied by an explanatory video.
Decided · The team decided against hosting dedicated GPU instances continuously because paying for idle time is uneconomical compared to calling APIs on demand via fal.ai.
Decided · Public content and technical papers from the project must be internally peer-reviewed and vetted by external domain experts prior to release.
Said to exist · Machine World runs across a multi-model aggregate rather than a single large model, using modular specialized mini-models.
Said to exist · Machine World is natively bilingual, supporting both Sinhala and English seamlessly.
Said to exist · The system design and ethics framework are directly governed by Abhidhamma and [doctrine term] Pakarana to prevent cultivating raga, dosa, and moha.
Said to exist · Inner Realm content is hosted and published via a dedicated YouTube channel and SoundCloud link.
Said to exist · Current generative Dhamma query retrieval accuracy operates at approximately 92% due to non-deterministic constraints.

Ideas and open questions¶

Idea · [name withheld] offers to help review and structure content points using screenshots once personal land deed work is completed.
Decided · A $35 subscription cost per user is too expensive, prompting the team to stick to native $20 tiers or local/OpenRouter setups.
Decided · Logging into the Chinese version of Seed Dance requires a Douyin account and SMS verification code, which fails on non-Chinese phone numbers.
Idea · Add a feature to download audio directly from a YouTube link, trim it between start and end timestamps locally, and automatically create the reel.
Idea · Prompt the AI focusing on user interaction and functional requirements rather than describing visual shapes or UI layout details.
Open · Speaker 2 raises a concern that AI-generated everyday examples of citta might contradict or misalign with the Tripitaka canon.

All details (42 topics, 86 more points)

Machine World Architecture and Vinaya Framework: Speaker 1 explains Machine World as a multi-model AI-native environment integrating physical agents, digital agents, and humans. The core framework embeds Abhidhamma and [doctrine term] principles to prevent training defilements (raga, dosa, moha) and support meditation by automating worldly chores to afford kaya-viveka and citta-viveka.
CLI Scaffolding, NotebookLM, and Onboarding: The team discusses how to interact with the system using CLI interfaces, Anthropic-based scaffolding, and NotebookLM knowledge bases. Speaker 1 notes that NotebookLM needs system prompt adjustments from discovery mode to concept explanation so collaborators can learn and formulate automation skills.
Dhamma Website Finalization before Vesak: The team reviews progress on an upcoming website for Dhamma teachings and aims to finalize and review content before the 30th (Vesak). [name withheld] plans to assist with content alignment after finishing personal deed-related tasks.
Marga Sakaccha and Inner Realm Integration: Speakers discuss using Machine World skills to automate video and audio pipelines for Marga Sakaccha podcasts and Inner Realm music tracks. They analyze how to represent Buddhist themes—such as non-self (anatta), mother-child attachment, and nama-rupa dissolution—visually and musically without falling into extremes of annihilationism or eternalism.
Video Generation Platforms and Seed Dance Evaluation: The speakers compare video generation platforms, including Seed Dance (Seedance 1.4, 1.5, 2.0), Douyin account login requirements, Chinese phone number restrictions, and pricing models. They discuss structuring prompts as captured reusable skills for various styles and scene shots.
Telegram Bot Video Automation Pipeline and Dhamma Content Integration: Miyuru demonstrates and explains a Telegram bot automation pipeline where an audio file and timestamped images are submitted to automatically produce a reel. The team discusses extending this to download and slice YouTube audio clips directly, and notes how this automation will streamline illustrating citta vīthi (thought processes) discussed in Marga Sakaccha.
Open-Source Models, fal.ai API Routing, and GPU Cost Analysis: The discussion covers self-hosting vs API routing for video generation models such as Wan 2.5, LTX Video, Luma, and Kling. They examine using fal.ai instead of paying continuous hosting costs for an H100 instance, deciding pay-per-call API endpoints are much more cost-effective.
Character Consistency Training and Video Upscaling: They also briefly touch upon upscaling using Topaz to achieve 60 fps output.
Interactive Visualization of Citta and Rupa: Speaker 1 instructs how to prompt AI to design interactive visualizations for mental states (citta) and material phenomena (rupa). Instead of specifying visual layout details like shapes or circles, the prompt should describe how the user interacts with, navigates, and investigates mental states and causal conditions.
Canonical Validation of Generated Dhamma Examples: The speakers discuss the risk of AI generating inaccurate Dhamma examples, such as wholesome thoughts arising on the road. Speaker 1 explains that generated examples must align with the Four Noble Truths, Paticcasamuppada, and [doctrine term] relations, checked through a canonical validation loop.
Digitized Tripitaka Integration and Scholar Monk Review: The system utilizes accurate digitized Tripitaka and commentary sources to surface insights without altering canonical content. Speaker 1 notes that external validation by scholarly monks will be carried out after developing an initial interactive design prototype.
Visualization and Interactive System for Citta and Cetasika: Speaker 1 outlines a requirement to visually navigate Citta (mental states) and Cetasika (mental factors) and their interconnections. The system should allow users to visually inspect and engage with how practices and mental states work. The goal is an interactive system that allows the user to engage behaviorally and visually.
Example Alignment with Core Dhamma Doctrines: The system's generated examples must strictly align with the Four Noble Truths, Paticcasamuppada (Dependent Origination), and the [doctrine term] Pakarana. If an example fits all three frameworks, its meaning is validated without altering the underlying raw data.
Canonical Validation and Tipitaka Digitized Sources: To avoid subjective or personal interpretations, the system must link with digitized canonical texts (Tipitaka and related commentary sources). The speakers plan to validate the canonical accuracy by consulting expert monks.
Prototype Development and Next Steps: Speaker 1 instructs Speaker 2 to create a functional prototype allowing navigation and visualization. Once a working design link is ready, both speakers will test and iteratively refine it.
Machine World Business Integration and Test Harness Architecture: Speaker 1 explains how Machine World integrates with external business systems such as retail POS and supply chain inventories. He advises building a test harness using synthetic sample data and MCP servers to model production flows and predict supply shortages before deploying live.
Automation Purpose: Freeing the Mind for Dhamma and Nāma-Rūpa Contemplation: The ultimate objective of reducing business operational stress through AI is outlined: providing physical and mental seclusion (kāya viveka and citta viveka). This enables human operators to observe the continuous oscillation of nāma and rūpa without being overwhelmed by daily operational burdens.
Multi-Loop Agentic Engineering and Validation: Speaker 1 details an agentic loop design utilizing three-tier loopbacks, memory artifact storage, and separate evaluator models to test outputs against original specifications. He highlights engineering iterative loops guided by goal commands to produce high-confidence agent execution.
Interface Evolution: Leap Motion, World Models, and Dynamic Generation: The discussion covers modern interfaces moving beyond legacy browsers into dynamic world models and real-time generation. Speaker 1 cites 3D spatial controllers like Leap Motion to physically manipulate digital software environments.
Abhidhamma Analysis and Digital Text Parsing Challenges: Speaker 2 reports difficulties with models parsing Sinhala Abhidhamma concepts like citta kṛtya and kusala/akusala states from digital notes. Speaker 1 suggests developing a dynamic interactive interface rather than relying on legacy waterfall-style document parsing.
LLM Selection: Cloud Providers, Fable 5, and Local Embedded Models: The speakers compare token costs, latency, and context sizes across Claude Opus, Sonnet, and huge architectures like 'Fable 5'. They explore deploying lightweight offline models running directly on mobile phones or edge devices for Sri Lankan localized use cases.
Machine World Concept Simplification & Auto-Publishing Pipelines: The team discusses fixing broken markdown link redirects in Machine World documentation and simplifying the concept into modular skills rather than purely CLI commands. They also review progress on automated video upload pipelines for Facebook and YouTube using OpenRouter and developer APIs.
AI Development Tooling & Cost Management: The speakers compare AI coding assistants and models including Claude Code, CodeX, OpenRouter keys, and Claude Team packages versus open-source hosted models. They analyze monthly and yearly tier costs and rate limits to determine a sustainable setup.
Marga Sakaccha Dhamma Documentation & Review Strategy: Miyuru highlights textual errors in Chitta-vithi and citta classification documentation derived from Ven. Rerukane Chandawimala Thero's texts. The team agrees on a structured review path consulting [name withheld] and Ven. [name withheld] Thero before publishing final content.
Machine World Frontend Bugs & AI Cyber Security Risks: Gehan and Miyuru analyze a frontend text animation bug in Machine World caused by font loading delays. The discussion concludes with insights on AI native search crawlers and supply chain cyber attacks targeting developer environments.
Machine World Test Harness & Business API Integration: Speaker 1 explains that before connecting directly to live company systems, a test harness with synthetically generated sample data must be built. The system should interface with business tools (POS, ERP, HR, finance) via APIs or MCP to generate dynamic real-time operational insights and automate business flows.
Multi-Agent Loop Architecture & Spec Engineering: The discussion covers engineering native multi-agent loops to automate problem solving. The system uses a primary loop for step execution and decision points, a secondary loop for memory fine-tuning and re-running, and a third loop for goal validation using an independent evaluation model.
Dhamma Application & Tipitaka UI Rendering: The speakers review the Marga Sakaccha Dhamma study tool. They address issues with Claude reading and rendering complex mental state data (akusala citta) from Tipitaka repositories and discuss building a live digital interactive UI rather than taking a waterfall requirement approach.
LLM Selection, Offline Models, and Token Cost Optimization: The speakers analyze various LLM options including Sonnet, Opus, and 'Table 5' (noting its immense 10-trillion parameter size and blade server server requirements). They decide to explore hard-binding and embedding offline local models like Gemma directly into the Machine World core application to eliminate high token costs.
Voice Integration and Model Independence: Speaker 1 instructs the team to stop typing prompts and adopt voice-to-voice interaction using Claude or Codex integrations. The core value of Machine World is preserving and synchronizing personal memory state across any standard or open-source model provider.
Content Creation and Problem Space Grounding: Speaker 1 cautions against introducing Machine World directly without first detailing the underlying problem space, test methodologies, and research findings. Demos and narrative case studies (such as physical inventory and supply chain disruptions) must show how Machine World handles real-world business needs.
Shared Memory Architecture and Conditional Computation: The discussion covers deploying a shared memory space, beginning with read-only access for team members. Machine World is defined as AI engineering based on conditional computation and memory cessation loops rather than traditional procedural coding.
Multi-Agent Collaboration and Quality Control: Speaker 1 outlines multi-agent, multi-human collaboration loops where agents perform heavy computation while humans review and guide. Public content and technical write-ups must undergo strict internal peer review and expert validation to build credibility.
Meditation Integration (Marga Sakaccha) and Cessation Engine: The platform incorporates spaces for Buddhist meditation retreats (3, 7, 10, 14, 21 days up to 3 months) via Marga Sakaccha. The underlying technical engine implements [doctrine term] (causality) and a cessation engine where states arise based on causes and cease when causes cease.
OS Stability, Rendering Pipelines, and Roadmap: Details surrounding a custom stable OS engine, agent harnesses, and replacing slow UI rendering (taking 45 minutes for a 3-second render) with dynamic live 3D pixel rendering directly connected to the memory state are discussed.
Physical Intelligence, Job Disruption, and Economic Realities: Speakers discuss rapid advancements in robotics and AI agent control (mouse/keyboard access, voice-to-voice native training, automated cooking and construction), warning of massive white-collar and physical job loss globally. They emphasize building self-sustaining revenue loops before personal funds run out.
Sri Lankan Market Strategy and User Onboarding: Speaker 2 shares details of their current software workplace handling dealership management systems (DMS/CMS). Speaker 1 outlines how Machine World should bridge local expertise and human-guided loops to help users build their own automated business solutions.
Machine World Memory Substrate and MCP Architecture: Speaker 1 explains that Machine World runs on an isolated memory substrate accessed via unique URLs over MCP. Compute is billed only when active, running on specific CPU/GPU instances rather than persisting as idle state.
AGI Systems as Grown Entities and Loop Optimization: Speaker 1 details how systems in Machine World are grown rather than coded statically. The system uses recursive self-improvement (RSI) loops to fine-tune operations and analyze cause-and-effect relationships within memory states.
Physical and Digital Interface Integrations and Origin Certification: The discussion covers integrating digital applications and physical devices into harnesses. Speaker 1 stresses the necessity of a certified chain of origin to ensure security and prevent unauthorized agent actions in critical systems like banking.
Pre-Alpha Release, Demos, and Cessation Engine: Speaker 1 outlines the pre-alpha release plan, including interactive demos like the reflection engine. The concept of conditional computation and a cessation engine is introduced to prune stale memory data and prevent hallucinations.
Research Publications, Content Strategy, and Funding Options: The strategy involves publishing benchmark research and weekly content to demonstrate the platform's uniqueness. Speaker 1 discusses self-funding with savings versus raising venture capital, which demands accelerated growth.
Planned · The team aims to finalize and publish the Dhamma website before the 30th for Vesak.
Planned · Miyuru and [name withheld] must build a Machine World skill to automatically convert audio recordings into referenced videos for YouTube.
Planned · The base system prompt on the NotebookLM knowledge base must be adjusted from 'discovery architect' to an instructional, explanatory mode.
Idea · External contributors to Machine World could receive token returns when others execute their defined skills.
Planned · Provide Claude Code CLI access to collaborators on their restored laptops to interact with system tools.
Planned · Inner Realm's new song will explore mother-child attachment, showing suffering and effort while depicting non-self (anatta) and avoiding annihilationism or eternalism.
Planned · Automate the complete media pipeline so that whenever a song is uploaded, video generation and distribution occur through automated flows via MCP.
Decided · The Telegram bot system runs locally on the user's machine and does not require opening ports on the router.
Said to exist · The Telegram bot accepts an audio file along with images tagged with specific display timestamps, and generates a video once the 'done' command is given.
Planned · Miyuru and [name withheld] plan to connect the next day around the same time to walk through the Telegram workflow and system setup.
Decided · fal.ai acts as an API router providing access to multiple video models including Kling V3 Pro, Sora, and Seed Dance.
Idea · Use Topaz to upscale generated video up to 60 fps.
Idea · The system should visualize citta (consciousness/thoughts) and rupa separately, showing their causal structures and conditions so users can navigate them interactively.
Idea · Validate generated Dhamma examples automatically within a loop matching Paticcasamuppada and [doctrine term] paccaya causal principles against canonical sources.
Planned · The generated Dhamma outputs and system interpretations will eventually be validated by learned Buddhist scholar monks.
Said to exist · The system has access to digitized accurate versions of the Tripitaka and commentaries to extract insights without modifying underlying text.
Planned · Speaker 2 is to build an initial prototype of the Dhamma query system and share the design link with Speaker 1 for testing.
Idea · Build an interactive visual system to navigate Citta, Cetasika, and their interconnections in Sinhala.
Idea · Examples generated by the system must align simultaneously with the Four Noble Truths, Paticcasamuppada, and [doctrine term] Pakarana.
Planned · Validate system outputs and Dhamma accuracy with learned Buddhist monks.
Decided · The AI system must not alter canonical text content; it should only help users search, surface insights, and link relevant suttas.
Planned · Speaker 2 will build a prototype system with interactive visual navigation for testing.
Idea · Build an MCP-connected test harness with synthetically generated data to simulate retail chain workflows before connecting Machine World directly to production clients.
Idea · Integrate APIs such as Daraz's API into the system to manage inventory, supplies, and dynamic operational optimizations.
Idea · Implement agentic workflows with a three-time loopback structure to pass memory artifacts across multi-agent session layers.
Idea · Use a dedicated validator model to continuously score loop outputs against the original goal specification up to an 80% or 90% threshold.
Idea · Utilize a slash goal command (/goal) and loop engineering artifacts to define targets for multi-agent workflows.
Idea · Incorporate spatial gesture hardware like Leap Motion into future interfaces to interact directly with 3D and world model visual components.
Open · The AI model is struggling to correctly interpret and parse Abhidhamma concepts such as citta kṛtya and akusala states from digitized text.
Planned · Schedule a one-hour collaborative session where Speaker 1 guides Speaker 2 through building the live interface.
Planned · Speaker 2 will start building the test harness based on Speaker 1's guidelines.
Idea · Deploy an embedded offline model within the local application rather than depending solely on Ollama as an access point.
Idea · Fable 5 is noted as an extremely large 10-trillion parameter class architecture requiring massive cluster infrastructure, costing twice as much as Opus.
Idea · A private access model named Sat Code offers a context window of up to 12 million tokens. (unclear in the recording)
Decided · Markdown links in the documentation currently auto-redirect to the home page when clicked.
Idea · Convert physical functions into reusable skills and scale them for mobile app integration.
Said to exist · Automated video uploads to Facebook and YouTube were established using developer page/group keys, Page IDs, and OpenRouter API keys.
Open · Automated uploading to TikTok could not be established successfully.
Idea · CodeX IDE was tested on its free tier for CLI usage but found inferior to Claude Code.
Decided · Miyuru identified factual discrepancies and flaws in examples within the Chitta-vithi and 89 cittas draft documentation.
Planned · [name withheld] Aiya will type and send updated Dhamma content based on noted corrections before final group review.
Planned · A group call will be arranged with [name withheld] to resolve approximately 80% of Dhamma content queries, followed by consultation with Ven. [name withheld] Thero.
Decided · Auto-generated Sinhala subtitles are inaccurate and unsuitable for automated video inclusion.
Planned · Miyuru will push the co-field code changes to a separate branch so Gehan can review them over the weekend.
Decided · Music tracks need to be rendered in both 9/16 vertical aspect ratio (for Shorts/Reels under 1 minute) and 16/9 full 4K resolution.
Decided · In Machine World, render map text animation displays before custom fonts load, causing visual formatting glitches.
Idea · Build a test harness with synthetically generated data to simulate company operations before connecting to live systems.
Decided · The agent system architecture utilizes a 3-tier multi-agent loop to execute steps, refine memory files, and evaluate original goal matching.
Open · The AI model is currently struggling to properly read and parse the Tipitaka Git repository data for specific unwholesome consciousness (akusala citta) mappings.
Idea · Embed a lightweight offline LLM like Gemma directly into the native Machine World application core.
Planned · Speaker 1 and Speaker 2 will schedule a collaborative session to set up the live interactive UI sample for the Tipitaka rendering app.
Idea · Team members should stop typing text prompts and switch to using voice integration in Claude or Codex.
Said to exist · Machine World keeps user memory synchronized on the user side rather than scattered across third-party model providers or agency tools.
Idea · Public presentation of Machine World should begin by clearly explaining the problem space, testing data, and research reports before introducing the tool itself.
Planned · The project will provide a shared memory space for the three team members, starting with read-only permissions.
Idea · The core technical paradigm of Machine World is conditional computation where memory states loop as conditions rather than static code execution.
Planned · A pre-alpha demo flow consisting of an MCP package and an introductory demo video will be provided to selected pre-alpha users.
Idea · Machine World will support Marga Sakaccha automated meditation spaces accommodating Vipassanā sessions ranging from 3 to 21 days up to 3 months.
Said to exist · Machine World is structured upon three pillars, utilizing [doctrine term] causality and a cessation engine to maintain state alignment and intentional forgetting.
Said to exist · The core engine runs on a custom, stabilized OS layer, with an open-source harness allowing users to plug in external agents.
Open · Current rendering approaches take approximately 45 minutes for a 3-second render.
Idea · Replace static code-based video rendering with a live data machine interface that renders the top memory state directly into 3D pixels.
Planned · Achieve a functional demo and pre-alpha release target before the end of November.
Planned · Build an internal wiki connected to the Machine World memory store detailing research, architecture, and founder insights.
Planned · Speaker 1 will host and configure the system and provide connectivity keys/passes to Speaker 2 and the team.
Said to exist · The core memory substrate allows multi-model, multi-agent, and multi-human collaboration on a single shared memory store.
Said to exist · Each user receives an isolated URL mapped to an entirely separate memory store on a CPU or GPU instance that only runs during MCP requests.
Decided · Billing occurs strictly for the GPU time used to process intelligence, with no cost when idle.
Decided · The system allows traversing memory states to evaluate cause and effect for specific decisions.
Said to exist · The development harness includes an RSI engine and an agent library.
Decided · Users interact with Machine World through a dedicated assistant rather than a standard web interface.
Idea · A monthly budget cap should be established for pre-alpha testing to control compute costs while providing a smooth experience.
Decided · The architecture requires human certification of agent origin and actions to validate authenticity to third-party systems like banks.
Planned · The pre-alpha release will include five interactive demos to demonstrate system capabilities such as memory state branching.
Decided · The RSI loop design is intentionally aligned with mental calm, physical ease, and internal reflection.
Said to exist · The model architecture runs using a micro CPU load and cross-model independent harnesses.
Decided · The first pre-alpha release maintains memory alignment as a shared ledger where users can adjust agent trust dials.
Decided · Agents are invoked using specific names/tags, ensuring human-centric conversation and clear scope of authorization.
Said to exist · A reflection engine demo lasting about two minutes has been prepared as an experiential introduction.
Decided · The system applies conditional computation to dynamically generate logic on top of memory states according to causes and conditions.
Said to exist · Machine World currently functions locally while formal research verification is being carried out.
Planned · Speaker 1 plans to host the MCP endpoint within the coming week or two.
Idea · The team needs to decide whether to approach venture capital funds for commercialization or bootstrap using savings until broad user adoption is achieved.
Decided · A voice agent will guide users upon connecting, providing updates on new features and available demos.
Planned · Machine World aims to integrate directly into tools like Zoom calls using real-time voice-to-voice interfaces without intermediate text.
Decided · The cessation engine removes stale data automatically to maintain managed memory and eliminate hallucinations.