1. Overview & Propositions

Ithax Labs has changed its name to Eternl.ai.

The new documentation is hosted in: docs.eternl.ai

-----------------------

Eternl.ai is developing infrastructure that enables simple and economical implementation of AGI gaming, where AI participates as companions, adversaries, or elements of dynamic environments.

We achieve this by offering integrated large language models (LLMs) and reinforcement learning (RL) solutions, and by alleviating the adaptability limitations of RL, thereby empowering more games with advanced AI capabilities and deep AI-human interactions.

Our AGI Gaming Ecosystem includes:

1. Gaming AGI: omnipotent AI that not only talks, but acts, combats, navigates, and interacts.

2. Compute Network: massive game inferencing network that is reliable and cost-effective.

3. Gamer Rollup: layer2 blockchain that enables interoperable AI assets and social graph.

4. Game Hub: the destination of hundreds of AI games with ownable omnipotent AI.

Our value propositions are to meet the vast unmet demand in producing AI games, harness the potential of blockchain to elevate AI games, and develop an integrated ecosystem that delivers both the best and the most extensive AI game experience.

Our Value Propositions

1. Solve Challenges in Developing AI Games

Challenges	Propositions—Infra to Produce, Operate, and Monetize AI Games
Hard to produce with tech and cost barriers	Gaming AGI empowers developers to build AI game experiences at drastically low technical barriers and costs.
Hard to operate with the inferencing backend	Compute Network provides developers with reliable and affordable game inferencing computes with no infrastructure management.
Hard to monetize with unproven PMF and cost	Gaming AGI Rollup enables developers to monetize from a global player base with additional channels such as AI assets NFTs.

2. Integrate AI Experiences in Games

Challenges	Our Propositions—Free Game-Engine SDKs for AI interactions
Limited AI performance in game context	An extensive library of SDKs that generate AI experiences to enhance immersiveness, dynamics, and sophistication.
Hard to integrate AI into game production	SDKs in the format of game engine plugins that take standardized and game-compatible input for output trained for game contexts.
Expensive to customize AI features	SDKs are free to use. No upfront cost until inferencing for miners. Required to launch a version non-exclusively on our Game Hub.

3. Optimize AI Performance in Game Contexts

Challenges	Propositions—SOTA Proprietary AI Models
Task Limitations of Large Language Models such as limited: memory, reasoning, output resolution, action capability, 3D generation capability, and animation productivity.	1/ Ability patchers adopt SOTA research concepts to tackle common limitations. 2/ Integrated framework enables our AGI to see, hear, speak, think, learn, decide, and act.
AI Hallucination is when AI models detect patterns or objects that do not exist or are not perceivable by human observers.	Adopt RAG to ground model's output to external sources, ensuring whitelisting and blacklisting for appropriateness.
Lack of adaptability of reinforcement learning agents to settings out of the training environments. The close coupling means that AI actions are hardly versatile to new games.	1/ Enhancing RL training efficiency with transfer learning and MLLM grounding. 2/ Identifying versatile but simple RL environment setups for gaming applications.

4. Address Limitations of Decentralized Inferencing

Limitations of GPU Networks	Propositions—Inferencing Use Case for Web3
GPU Memory Limitation: LLMs are too big to deploy on idled GPUs in DePINs and player laptops.	Lightweight Model: quantized and distilled models to reduce hardware requirements of compute nodes—RTX4090s sufficient for most features.
Security Concerns: hard to protect proprietary models nor verify inferencing in a trustless setup.	Obfuscated Lightweight Model: models are "unreadable" for core technology protection. Proof of inferencing is embedded into the models.
Not Cost-Effective: adding gas fee onto the inferencing cost makes decentralized compute expensive.	Minimal Gas Fee: highly concentrated and long-term compute purchasers of platform games. We post compute record per block but settle once a day.

5. Leverage Web3 Value Propositions for Gaming

Problem	Propositions—Blockchain elevating AI games
Web3 does not improve gaming experience	Blockchain allows players to own and trade AI agents interactable across games, and enables cross-game social networks.
NFT assets destabilize game economics	AI agents are cross-game "souls" that define identity and abilities, and game assets remain game-specific for a stable economy.
No quality web3 games to prove the market	Onboard high quality web2 games by meeting the underserved developer demand with our free-of-charge AI empowerment.

6. Harness Web3 Value Propositions to AI

Problem	Propositions—AI Assets and Web3 Inferencing Use Case
Innovate a new asset class of AI agent NFTs	An asset of usage rights to selected prompts, RAGs, and model weights for an AI model operating in a distributed setting.
Bring viable inferencing use case to web3	Bring a massive web3-tailored inferencing use case enabled by our obfuscated lightweight model and simple settlement structure.