Still don't know what to do with AI? Try using AI to arbitrage and earn back your subscription fee.

Author: Huang Shiliang

I have been advising a few relatives and friends to make use of chatgpt, but they all feel that there’s no need to pay for it, probably because they are reluctant to spend money. I keep saying that this small subscription fee can help you earn it back with AI, and I have been teaching one of them, trying to cultivate a successful case, but I haven’t succeeded yet.

Many years ago, a friend who was into quantitative arbitrage tried to sell me his strategy. The core of his PowerPoint presentation was a partial differential equation, and I couldn't understand it at all. But he said this is how to refine gold from mathematics.

In 2017, the mainstream private key management method for ETH was not mnemonics, but keystore(json) files. I often imported the same set of keystore files into myetherwallet and metamask wallets to call different contracts, because at that time, the wallet functions of Ethereum had their own strengths. To achieve arbitrage across different contracts, it was often necessary to import the same set of private keys into different wallets for use.

There was no way, the infrastructure back then was too bad. Looking back now, it was purely luck that I didn't encounter phishing sites or clipboard trojans at that time. Such high-risk operations, even if someone pointed a gun at me now, I would never dare to do it again.

Two years ago, when ordi and other inscriptions became popular, I had no idea how to get involved in arbitrage. Once, by a very random chance, I saw a post on X saying that there were many inscriptions coins that had long been abandoned, but there were still people placing orders to buy them on the dex, along with screenshots. I thought to myself, isn't this a perfect arbitrage opportunity? I could just buy them on the cex and sell them to the order on the dex. Because doing buy and sell orders for inscription-type assets on the dex is quite strange, using PSBT for transactions, some people might just forget about their orders after placing them. After that time, I spent a long time searching for junk in the inscription graveyard, but in the end, I didn't make much btc in sat.

The three stories I have experienced above reveal the three barriers to entering the cryptocurrency arbitrage: difficulty in mathematics, difficulty in tools, and no guidance from experts (mathematical models, tool barriers, experience barriers).

Arbitrage is the theme in the cryptocurrency field that comes second only to trading & investing, I feel it is also an excellent entry point for most ordinary people to enter this industry, and it still is.

In the past, arbitrage indeed had a high threshold. But now with AI, after I tried to delve into using AI to find arbitrage opportunities, it became very clear: the high walls that once blocked ordinary people are being quickly flattened by AI.

Here are my three thoughts on “How AI Reconstructs Arbitrage.”

Threshold One: The “Disenchantment” of Mathematical Models

The biggest obstacle to arbitrage is undoubtedly the establishment of mathematical models.

In the primitive era of the cryptocurrency world, arbitrage models were simple and crude. For example, the earliest exchange arbitrage was simply “addition and subtraction”: when the price on Exchange A was low and the price on Exchange B was high, one could profit by buying and selling manually. Later, with futures and spot arbitrage (the price difference between perpetual contracts and spot prices), although concepts like funding rates and leverage were introduced, it still remained at the level of “addition, subtraction, multiplication, and division.”

However, with the development of the industry, complex derivatives have emerged. Synthetic options, multi-angle arbitrage, and arbitrage between on-chain and CEX have made the models increasingly obscure. For those without a mathematical background, just understanding “how to synthesize a spot with Call and Put” is likely to be difficult for 99% of people.

Now, AI has become the strongest mathematical translator. AI is best at handling these logic and formulas. No matter what math problem I encounter, I just ask AI directly.

In the past, we used Excel spreadsheets to calculate and verify the components of the Uniswap liquidity pool. Now, we can simply ask ChatGPT, tell it the parameters, and it will calculate it clearly for you immediately.

Let me give another recent example that I studied.

Recently, PolyMarket has been very popular, for example, the prediction about “When will Trump disclose the Epstein list?” I am not familiar with the arbitrage models for probabilistic events, so I directly took a screenshot and sent it to AI, asking it: “Is there any arbitrage opportunity here?”

AI instantly provided the answer, explaining to me the game relationship between “forward probability” and “recent probability”, what yes + no = 1 means, and listed all the mathematical logic on how to operate under specific sudden events.

I won't copy the AI's reply, you can ask it yourselves.

With AI, you will never regret sleeping in your advanced mathematics class during your freshman year.

Threshold 2: The “Dimensionality Reduction” of Tools and Environment Setup

The second threshold is the use of technical tools.

If you experienced the ICO of 2017 or the DeFi Summer of 2020, you would know how high the barriers to “operations” themselves were. The Ethereum tools at that time did not have user-friendly UIs, and calling contracts required manually filling in addresses and parameters; to claim airdrops, one needed to configure anti-association fingerprint browsers, IP proxies, and so on.

However, it is precisely because these learning curves are extremely steep that early profits are so substantial.

Now that there is AI, all product usage issues have completely disappeared. If you don't understand something, just ask chatgpt, its answers are too perfect.

In the strategy of farming rewards, setting up the environment is still a challenge for most people. However, now, if you send the strategy intention (for example, wanting to farm potential airdrops from polymarket) to ChatGPT, it will generate a complete execution plan, including the usage methods for all tools.

Now this knowledge really has a zero threshold; all you need is intention, and the rest is just to follow step by step. If you encounter something you don't understand, just ask AI.

You can now ask AI if there are similar strategies in the cryptocurrency exchange for inviting newcomers and trading mining (like Binance's Alpha). Each exchange has similar registration reward strategies for inviting users, and this strategy of consuming user accounts is difficult to monopolize; you can ask AI how to implement it.

AI can also assist in some very simple and auxiliary programming tasks, such as writing a script to monitor on-chain prices or designing a visual data dashboard, which previously might have required a programmer's help, but now AI can handle it all.

But don't expect AI to write code for you or create a quantitative program to compete with various robots for profits; that's unlikely. AI is very unlikely to help you quickly become a quantitative expert or a super blockchain scientist from scratch, but it is enough to help anyone overcome the initial barrier of “not knowing how to get started.”

Now, AI has virtually eliminated the technical barriers.

Threshold Three: The “Decoding” of Experience Transmission

Arbitrage highly relies on experience.

For example, when Binance experienced a significant drop in the prices of wbETH and ETH on October 11, experienced veterans would not hesitate to exchange ETH for WBETH. This kind of sharpness used to require “a master to guide a disciple” or spending years learning through personal losses.

Now, AI is the best “master”.

There are many big shots in the crypto community who love to share, but it still takes quite a bit of effort to extract executable strategies from their sharing. However, with AI, it becomes much easier to extract valuable insights from the big shots' fragmentary remarks.

Every time I see a profit-sharing post on x, I send the post directly to chatgpt, along with the prompt “Please analyze this tweet. What arbitrage models have historically existed for the cryptocurrency and protocol mentioned by the author? If he is correct, what is the underlying logic chain? What tools do I need to prepare to execute this strategy? Where are the potential risks?”

We just need to be diligent “porters”, reading articles every day, reading x, looking at the shares in the group, and then feeding these clues that should originally be commercial secrets to the AI, which can quickly guide you through most mature arbitrage logic. This learning efficiency is incomparable to the traditional “apprenticeship system”.

The Last Bastion: Opportunity Discovery

Never mythologize AI. If AI could make money on its own, it wouldn't need you or your subscription. Why would Sam Altman still be asking you for that 20 dollars? Otherwise, a few capital giants would have already built hundreds of intelligent agents to siphon off market profits.

Currently, AI has only significantly reduced the cost of execution, but it cannot replace the ability to “ask questions” and “perceptive awareness.” AI needs you to give it instructions, needs you to tell it where to analyze. That kind of perception of market sentiment and sensitivity to sudden events still remains the domain of humans. At the beginning of opportunity discovery, it still relies on our diligence, which may also be the value of us humans at this stage.

The focus should be on arbitrage in “non-standard products.” For those highly standardized and data-transparent sectors (such as arbitrage with mainstream coins), quantitative trading bots have long driven profits to zero, and there's no need for chatgpt to play any role. The places where profits may still exist are those that mainstream strategies and large funds overlook.

AI is a dragon-slaying sword, but if you only use it to write weekly reports, it becomes scrap metal. In the context of AI, the competition for arbitrage is no longer about who has better math or who writes better code, but rather about who more diligently feeds new things from the market to AI. You don't need to be smarter than quantitative institutions; you just need to be a step faster than retail investors who still don't know how to use AI — even if it's just half a step, the profit is yours.

Try to set a small goal: earn back the subscription fee for AI through arbitrage. ChatGPT Pro plus Gemini Ultra costs $450 a month. Aim to give yourself another $50 salary, and find a way to steadily earn back this $500 from the market every month.