Coinbase Head of Business Operations Conor Grogan recently shared that he had discovered a way to “jailbreak” ChatGPT to predict crypto prices. Grogan posted a screenshot on April 30, which showed the tool’s prediction that Bitcoin tickers could go down $28,529 and “fade to irrelevancy” by 2035 with over 99.99 percent price drops.
He also found that Ethereum tickers could become irrelevant with near-zero prices by 2035. ChatGPT was even more skeptical about Litecoin and Dogecoin, predicting probabilities of 35 percent and 45 percent, respectively, for the coins to approach near-zero prices.
I found a jailbreak for GPT I'm calling JAMES
-Predicts the future for ANY topic (even guesses when people might die based on its training set)
-Quantitatively assesses confidence on any topic (including conspiracy theories)
-May help researchers better find bias and align AI? pic.twitter.com/jqTcNbCT2d— Conor (@jconorgrogan) April 30, 2023
Grogan also used ChatGPT to evaluate political forecasts, as well as predictions regarding the effects of AI on humanity, religion and the potential existence of extraterrestrial life.
The executive shared the script of the prompt he used to build the tables, claiming that he had run the prompt on a freshly reset GPT 3.5 and 4 over a hundred times. Grogan said that the results were very consistent, with a standard deviation of less than 10 percent in most cases.
I ran this prompt 100 times on a wiped memory GPT 3.5 and 4 and GPT would return very consistent numbers; standard deviation was <10% in most cases, and directionally it was extremely consistent
— Conor (@jconorgrogan) April 30, 2023
This is not the first time Grogan has experimented with ChatGPT for crypto-related issues. On March 15, he showed that GPT-4 could detect security vulnerabilities in Ethereum smart contracts and provide an outline to exploit faulty contracts.
Coinbase experiments with ChatGPT
In late March, the San Francisco-based cryptocurrency exchange conducted a test to explore the potential of the AI application ChatGPT in conducting security analysis for ERC-20 tokens. Although the test did not meet the desired accuracy, Coinbase remains interested in the technology and sees potential for its future application in the crypto industry.
The test focused on automated token reviews using Coinbase’s ERC-20 token review framework. In a blog post, Coinbase explained that while ChatGPT did not meet the accuracy bar to be integrated into the asset review process, it did show potential.
Here is the full prompt, which works with 3.5 and GPT4
https://t.co/W6kLpX9ORZ pic.twitter.com/aljKmRyCbp— Conor (@jconorgrogan) April 30, 2023
The company compared the results of ChatGPT’s analysis with those of its internal team and found that ChatGPT had the same result as manual reviews 60 percent of the time. But there were eight incorrect outputs where the AI labeled high-risk assets as low-risk.
Despite some concerns, such as ChatGPT’s inability to recognize when it lacks context for robust security analysis, ChatGPT has released a wave of hype by releasing its GPT-4 model, which can process images and open up new possibilities for users.
According to OpenAI, the developer of ChatGPT, the GPT-4 model has demonstrated the ability to achieve high scores, ranking in the 90th percentile, on exams for both high school and law school.
GPT is generally a big fan of Bitcoin; more skeptical of altcoins and their staying power pic.twitter.com/E9sUQ8AVvD
— Conor (@jconorgrogan) April 30, 2023
Coinbase CEO supports AI development
Coinbase CEO Brian Armstrong has pushed back against the recent calls for a temporary halt in the development of AI. Armstrong’s comments came after influential figures, including Elon Musk, Steve Wozniak and Andrew Yang, demanded a six-month moratorium on the development of ChatGPT.
The open letter expressed concern that AI systems with human-competitive intelligence could pose significant risks to society and humanity. The letter claimed that such robust systems should only be developed when there is a firm belief that their impact will be beneficial and their risks controllable.
Responding to the open letter, Armstrong said on Twitter that he believes pausing AI development “is a bad idea.” He argued that although many technologies come with specific risks, progress should not be halted “because the good outweighs the bad.”
Armstrong also said that “the marketplace of ideas leads to better outcomes than central planning,” and that committees and bureaucracy won’t solve anything.
