Hi, I’m Yves. As we wait to see what happens next in the escalating Middle East situation, I thought it would be useful to take a look at some key real economy issues, a big one being the outlook for AI, and in particular OpenAI.
Reviewed by Ed Zitron, His compelling case against OpenAI In a weighty post last week (estimated 31 minute read), I wrote: His arguments are so multifaceted, detailed and well documented that I worry that a summary here will not adequately address his voluminous work. Therefore, I urge anyone who disagrees with Zitron’s case to: Read his post To see the obvious drawbacks is because I had to cut out large parts of his arguments.
Before moving on to Zitron’s compelling criticism, the fact that AI’s usefulness has been overstated does not mean that AI is useless. In fact, AI can have applications in small business environments. The big fuss a few months ago about AI posing a danger to humanity was to justify regulation. The reason is that AI advocates have woken up to the fact that AI has no barriers to entry. Even the smallest players can come up with useful applications based on very small training sets. Imagine a professional services firm that uses AI to create boilerplate letters to clients.
Some hedge funds are developing much more advanced applications of so-called black box trading. I confess that I have never seen performance statistics for the various strategies (so-called quantitative vs. “event-driven” strategies, merger vs. market-neutral vs. global arbitrage, and several other varieties). But I cannot think of any that regularly outperform, much less an AI black box. If there had been any success in this field, it would have been covered extensively in the press.
Returning to Zitron, he portrays OpenAI as the mother of all problems, saying it has to do a ton of impossible or near-impossible things to survive. Remember that fatal cumulative probability calculation that applies to young startups? If a company has to do seven things to thrive, and each one has a 90% chance of success, it’s a winner, right?
No. Get out your calculator. .9 x .9 x .9. x 9 x 9. x .9 x .9 = .478, which means your chances of success are less than 50%.
He also compares OpenAI to Uber, very unfavorably. I have to take issue with his generous portrayal of Uber as fulfilling a consumer need. That portrayal becomes suspect when you realize that Uber is essentially a high-cost provider with no barriers to entry. Uber’s popularity is largely due to investors heavily subsidizing ride fares. What’s not to like if you can get a truly low-cost service?
One mistake we made in our analysis of Uber was not viewing it primarily as an investment. In the 1800s in the US, railroad companies were founded one after another, some with directly competing routes. Yet more new railroad companies laid track, despite almost inevitable bankruptcies. Why? These were stock market transactions (some say frauds), and many investors jumped in despite their record of failure.
Uber and other recent unicorns saw their value inflated dramatically due to shoddy valuation procedures by venture capital investors. These investments look much more attractive than they actually are..
Abstract from Zitron’s paper:
In order for OpenAI to survive beyond two years, I believe the following needs to happen (in no particular order):
- Navigate a complicated and troubling relationship with Microsoft that is both a lifeline and a source of direct competition.
- We’ve raised more money than any startup ever has, and we continue to raise at a pace never seen in the history of fundraising.
- We will achieve significant technological advances that could reduce the cost of building and operating a GPT, or its successor, by thousands of percent.
- GPT has made such significant technological advances that it can address new, never-before-seen use cases that are currently not possible or even assumed to be possible by artificial intelligence researchers.
- These use cases will create new jobs while at the same time fully automating existing jobs and justifying the huge capital expenditures and infrastructure investments required to continue doing so.
Ultimately, we believe OpenAI in its current form is unsustainable: there is no path to monetization, the burn rate is too high, and the technology, known as generative AI, requires too much energy for the power grid to maintain. Additionally, training these models is similarly unsustainable, both due to ongoing legal issues (due to theft) and the amount of training data required to develop the models.
And, simply put, any technology that costs hundreds of billions of dollars to prove its worth is built on bad architecture. There is no historical precedent for what OpenAI needs to achieve. No one has ever raised the amount of money needed, and no technology has ever required such massive amounts of money and systems power, such as rebuilding the US power grid. survive, Moreover Prove that the technology is worth such an investment.
To be clear, this post focuses on OpenAI and not on generative AI as a technology. However, I believe its existence is essential for companies to continue to be interested in and invest in the space…
My point here is not to say that OpenAI will definitely collapse, or that generative AI will decisively fail. My point here is to provide a sobering explanation for why OpenAI, in its current form, cannot survive more than a few more years absent an incredible blend of technological breakthroughs and financial magic, some of which are possible, but much of which is historically unprecedented.
Zitron begins by examining the opaque but apparently messy relationship between Microsoft and OpenAI and how it impacts valuation. This is a bit too much for the general reader, but it’s informative for both the tech and finance industries. This part is necessarily a bit difficult, so I encourage you to read Zitron’s post. Read the full story.
The discussion now turns to the issue of funding. The key points here (emphasis in original):
Assuming everything exists in a vacuum, OpenAI needs at least $5 billion a year in new capital to surviveThis will require raising more capital than any startup in history has ever raised, likely in perpetuity, and therefore access to capital on a scale unmatched by any other company in the history of business.
Zitron lists a small selection of companies that have recently raised large amounts of funding and argues that OpenAI is by far the biggest money-sucker, simply in terms of the rate at which it will burn money and the expected duration of the burn.
He then delves into the profitability, or lack thereof, exacerbated by what would previously have been called the build-out problem.
As I’ve written repeatedly, generative AI is highly unprofitable and, based on Information’s estimates, its cost of sales is unsustainable.
OpenAI’s costs will only increase over time, as will the costs of making these models “better,” and, in the words of Goldman Sachs’ Jim Covello, they have yet to solve a complex problem that would justify the cost… Since November 2022, ChatGPT has become more sophisticated, its generations have become faster, and it can ingest more data, but it has yet to produce a true “killer app” like the iPhone.
Furthermore, transformer-based models have become heavily commoditized, resulting in a price war that has already begun…
As a result, OpenAI’s revenue may increase, but it will likely do so by reducing the cost of its services, not its own operating costs…
As mentioned above, OpenAI, like all other Transformer-based model developers, needs large amounts of training data to make their models “better”…
Doing so would likely lead to permanent legal action…
And, frankly, I don’t know if there is enough training data To pass these models on to the next generation, even if AI companies were free to legally download any text or visual media from the internet to train these models, it doesn’t seem to be enough…
And there’s a huge, troubling problem: generative AI doesn’t have the product-market fit at the scale necessary to support its existence.
To be clear, I’m not saying that generative AI is completely useless or that there’s no product-market fit at all…
However, at present, necessary.
Generative AI, you It is absolutely necessary to integrateThis wouldn’t be a problem, other than the feeling that the company would be “behind” if it wasn’t using AI. If the costs of running generative AI were a tiny fraction of what they are today, say tens or hundreds of thousands of percent, but as it stands OpenAI is effectively subsidizing the generative AI movement. nice and usefulGPT is only changing the world as much as the market will allow.
He still has a lot to say on the subject.
Oh, and that’s before we get to the little issue of energy, which he also analyzes in depth.
He then goes on to explain again what OpenAI needs to do to overcome this obstacle, and why it seems extremely impossible.
Again, if OpenAI or AI in general is a topic of interest to you, Be sure to read Zitron’s postAnd be sure to distribute it widely.
