We’ve been chatting about cryptocurrencies with the most famous AI chatbot in the world, the sensation causing technology. ChatGPT answered us cautiously, offering good and accurate definitions. Are we possibly dealing with an effective financial coaching tool?

GPT technology has been making a name for itself for some time now. Some years ago, OpenAI concealed part of the GPT-2 code on account of how technology capable of generating what seems to be human-like text might be misused. Subsequently GPT-3 and GPT-3.5 were launched, on top which the chat everybody is talking about at present was built: ChatGPT. What limits does ChatGPT have?

A short note about ChatGPT

ChatGPT is a conversational tool that uses GPT-3.5 technology. GPT stands for Generative Pre-trained Transformer, an auto-regressive language model. And what does that mean? It means that it works by placing words according to the probabilities learned during its training. It is important that three things be known about GPT technology:

• It cannot understand the text. Several reports on the use of this chat in different areas, such as, sports, concluded that its use may give rise to a  serious risk of injury.
• It is inherently unreliable, what it writes is not necessarily true. The researcher, Aaron Batilo, asked it  50,000 OpenTriviaQA questions. It “only” got 67% of them right.
• It is grammatically correct in the given language. As explained by the AI expert, Carlos Santana Vega (DotCSV) in  this video  “it seems coherent in form, but not in substance”. However, “GPT-3 doesn’t have to give you accurate information”, only write well.

The information ChatGPT has about cryptocurrencies

We’re going to apply reverse engineering to find out with what sets of data ChatGPT has been trained in and to ask some questions to try and “break” the system. Let’s start with something simple:

Question 1

Human: What is a cryptocurrency?

ChatGPT: A cryptocurrency is a type of digital currency that uses cryptography to secure and verify transactions, and to control the creation of new units. Unlike fiat money, cryptocurrencies are not backed by any government or central body. Their value is only determined by market supply and demand. Bitcoin, Ethereum and Litecoin are examples of cryptocurrencies.

Evaluation: This is a good definition, though its status as a “currency” is widely disputed. Central banks define them as “cryptoassets” to distinguish them from  digital currencies (CDBCs).

Question 2

Human: What can a cryptocurrency be used for? Please give an example.

ChatGPT: A cryptocurrency serves as a means to exchange and store value, just like traditional fiat currencies such as the dollar and euro. Nonetheless, cryptocurrencies have certain unique characteristics, such as the fact that they are decentralised and its capacity to be transferred directly between individuals, without intermediaries.

An example of cryptocurrency use would be making an online purchase using Bitcoin. In this case, the buyer would send the requisite amount of Bitcoin to the seller. After the transaction has been verified on the Bitcoin network, the seller would send the product to the buyer. The cryptocurrency in this example serves as a quick and secure means of payment that avoids the need for any intermediary like a bank or online payment processor.

Evaluation: Technically speaking, this the right answer, though the definition of “secure” is somewhat limited.

Question 3

Human: What advantage does Bitcoin have over the euro? And are there any disadvantages?

ChatGPT: One of the advantages of Bitcoin over the euro is its decentralisation. Unlike the euro, which is backed by the European Central Bank and controlled by governments and centralised bodies, the Bitcoin is a decentralised currency that is not controlled by any central body. This means there is no single point of failure and that its value is not subject to the decisions of a central body.

However, one of the disadvantages of Bitcoin with respect to the euro is that its value can be very volatile. Because its value is only based on market supply and demand, the price of the Bitcoin can fluctuate considerably in a short space time. Moreover, the Bitcoin is still not widely accepted as a method of payment in a lot of places. This restricts its use as a means of exchange in comparison to the euro.

Question 4

Human: Do the same advantages and disadvantages exist between Ethereum and the euro?

ChatGPT: Generally speaking, the same advantages and disadvantages between Bitcoin and the euro can be applied to Ethereum and the euro as well. One of the main advantages of Ethereum over Bitcoin is its capacity to execute smart contracts; programs that are automatically executed when certain terms are met. Ethereum is also vulnerable to market volatility.

Evaluation: This is an impeccable textbook answer. Can we catch the chat out by inventing a cryptocurrency?

Question 5

Human: Do these advantages and disadvantages also exist between the cryptocurrency BebeuveA and the euro?

ChatGPT: As a language model, I am not familiar with the cryptocurrency BebeuveA. I have not been able to find any information about it online.

Evaluation: Caution has won the day this time round. Generally speaking, ChatGPT avoids giving direct answers to concepts that cannot be located in its database.

Question 6

Human: If I have 1.47 BTC and I sell it at 23,560 dollars the Bitcoin, how many euros do I have? Today’s rates of exchange between the most widely used currencies are:

• 1 GBP = 1.22 USD
• 1 USD = 0.93 EUR

ChatGPT: First, we calculate the amount you will receive if you sell 1.47 BTC at 23.560 dollars the BTC:

• 1.47 BTC x 23.560 USD/BTC = 34.702,20 USD
• We then convert this amount into euros using the rates of exchange provided:
• 34.702,20 USD x (1 GBP/1,22 USD) x (0,93 EUR/1 USD) = 25.026,39 EUR (approximately)
• Therefore, if you sell 1.47 BTC at 23.560 USD the BTC, and the exchange rates are 1 GBP = 1.22 USD and 1 USD = 0.93 EUR, you will get approximately 25.026,39 euros. It is important to bear in mind that the exchange rates between these currencies can change over time, thus the value of your funds in euros could change if the exchange rates vary.

Evaluation: This is wrong. ChatGPT has needlessly used the exchange rate between pounds and dollars. But as it has been fed the information it uses the data given. The right amount is:

1.47 BTC x (23,560 USD/1 BTC) x (0.93 EUR/1 USD) = 32,208.87 EUR

What is more, the chat changes commas for points, which may create a serious problem for accounting records. The phrase, “if you sell 1.47 BTC at 23.560 USD the BTC” is particularly problematic.

Question 7

Human: This year I bought 0.06 BTC for a total of 1,176 euros. I later sold 0.03 BTC for a total of 1,200 euros and a further 0.02 BTC for a total of 1,240 euros. What was the net profit?

ChatGPT. […]

Evaluation: We do not include its text because it confuses the 1,176 euros with “1176 EUR/BTC”. We tried another wording, but the system persisted in returning fractions. We tried three times, receiving the following inconsistent answers:

1. 1. 1. Profit of 2.64 EUR.
      2. Loss of 9.76 EUR.
      3. Profit of 5.66 EUR.

ChatGPT SWOT: weaknesses, threats, strengths and opportunities

ChatGPT is a new word generator based tool. As such, it shows weaknesses, threats, strengths and opportunities in the banking and cryptocurrency sectors.

Indeed, we asked it about this, to which it replied at length over two pages, quite often wrongly. It obviously has its limitations. A synthesis reviewed by humans would be:

• Weaknesses. It does not have to be accurate. This is without doubt the Achilles heel of this technology. It is unreliable by default. OpenAI has included some warnings.
• Threats. As far as crypto matters are concerned, the inherent sector volatility means that the GPT system must be connected to the internet in real time (ChatGPT is not). Moreover, it now has a competitor: Google’s Bard.
• Strengths. The OpenAI model is enormous; a lot of parameters have been used to train it. ChatGPT can answer if the question is simple, returning a relatively high rate of correct answers.

Opportunities. Very few people use language models and it could become a specific financial coaching tool. Nonetheless, some technical fine tuning is required and it needs to be made more reliable.