0) 🧵 a autonomous world engine based on chatGPT

TLDR: LLM allows us to 1) use intuitive interface to code & customize complex game logic, and 2) simulate not only "physical" interactions, but also intellectual interactions, thus creating an indefinitely *deep* world.

1) I define autonomous as the part of the game that is created by game engine itself.

e.g. the main story, the designed dialogue, and the missions of GTA5 are not autonomous, as they're produced by humans. Only maybe a small part of the game is produced by organic NPC behaviors


2) on the other hand, a famous example of autonomous game is Dwarf Fortress, in which every NPC and game object has its own "autonomy" and can create its own interaction with each other.

3) For example, in a random game, two dwarf doctors decides to get married while their patient is lying on the bed with the body open, because the enemies are coming and they think getting married before dying it the most important thing to do.

source: https://www.ign.com/articles/dwarf-fortress-review

4) It's obviously fun, as the world is very *deep*: everything the player interacts with has a rich history, and this history is infinite, because it's generated by the game engine. While it's impossible to traverse every story in Witcher 3, it's not possible for Dwarf Fortress.

5) However, there are several limitations:
First, the game logic is too complex for both dev & player. It takes lots of hours for a player to get to know the basic mechanics, and it takes the founders 16 years to make it.

This complexity also limits customizability.

6) Second, the interactions are still "physical", as you can simulate the basic physics, wars, fights, and other conflicts, but it's very hard to simulate the multi-agent game theoretical interactions based on each one's unique ideology and perception to the situation

7) a LLM-based game engine might be able to overcome those. For the dwarf doctors example, we can just input the environmental input with each agent's character, and chatGPT can easily generate a detailed account of their action, which can trigger further interactions.

8) this game engine is extremely intuitive: every player can literally read the source code (which are just sets of prompts), and they can customize the world in whatever way they want.

Want to add blockchain into the Dwarf Fortress? Just describe a paragraph and it will evolve.

9) It's also extremely powerful at simulating intellectual things: what if a Marxist male dwarf marries a Capitalist female dwarf, and had a nihilist kid, and each of them controls a country. What will be the diplomacy scene?

LLM can create arbitrary of such details

10) One caveat is that LLM still suck at remembering & calculating physical states (e.g. location & number of soldier) and to produce such a world we need a combination of traditional game engine and LLM engine. But it definitely unlocks some new, revolution gaming primitives.

AI-based game engine

1/ Thoughts on a Trustless Centralized Exchange or how can we build a Binance/Coinbase that can never rug user

Rather than giving CEX the custody of all your assets, it will be deposited into a vault (smart contract). Fund can only be transferred when both CEX and user consents.

2/ How does trading work?

A maker A posts a limit order to sell 1 ETH for 2000 USDC, she also signs a signature S1 that represents this order. When a taker B takes the order, she will send signature S2.

3/ By combining S1 and S2, they can settle this trade on chain, and the vault will give A 2000 USDC and B 1 ETH.

4/ In reality, many trades will be settled together, and the CEX needs to aggregate a large number of signatures so that many transactions can be settled together, saving blockspace and gas fee.

5/ User's trading experience is just like a regular CEX, because the order book and matching engine are fully web2. But when you settle the transactions on chain, you aggregate all the signatures to unlock funds trustlessly, rather than a CEX send you money on their discretion

6/ a great characterization by @_weidai is "multi-party state-channel" that directly encodes order book logic. One caveat is that the system is not censorship resistent, but accepting it could bring CEX trading experience, while every transfer of fund must be signed by user.

A Trustless CEX

I have been rethinking the thesis to build products that "onboard web users." Many of them focus on lowering the onboarding barrier and to make UX more seamless.

It is legit only if the following assumptions are true:

1. The constraint of crypto mass adoption is not the lack of innovative + attractive product, but the friction for user to access those products.

I'm a web2 user not into trading nor understand the web3 jargons like NFT, if there anything fun I can meaningfully do?

2. The unfamiliar experiences for web2 users are *bugs* not *features* AND you can meaningfully abstract them out.

Some technical details CANNOT abstract out, and users need to understand them.

If web2 users can learn WiFi/ISP/iCloud, they can also understand tx hash (?).

Crypto mass adoption?

We built a smart contract wallet at @ETHGlobal SF Hackathon and won $4k in prize. Here's what's so unique about it:
1. it's **natively cross-chain**
2. it has **customizable modular security**
https://ethglobal.com/showcase/gani-wallet-838so

We all know Account Abstraction is going to replace regular EOA wallets: it's much more flexible and fully programmable, allowing things like using ERC20 token to pay gas or social recovery

@VitalikButerin and EF are actively pushing this framework

However, it's currently un-usable, mainly due to:
1. High gas fee, due to on-chain wallet logic
2. no native cross-chain. Since all logic is written in smart contract, you need to re-deploy on all chains your wallet is on.

A Simple graph to explain the idea:

Wallet contract is deployed on every chain (including those with high gas cost) and any update on wallet logic must be re-deployed on every other chain for it to have cross-chain consistency.

Collaborating with @Hyperlane_xyz, we proposed this brand-new wallet architecture.

We designate one chain as the "wallet chain", on which all wallet logic is implemented. We can choose a chain that is fast & secure (e.g. Optimism in the hackathon).

After the security modules are validated on the wallet chain, we relay the transaction data to whatever execution chain and executed accordingly.

Thus, the most gas-intensive operations can be done efficiently.

Another benefit is that we could customize the wallet logic without bothering to update smart contracts on other chains, and thus we are able to implement this cool concept called **modular security**

We did 3 security modules:
1.  Password-protected private key (like MetaMask)
2.  Guardian wallet verification (use your hardware wallet to sign transaction)
3.  One-Time Passcode (UX is like Google Authenticator)

The cool thing is that you can **compose** them

E,g, for all tx worth more than $1000, I require security module 2, and for tx worth more than $1M I require 2 + 3.

All logic is implemented trustlessly on chain -- without trusting a third-party (like what many MPC solutions are).


It was incredible to achieve all these features in just 3 days. Kudos to the team!

And also special thanks to the great advices from @_weidai on cross-chain mechanism as well as the technical integration support from @Hyperlane_xyz

We will continue exploring the thesis of "interface of blockchain (📱)" and do technical research in the wallet space.

For anyone who is interested to talk more, my DM is always open

Gani Wallet: Cross-chain AA Wallet with modular security

I just published some thoughts on web3 social and consumer apps.

The core argument is: web3 social apps should not adopt the same product form as web2. For example, making a new Twitter with web3 tech (e.g. Lenster) does not work.

https://onjas.substack.com/p/web3-facebook-will-not-have-feeds

It's because *product forms are created by the set of underlying technology.*

Web2 tech specializes in information transmission while web3, as a reaction to web2, focuses on a self-sovereign software and value system.

Each of them suck at doing the other's job

Adopting the same product form mean adopting the same criteria. As for Twitter-like products, users value "content liquidity:" the the more content and the more updated the content the better.


Twitter can use efficient database and server system to provide great content liquidity. However, a web3 Twitter must rely on inefficient database (e.g. blockchain), and can hardly compete with Twitter.


The solution is to make our own rules: web3 products like token, NFT, dApp all value data/asset sovereignty, interoperability, and composability, which are not not available with web2 tech at all.

We should also focus on creating web3-native social products, like ENS.

More details in the article.

Special thanks to @JoshuaRSS3, @ianDAOs, @MapleLeafCap ,@_bitbear, @odysseus0z, @JiahuiL37633083 for the discussion or inspiration.

The cover image kindly crafted by @kzdagoof, whose product Curio inspired me on the thoughts around on-chain games.


Article: Web3 Facebook will not have feeds

Thoughts on NFT and fashion industry.

Fundamentally, both are Status-as-a-service (cr. @eugenewei) i.e. you pay money to gain social status (have BAYC profile ≈ wearing luxury suites). 

However, with different mechanism, the business & future prospects are different.

Similarities:

1. Both NFT and fashion goods indicate social status, and people  pay for that.
2. Status can only be established in social scenarios: luxury suites: formal events, NFT PFP: social media.

PS: yes, NFT can be art, but CryptoPunk as digital art is much cheaper than it as PFP. Value comes from consensus, and social scenarios generate consensus most efficiently. 

3. Both NFT and fashion are driven by fads, which are inherent elements in status games.

NFT fads are composed of different collections/brands. 

Fashion fads are different styles (color, design pattern) of each season, and the brands don't change that much.

Note: why are fads inherent for status games? 

High-status stuff will quickly attract low-status imitators, making it less fancy. Thus, high fashion industry must come up with something new to distance itself from consumer brands. 

Differences between NFT & fashion industry:

1. Blue chip NFTs do not need to create fads, because they have a natural price barrier. NFT fads usually take place at lower price level (like the recent black cats).

2.  fashion goods have higher consumable utility and are perishable, while NFTs have higher investment value and are persistent & fraud-proof. 

In the same scenario, the latter is more efficient in generating status.

3. fashion brands usually have numerous product lines, while currently 1 NFT brand ≈ 1 NFT collection. 

OK, those are observations, here are my conclusions/guesses:

1. I expect successful NFT brands to have more collections (and blue chip NFT project are already doing so), because it reduces cognitive burden and produces tools to gain social status more efficiently.

e.g. most people in crypto have heard of BAYC, and that's the result of numerous marketing & branding. However at most 10k people can enjoy the status premium. That's inefficient. 

2. Ultimately, market share of NFT vs. physical fashion brands depends on the relative importance of metaverse and the real world in our social life. 

More specifically, as humans have similar social demands, the more time we spend in the metaverse, the less important are real world fashion brands, vice versa.

3. We might see the NFT counterpart of LV/Gucci as well as Nike/Adidas producing a larger number of collections that can be used by millions of people, given that humans in general spend more time in the digital world.