Every day on the timeline, you see people talking about liquidity mining, leveraged liquidations, and yield from restaking like they’re macroeconomists. Meanwhile, over in Silicon Valley, the default context window for large models has jumped straight to 500k, nearly doubling the amount of documents and code you can feed them. Honestly, on-chain finance is every bit as complex as the code behind large models: nested protocols, multisig rules, dynamic interest rate models—it’s a maze with twenty layers of nesting.
A few days ago, I tried using a large model to help me make sense of a governance proposal and audit report for a long-established DeFi lending protocol. It hadn’t even finished reading the first few tens of thousands of words before it ran out of context and started “forgetting” things. I was so mad I nearly smashed my keyboard. Now, though, it comes with 500k by default. From now on, when analyzing dozens of pages of black-swan postmortems or complex lending liquidation logic, at least AI can swallow the entire “ledger” in one go, without me constantly having to remind it, “You forgot what we were talking about earlier.”
But then again, if the tools are getting stronger, shouldn’t we be getting better at being the suckers on-chain too? People used to complain that DeFi had a steep learning curve because understanding a protocol white paper took so much effort. Now AI can gulp down millions of tokens in no time. Before long, we’ll be able to use large models as on-the-go advisors for everything from on-chain data analysis and smart contract vulnerability scanning to backtesting arbitrage strategies.
Those dog-themed projects that do nothing but shout slogans and post memes will probably have their dirty laundry exposed in full in the face of this level of information-processing power.
When you’re doing DeFi research, do you usually hand the data over to AI to crunch? Or do you think the whole thing is unreliable and prefer to calculate everything yourself, one number at a time? Let’s chat in the comments: what are the most useful on-chain analysis tools you’re using these days?
A few days ago, I tried using a large model to help me make sense of a governance proposal and audit report for a long-established DeFi lending protocol. It hadn’t even finished reading the first few tens of thousands of words before it ran out of context and started “forgetting” things. I was so mad I nearly smashed my keyboard. Now, though, it comes with 500k by default. From now on, when analyzing dozens of pages of black-swan postmortems or complex lending liquidation logic, at least AI can swallow the entire “ledger” in one go, without me constantly having to remind it, “You forgot what we were talking about earlier.”
But then again, if the tools are getting stronger, shouldn’t we be getting better at being the suckers on-chain too? People used to complain that DeFi had a steep learning curve because understanding a protocol white paper took so much effort. Now AI can gulp down millions of tokens in no time. Before long, we’ll be able to use large models as on-the-go advisors for everything from on-chain data analysis and smart contract vulnerability scanning to backtesting arbitrage strategies.
Those dog-themed projects that do nothing but shout slogans and post memes will probably have their dirty laundry exposed in full in the face of this level of information-processing power.
When you’re doing DeFi research, do you usually hand the data over to AI to crunch? Or do you think the whole thing is unreliable and prefer to calculate everything yourself, one number at a time? Let’s chat in the comments: what are the most useful on-chain analysis tools you’re using these days?