Let's skip the AI history part (there are countless reads out there) and start from the present day: AI is everywhere, whether you know it or not.

Take your phone, for example. The effects on the picture you take, your smart assistant, and your GPS app telling you there's a faster alternative are all governed by algorithms that can learn.

One thing that is often overlooked is how fast AI is evolving and, consequently, how fast the market is adapting to it.

2022/2023 saw the great rebirth of AI, and money flowed in as debates intensified. While AI bros were deep in the trench between the layers of their neural networks, buzzword seekers began betting on the ontological value of "Artificial Intelligence."

That is how the narrative was born.

While most of the market was heading the other way, global funding for AI reached $50 billion in 2023, a 10% increase from the previous year.

In anticipation of a new crypto cycle, the blockchain gold rush towards artificial intelligence was inevitable, with the final goal being openness, trust, reliability, making a lot of money.

This created a perfect environment for a species that thrives on opportunities: the LARPs.

In the tech world, a company offering "tailor-made AI solutions integrated into your system" usually means "we have no idea what to do, but we like saying that we do AI"—that's the product they're trying to sell you.

So, back to the blockchain, here's a revelation for you:

In a world full of LARP, most AI tokens are complete shit.

In the ever-expanding token pool, even just focusing on projects that have working products, red flags emerge while their marketcaps reach the GDPs of small nations.

So, for AI Due Diligence 101, this is how you can assess whether a project exploits AI in a good or bad way, but first a couple of notes:

No token names will be used, and nothing should be considered financial advice because it's not.

Good use of AI doesn't necessarily mean the project is legitimate, and bad use of AI doesn't necessarily mean the project is a scam. Always do your own due diligence.

Here's a series of examples and elements to look for when you think you might've found an AI gem

1) Vibe check

Try to understand how AI is used in whatever project/token you're exploring. Riding the wave with AI-based projects might give you an edge in the market, but the risk increases if the AI part is purely speculative.

The crypto market loves new tech but loves money more: while AI integration might drive in buys now, speculative profit is ALWAYS a good reason to sell.

2) Presence of technical documentation and code

If the project passes the vibe test and does not seem purely speculative, the first technical sign should be easy to spot: look at the whitepaper. If you can't find one, look for a pitch deck or try to find information on social media.

Consider that documents are extremely informative and should be easy to find. Hard-to-find information is a red flag.

                                                                 $$\sum_{i=1}^{n} \mathcal{L}_i * S_i$$

Another extremely important indicator is Github:

• When was the last update on the repo(s)?

• Did they fork something? If so, what did they fork?

• If they forked, what did they commit?

• Do the contributors to the Github have done other works in the past?

A repo full of forks might make you suspicious, but it all depends on what the forks are for. If they're a necessary component for the project, there's nothing wrong with them. If the entire project is a fork, well… take your guess.

Note that a Github (or any code repo) might not always be present or complete, especially for closed-source tech.

3) Evaluate the AI's role and necessity

Once you've obtained some technical documentation, it's time to put on the Sherlock Holmes hat and find out how the project you're studying integrates AI. This is where things can get really interesting (or sketchy, depending on what you find).

First, you need to determine if AI is a core component of the project or if it is just being thrown around as a buzzword to attract hype. You will find yourself in one of these four scenarios:

• There are mentions of AI without actually talking about what it is used for (🚩🚩🚩)

• The technical docs are oversaturated with AI-related terms. Stuff like: "Our Convolutional Neural Network will be used to assess the degrees of freedom in an LP space, transforming parameters into Eigenvectors mapped to e8-lattice to find out the next paradigm" - all of this with no mention of system architecture or information on actual implementation. (🚩)

• You can find the system architecture, most likely filled with LaTex and diagrams that make sense (more or less) or technical jargon supported by publications, citations, etc. (✅)

• The project doesn't use an in-house AI system but relies on existing tools/models/algorithms, quoting them and explaining their integration (✅)

The first scenario is the majority of tokens/projects you can find on the blockchain. 99% of AI-themed altcoins (shitcoins) fall into that category.

Scenario 2 is hit or miss, but usually technical language without publications/architecture/VMS isn't worth much.

Scenario 3 is the most promising, as it offers actual studies supporting the development of the needed tech for whatever you're studying.

Scenario 4 is also good; you can consider it "not needing to reinvent the wheel." If there's already a good solution, implementing it instead of developing it from scratch is always the best choice.

After that, ask yourself if AI makes sense for what the project is trying to achieve. Does it bring value to the table? Is it solving a real problem or enhancing some aspects of it in a meaningful way? Or is it just being shoehorned to make the project seem more cutting-edge? Please exercise extreme caution if the answer is unclear or if AI feels like an afterthought added to the project.

Now, let's dive deeper into the technical aspects of your due diligence. If you get your hands on the system architecture, pay close attention to the choice of tools used and their suitability for the task at hand. For example, if the project claims to use a Convolutional Neural Network (CNN) for natural language processing tasks, that's a red flag, as CNNs are primarily used for image and video processing. Similarly, if they mention using a Recurrent Neural Network (RNN) for image classification, that's another red flag, as RNNs are better suited for sequential data like time series or natural language. You don't always need complexity in a system, and as Occam might show, most use cases are better solved with algorithms that are less computationally hungry than a neural network.

Deep learning can offer a more customizable architecture by introducing activation functions, loss functions, and optimization algorithms. If the project uses unconventional or outdated approaches without proper justification, it might indicate a lack of expertise or understanding of AI's current state of the art. Look for the most common loss functions, like cross-entropy for classification tasks (is this an orange or an apple?), use for regression, and optimization algorithms like Adam, RMSprop, or SGD with momentum.

If the data used for training are public, the project should provide information on the dataset size, diversity, and quality. If they claim to have trained their models on a large dataset but provide no details on its composition or origin, that's no bueno. Additionally, if they don't mention any data preprocessing, normalization, or augmentation techniques, it might suggest an oversimplified approach that could lead to poor generalization.

Finally, pay attention to the evaluation metrics used to assess the model's performance. If the project only reports accuracy without considering other relevant metrics like precision, recall, F1-score, or area under the ROC curve (AUC-ROC), it might indicate a lack of rigor in their evaluation process. A comprehensive evaluation should include multiple metrics and ideally be performed on a separate validation or test set to assess the model's generalization capabilities.

4) Assess the team's expertise

The next step is to research the people behind the project. Starting with the fundamentals, a good team dedicated to a good project has little reason to stay on.

Fetch their LinkedIn (ugh) and look for the team members' backgrounds in their fields.

Do the tech people have any experience in machine learning/deep learning/data science/AI?

What were the past roles of the management? Have they all (or most of them) worked with AI before?

A team with a solid background in the field is more likely to develop a legitimate AI-based project (DUH).

And while you're there, if you find any papers, crosscheck the names. You have no idea how many projects list scientific papers in their work, while not even the 10th author is part of the team—red flag.

The following points are considered best practices for due diligence. TLDR: Check if the team is working, check if the community is smart, and draw conclusions.

5) Examine the project's progress and roadmap

Now that you're done with the technical part, it's time to resume due diligence as you know it. Check and see if the project is even making moves. You can find those answers in the roadmap. If they don't have one, it's a bad sign.

Look at their current stage of development and what they have planned for the future. Is there a working prototype or beta version that you can check out? Or does it fall into the buzzword hype? Steady progress inspires confidence; otherwise, if the team makes grand claims without evidence of actual development, that's a red flag. Results > hype.

6) Check everything orbiting around

In other words, have a look at partnerships and community.

Great partners can help a project achieve greatness, and as stated before, there's no need to reinvent the wheel. So, finding a project that uses non-proprietary tech meaningfully to exploit its product is a good sign. If the project works with well-known companies or institutions, that's also a green flag.

But as with the tech part and roadmap, be careful if they make vague claims about partnerships without concrete details or proof. Name-dropping without proof is worth nothing, so if in doubt, try to confront the team—or directly the partners they claim to have.

After that, go check their userbase, look at what the people are saying, measure the involvement in joining the project's socials, fetch info on Reddit, and do whatever it takes to grasp the sentiment. A good KPI for that is how communicative the team is—people need to be reassured, and clarity is all it takes, while a lack of it (especially in a hype scenario) might be a sign of a downfall.

Another good indicator is how knowledgeable the community seems: good communication leads to good understanding, which leads to happy users. Avoid the moonboys and overhype, stay focused and cold, and try to grasp as much information as you can.

Information is how you win.

Note: Nothing you’ve read should be interpreted as a financial advice. Do your own research. Dingboard is an experimental AI tool for image editing made by Yacine.