I’d define a tech bubble as: collectively over-estimating the impact of a technology on financial outcomes. And by that definition, we are in a gigantic AI bubble.
Some quick thoughts behind the AI bubble
# LLMs are unpredictable, fuzzy, and stochastic. They’re fundamentally incapable of providing material business outcomes (except straightforward automations) where nuanced judgement, accuracy and accountability is needed.
# In SaaS, the marginal cost for one more user is minimal. A large part of the cost is tied to customer support and cloud. But in LLMs-based software, the compute cost keeps rising with token usage, suppressing the margins.
# Chat GPT’s conversion (from free to paid users) is 5%. This is abysmally low and it points towards a possibility that people are not getting outcomes they’d pay for.
# LLMs are very cool. But where businesses are struggling is achieving ROI on them. In most of the cases, there is no concrete $ savings using AI, yet.
# Even in the most straightforward use-case which is coding, AI is still not moving the needle massively. Tech leaders often report that the AI-generated code is 10-30% and it often leads to tech debt.
# With the rise of AI slop on the web, further training on already regurgitated data implies that LLMs are hitting a plateau. A better route for fundamental model creators could be to explore other frameworks apart from Transformers, and that pursuit to hit jackpot might take a while.
# AGI seems to be a high vanity pursuit. LLMs have a long way to go still, and AGI looks like too big of a leap, a moonshot. If AGI was in sight, OpenAI wouldn’t be hiring thousands of sales staff across the globe and pushing for B2C use-cases.
# OpenAI is releasing product updates to acquire more users and increase their retention. A company that genuinely believed it can achieve AGI would solely focus on it, rather than these non-fundamental updates like these B2C use-cases.
Why is the hype not ending?
Within corporations, leaders are hyping up the impact of AI for mainly two reasons. They either (1) have to keep selling AI or compute (2) or have to show high AI adoption numbers in their orgs to their investors who are not technical.
AI startups are hyping AI because their funding depends on it: VC’s are investing in anything that sounds like AI, to avoid missing out on any upside. And founders are having fun building ‘cutting edge’ products and raise funding.
Even if these AI startups fail (99.9% of them will), they’ll get to write things like “Ex-Sequoia funded startup” on LinkedIn and get in a big-tech company. It’s a win-win for founders and early stage VCs and a jackpot for shovel sellers like Nvidia and YouTube AI gurus.
Additionally, political leaders are non-technical, so they seldom know what they’re talking about. They’re hyping it up to gain soft-power.
All is not bad
Gen AI is incredibly useful, just not at replacing humans. It’s an incredible invention that allows us to transfer thoughts and retrieve data using language.
Also, bubbles are not an evil phenomena but just a side-effect of Capitalism. Humans are greedy, selfish, and irrational, and that’s why markets go up and correct and then crash and bubbles form and then burst, shaking down economies, pension funds, and livelihoods.
