“Hey ChatGPT the market is pumping, I’m sidelined and coping, find me the next 1000x shitcoin I can ape my entire net worth into and find it now!”

Of course, we’re not redacted enough to ask a Large Language Model such a stupid question, right?

The problem is, if you did ask such a question, the most likely response would involve something about the irresponsible nature of gambling your entire networth on sub $100k market cap shitcoins.

It seems ChatGPT simply doesn’t understand having one SOL and a dream.

Degenerate questions aside, there are some real questions that definitely need answering.

Are Large Language Models (LLMs) actually useful in crypto research? Can they help you find potentially great investment opportunities? And more importantly, can they help you save time?

The obvious answer is yes, and if you’re not using them to some extent in your crypto journey, you are likely working harder, not smarter.

So, how do we make the most of our AI friends? What kind of information can they access? And how can we use that information to reach the promised land?

Read on to find these answers and hopefully turn yourself into an AI-fueled, bet-slinging ape.

Common LLM use-cases

There are a number of use cases that stand out from the get-go. These are ways to use LLMs to analyze data that can be useful to you as a gambler of magic internet beans, and can help give you a slight edge over the average left-curver who just closes their eyes and throws darts at DEX Screener.

Code

If there is one thing that AI excels at, it's code.

Microsoft CEO Satya Nadella recently said that 30% of Microsoft’s code was written by AI. The company's CTO, Kevin Scott, said he expects that 95% of all code will be AI-generated by 2030.

With that in mind, it makes perfect sense to put this ability to work in the form of smart contract auditing.

Many shitcoins have dodgy code. Ever had the privilege of buying a token that literally went up, only to realize that you couldn't sell?

Ahhh yes, the classic honeypot.

Or a token that claims to have a finite supply, but buried deep in the code was a function that allowed for endless minting of new tokens? Rugged.

Most of these vulnerabilities can be avoided by running the code through the LLM of your choice and having it identify the red flags for you.

Yes, you may need to ask the right questions to get the information you need, but you can also ask the LLM for a list of common red flags to look for and have it scan for them.

This type of auditing becomes especially useful when aping new coins that have just hit the market and could be rife with dangerous code that is going to wreck you in a flash.

Auditing is only one part of LLMs and their crypto coding capabilities. Writing the code for smart contracts is another.

We have recently seen the birth of “vibe coding” and with it, platforms allow users to spin up new tokens and applications using natural language prompts through an LLM.

They then pump out code on the backend and build out whatever your heart desires.

Some seriously interesting and profitable shitcoinery has gone down via the use of vibe-coding platforms. If you want to know more, then be sure to check out our top-notch coverage of the subject here.

Who knows, you might just build the next AAVE and print your way to a Lamborghini and a yacht parked along the coastline of Monaco.

Transaction analysis

Another area where LLMs shine in crypto research is transaction analysis. If you’re an on-chain sleuth, you need to use these LLMs to work for you, or you simply won’t be able to keep up.

As it turns out, LLMs are far better at interpreting complicated transaction graphs, identifying patterns and providing detailed insights into user behaviors than any of the traditional methods.

The stats speak for themselves.

LMMs have shown 98-100% accuracy in identifying in/out degrees and transaction values within node-level analysis, or in simpler terms, the number of connections and amount of money moving around within a network.

They also have a 95% detection rate for identifying high-value outlier transactions (e.g. whales setting up unusual buy and sell orders). They can even figure out patterns within mixer services with a precision rate of 89%.

LLMs take the cake over any human effort by a country mile, even for simpler tasks like identifying exchange wallets and wallets with high win rates.

Market analysis for trading

With the recent rise of AI agents like AIXBT and MOBY, it has become clear that the future of market analysis is firmly in the hands of our robot overlords.

The work-rates that these agents can produce leaves the energy drinking, nicotine addicted shitposters in the dust and trying to keep up with them seems like a futile task.

LLMs play a large role in this type of analysis and can be used for identifying everything from technicals on the chart to overall sentiment in the marketplace.

Platforms like Kaito have leveraged this data exceptionally well and have established themselves as stand-outs in AI-powered data analytics.

The good news is that for the average pleb swimming around in a sea of unfiltered crypto data, the standard LLMs can actually do a pretty good job of deciphering it all.

A good example of this is GPT-4, which has the capabilities to scan through millions of social media posts to accurately gauge bullish or bearish sentiment levels.

This can be useful for detecting spikes in positive sentiment during periods of flat price action that may give users a hint of an upcoming pump.

The same tactics can be used in the opposite direction as sentiment starts to drop off during a strong run up and may signal some kind of a reversal or stall-out in price.

There are even a number of frameworks on offer, like FS-ReasoningAgent, that do a pretty good job of separating factual evidence (like price data) from subjective evidence (like sentiment).

These kinds of frameworks have seen an improvement in P/L of up to 10% on majors like BTC, ETH and SOL.

The numbers seen below speak for themselves. It seems LLMs can really be used as pretty solid trading assistants.

Technical analysis is also within the scope of many LLMs. Some will allow the uploading of charts and then provide insight into key levels and overall price trend to give users an idea of where the price could be heading in the short, medium, and long term.

This can be especially useful for identifying times of increased volume and volatility ahead of major moves.

As useful as these LLMs are at analysing market data, they are still just a tool to add to the kit and probably shouldn’t make up your entire trading strategy, just yet.

Although, in saying that, most of us have no idea what we are doing anyway.

Why should you use an LLM?

Now that we know the most common use cases for LLMs within the world of crypto research, let’s take a look at why you should use them and when is the best time to do so.

Upsides of using AI research assistants

There are a number of solid reasons to start using LLMs in your crypto research. Who knows, they might just give you the edge you have been looking for.

The major advantages of doing so are that LLMs can contextualize data that would otherwise be hard to understand, especially when dealing with transaction data and code.

They are also able to handle large data sets and multiple different sources at once, making it far easier to aggregate information in one place.

Here are some of the most important reasons to get prompted today.

Speeding up the process

Possibly the most profound reason to incorporate an LLM into your crypto research is to simply speed up the process of skimming through ungodly amounts of information.

Crypto moves fast, and keeping up as a humble ape is a hard task, even for the most terminally online individuals.

Using LLMs to triage information, summarize complex documentation, categorize and highlight key information points is an invaluable way of speed running your research process.

New project just launched? A new upgrade just happened on your favorite L1? Just run the whitepaper or documentation through the LLM of your choice and have it give you a detailed overview of the key points, minus all the crypto jargon, in a matter of seconds.

Automation of tasks

If you find yourself doing the same repetitive tasks over and over again in order to get ahead in the casino you might be better off having an LLM do them for you.

Whether it's digging around on block explorers or trying to find the best yield farming opportunities across DeFi, there might be a place for an LLM to improve the efficiency of the task you are doing.

There are even a number of plug-ins available, like RSS3’s Web3 User Activity ChatGPT plug-in, that allows users to input a wallet address and have it spit out full, detailed summaries of its on-chain activities.

As mentioned earlier, one of the standout automation tasks that can be done with an LLM is reviewing code and auditing contracts of shitcoins before you ape in.

These LLMs are incredibly effective at spotting code-based anomalies and picking up on red flags.

Let’s be honest, most of us aren’t super shadowy devs, so having the ability to give a quick overview of a token's code can save us all a lot of time and maybe a lot of our hard-earned money!

Building out your own LLM

Although this is just a beginner’s guide to the world of crypto LLM use, it would be stupid not to mention the fact that you can build out your own LLM to perform crypto-specific tasks if you have the technical ability to do so.

By building out your own LLM to take care of tasks, you will get superior results than you would by simply using the generic products that are currently available at this point in time.

There is a three-step workflow required to build an LLM that one needs to understand before starting this task.

1. Feeding data into the model to make sure it gives the required outputs. This is commonly known as embedding data and involves inserting files, generally fragmented, into a database. This database is known as a vector database.

1. Creating the prompts needed as input into the language model. When a user submits a search request, the LLM itself will produce a series of prompts that can query the vector database, and then a final prompt that combines all this information together.

1. Executing the prompts to give an output. This involves handing off the information gathered to an existing LLM model for what is known as Inference.

This process usually involves the original prompts being broken down into numerical representations that the model can understand. These numbers are then decoded and turned into human-readable text before being given as the final output.

Obviously, this is an incredibly oversimplified explanation, but it is hopefully enough to get your degenerate brain around the basics of how these models work.

If you’re not feeling up to the task of building your own LLM, there is also the option of integrating the APIs from your favorite websites like DefiLlama and Coingecko into existing models like ChatGPT so they can more readily access real-time information as needed.

One could expect to see many protocols in the not so distant future looking to integrate LLM style features into their blockchains for tasks like regression, classifications, text completion, and AI generated content (AIGC) on-chain.

Notable crypto-related LLM tools

If you’re looking to dive deeper into using these large language models in your crypto research, it is probably worth taking a look at some of the tools available to help you so.

There are a number of different tools and plug-ins in the market, and this list is just a drop in the ocean of what is available, but it's not a bad place to start if you’re just getting warmed up.

LLM4FUZZ

LLM4FUZZ is an advanced fuzzing framework for use in smart contract auditing. This tool takes traditional fuzzing to the next level and makes up for some of the missed vulnerabilities that occur from using these traditional methods.

This works by using the LLM to focus on high-value areas of the code base where vulnerabilities are most commonly found.

It then converts the code into structured representations like syntax trees, from which it can predict different sequences of function calls that may trigger complex vulnerabilities within the smart contract itself.

With the help of LLM4FUZZ, you can rapidly become the auditing pro you always dreamt of being.

BlockGPT

BlockGPT is basically a crypto specific LLM that has been trained on blockchain data that allows users to access real-time insights and tailor-made tools that are specific to different crypto-related tasks.

The BlockGPT platform provides a solid base for sentiment analysis, wallet tracking, yield farming opportunities, news, real time price analysis, and pretty much anything a user can dream of that relates to blockchain information.

This is all done via a ChatGPT-style natural language interface.

Other note-worthy projects

There are also a number of projects out there that fall under the AI agent sector that integrate some form of LLM-style interface into their frontend to help users get the best experience from their products.

Some of the leading names in this sector include Griffain, HeyAnon, Swarmnode, and Orbit.

These projects aim to simplify a host of DeFi-related tasks for users and can perform actions like trades, swaps, staking, accessing borrowing and lending strategies. In fact, they can also give out general information on a variety of crypto-related data points.

It is very likely we will see integrations of these types of projects into block explorers, websites and wallets at some point in the not-so-distant future.

If you’re looking for an easy to understand, ape friendly guide on the world of AI agents be sure to have a read over our AI Agents Beginner’s Guide for everything you need to know to get you started.

Drawbacks of crypto-LLM use

Despite the endless use cases for language models to supercharge your research processes, there are still, as always, some significant drawbacks to consider.

The first issue that comes with the use of LLMs is the fact that these platforms have token limitations.

These limitations can restrict the amount of data they can process at one go, meaning that large code bases and multiple pages of complex documents might overload them.

The workaround here is to break down your tasks into chunks so that the interface can handle the amount of work you are throwing at it, which is still likely to be much faster than doing it yourself but may require a little more effort on your part.

The next issue is the potential for inaccurate explanations and insights given by the model being used.

Most of these models have cut-off dates for the data sets they have been trained on and some of these cut-off dates can be quite old when compared to the fast-moving pace of crypto.

It always pays to double-check any outputs given by an LLM if you are suspicious of its accuracy, as “hallucinations” are possible.

A hallucination is basically when an LLM makes up its response, kind of like that know-it-all friend we all have who claims to know the answer to everything and is full of “trust me bro” sources.

Finally, there are some concerns around data privacy that need to be considered.

Smashing your favorite LLM with sensitive and personal crypto data may put the more cyber-sensitive types on edge.

Where is this information being stored? Who is it being given or sold to, now or in the future? Can you really trust these third-party APIs?

These are all reasonable questions to ask yourself before you give some robot on your screen all your wallet addresses and transactional data.

Conclusion

The combination of crypto and AI is really a match made in heaven. Many of the core fundamentals that make blockchain technology so groundbreaking also have major use cases in the world of artificial intelligence.

It is very likely that the biggest future breakthroughs in AI will be made either on-chain or with some sort of crypto integrations baked into them.

LLMs themselves are already transforming the way we do crypto research and will only get better and more specialized as time goes on.

The more crypto-specific, domain-specific tools and integrations that come out will likely 100x the capability of the average user and turn us all into basement-dwelling research analysts.

Using LLMs for complex, high-volume tasks and then iterating on their outputs to get to the exact data points we are after is a real no-brainer for anyone who wants to become more efficient at digging through the crypto trenches.

We are all aware of the exponential speed that crypto moves at and AI is no different. It seems that almost every week there is a new model launched with new capabilities that far surpass the previous ones.

If you are someone who wants to keep up with all this fascinating AI tech, make sure you check out our Big Machines updates and weekly newsletter that keep track of all the latest happenings in the AI world in the same laid-back blocmates style you are used to reading.

May the green candles be with you.