Tech-bros Adopted AI After Killing Crypto

Oops, they did it again.

Everyone has the friend who is constantly pitching the next money maker. We all had at least one chance to buy Bitcoin before it hit $10,000, $100, or even $1. If only we had listened to our crazy friend maybe we would be rich today. Certainly the monetary value tied to Bitcoin has exploded beyond what even those early adopters may have imagined. The value in Bitcoin seems, at least to me, to be missing.

Following closely on the heels of the crypto craze has been the rapid advancement of AI, namely LLMs (large language models) and the pursuit of AGI (artificial general intelligence). Much like cryptocurrency, the field of AI has accrued a cult-like following. This hype train makes it hard to tell if the recent progress is all its worth or if it falls short of the tech-bro promises. The parallelism between the cryptocurrency craze and AI rapid adaptation extends beyond their technical obscurity. It turns out all those crypto-shillers found a new home in overselling AI.

Absence of inherent value is what allowed tech-bros, grifters, and influencers to monetize the hell out of cryptocurrency in recent years. What they are actually playing on though is the lack of value compounded with the complexity surrounding cryptocurrency. The blockchain definition alone is full of smoke and mirrors in the form of tech-y vocabulary:

Tie that to a bunch of other jargon—cryptography, alt-coin, cold wallet, mining, non-fungible token (NFT), decentralized finance (DeFi)—and you have a system so convoluted it would take a masters degree in shit posting to figure out what is going on.

This is where tech-bros entered the chat by offering high level over-generalization followed by inducing of maximum FOMO. We were told over and over, don’t worry about the risks or the absence of value, this shit is going to the moon.

We are slowly realizing cryptocurrency might not have much going for it beyond rug pulls and money laundering. AI on the other hand does have real life application. In the past couple of years we have seen AI models passing the bar exam, re-inventing the centuries old game of Go, and driving the streets of San Francisco.

Tech-bros saw an opportunity in AI and it came at a perfect time. Look at the google search trends for “cryptocurrency” and “ai” over the past five years.

The downfall of crypto bled perfectly into the uprise in AI. Now this transition may be entirely a coincidence. In November 2022, OpenAI released ChatGPT which to be frank, was a massive advancement in the realm of AI. However, it was still full of hallucinations and the technology of an LLM did not really make sense. Plus, it is really even cool the computer talks back when we saw that as early as 2000 with AIM chatbots? All of this makes it hard to believe AI would become a household topic of conversation seemingly overnight.

Luckily (for big tech), there was a huge population of people already well versed in simplifying complex tech into digestible bites of advertisement. The transition from crypto-bro to ai-bro was seemless, and the overselling began.

Immediately following the release of ChatGPT fear spread that it would soon be taking our jobs. The possibilities were endless using these new LLMs. If you feed it the right data, it can produce anything your dreams desire. Since 2022 we have seen bots tackling software development, graphic design, video editing,and customer service. Candidly, it has improved drastically. No longer are the days of pictures with 14 fingered mutant people; although occasionally it does still hallucinate and tell you to eat rocks.

At the same time it is hard to believe AI could steal my job. Day-to-day I am a software engineer. I have used most of the bots out there which claim to take my job one day: Github Copilot, ChatGPT, Claude, etc. While there are some tasks they are really good at, I still run into pretty hefty bugs in generated code often. While it assists my workflow, I could never see it replacing me. This is where the tech-bros come to AI’s rescue. It is not about AI replacing your job, it is about AI becoming your job.

Wait a second, this tweet format looks familiar.

quod erat demonstrandum

Beyond tech influencers, tech companies have begun pouring massive amounts of resource into AI. Google, Microsoft (and OpenAI), Facebook, Apple, Nvidia; they all want to be in the news for their next big breakthrough. This is a complex relationship because it is hard to tell if the hype was inspired by these tech influencers or if they are riding the coat tails of the big brands. The problem either way is the potential of LLMs and the prospect of AGI may have been oversold.

An intersting paper released in April of this year, No "Zero-Shot" Without Exponential Data: Pretraining Concept Frequency Determines Multimodal Model Performance (Vishaal Udandarao et al.), outlines that we may be further from AGI than we are led to believe. The analysis uses a common test of progress towards AGI called “zero-shot” performance. Essentially, the user asks for information regarding a topic not in the training data (or potentially was a very small subset of the data) but may be closely related to the training data. For example, if a model is trained on pictures of birds, then you would ask it to identify (or generate) an image of a very specific rare bird. The paper finds that often these models cannot produce promising “zero-shot” performance.

Interestingly, the paper finds not only do current models struggle with this task but that they would need near infinite amounts of data to come close. The amount of data required vs the performance likely follows a log-linear scale. An exponential amount of data may only lead to a marginal gain in accuracy in “zero-shot” performance.

What becomes more troubling is that tech giants are likely aware of this issue. At the moment the best way forward appears to be to attempt to meet the needs of the model by feeding it endless data. This means more cases of theft of intellectual property, mining user data from your favorite applications, and overhyping lackluster results.

The fear I have with AI fodder is it will eventually lead to a boy who cried wolf scenario. The public is already becoming tired of the headlines providing false promises for the future of AI. The well of attention and financing will soon run dry when LLMs inevitably fall short of AGI. As a result, resources devoted to AI research will also slim.

One day though there might be a breakthrough which does propel AI to within a hairs margin of AGI. The world may be so accustomed to click bait that we will fail to see it for its truth. Should this point come it may be too late and AI will already have taken our data and our jobs past the point of return.

The popularization of LLMs and the pursuit of AGI in particular is saturating the domain beyond return. It is saddening because there are real specific applications for AI already. Companies like Recursion Pharmaceuticals have been leveraging AI models since 2013. Machine learning has been used to track fraud and identity theft for over two decades. These are real applications of AI which are likely to see a decrease in funding if LLMs fall off the hype train.

All said, I may be wrong. I too remember the days when Bitcoin was under $1000 and I thought surely it would never go any higher.