Hi! In case you are new around here, I’m Om. In this letter, I share my latest thoughts, articles worth reading from around the web, my recommendations & occasionally, my photography. In this issue, I address the following:

• A quick personal update
• OpenAI & the hype curve
• Five good reads

The recent mini-break was refreshing and rejuvenating. It allowed me to reflect on my friend Sean Gourley’s wise words. Just as in fast food, if the time taken to consume information surpasses creation time, it’s not worthwhile. This notion was the backdrop of my writing hiatus and deliberate distancing from social media. 

My time and energy were devoted to diving into long-form journalism, perusing magazine articles, and whittling down my hefty reading list. Concurrently, I took neglected books off the shelf, and with a preference for the traditional fountain pen, I embraced the tactile practice of jotting down thoughts, nurturing ideas, and devising an editorial schedule.

While on this break, I put the finishing touches on the website. All my creative work sits here — my photography, essays, interviews, and now my recommendations – OMakase. I have a new and fresh approach to this recommendations-oriented blog.

It is good to be back. 

---

OpenAI & the hype curve

OpenAI, the belle of the ball in 2022, is already turning into a witch — or at least the subject of a witch hunt in the media. The plaudits have been replaced by bricks, echoing an all too familiar hype cycle in Silicon Valley. 

***

This past weekend, an obscure publication, citing data from another source (without context), came to a shocking conclusion: OpenAI, a company that managed to get $10 billion in investment from Microsoft less than a year ago, would be bankrupt in 2024. It soon made its way into other publications. Soon, AI promoters on Social Media were quick to rewrite the post and share it with their followers. No one bothered to pay attention to the fact that the original article lacked rigor around numbers. 

I am not privy to OpenAI’s financials, nor do I claim to be an expert on the company and its affairs. However, as someone who has written about startups, has a little grasp on the financial matters at companies, and has been an investor in companies — I read the article and came to a quick conclusion. I am being generous when I say that the article was of dubious factual provenance. 

As is my habit, whenever I read such articles, I try and do some basic back-of-the-envelope math to see if it passes the smell check. The original article — trust me, it is not worth linking to — assumed that the company spent $700,000 daily. That’s about $255 million a year — not the kind of burn to put a company that raised $10 billion from Microsoft on the road to bankruptcy. The $ 700,000-a-day estimate came from Dylan Patel, an analyst at SemiAnalysis, who estimated that it was about 36 cents per query for ChatGPT in an interview with The Information. The number was taken by Analytics India, which didn’t cite Patel or The Information. FirstPost, a website owned by Indian conglomerate Reliance Industries, reposted the Analytics India article before it was shared countless times on social media. 

***

When announcing the news of its investment earlier this year, Microsoft said it would invest in its supercomputing infrastructure to help OpenAI and be OpenAI’s exclusive cloud provider. Since then, there have been rumblings around Silicon Valley that Microsoft’s investment was part cash and part access to its Azure cloud — and the hard-to-get-by GPU resources. It is strange not to have read this in Microsoft’s filings if it is true. [As a historical aside, such deals were known as “vendor financing” and were commonplace during the Internet 1.0 bubble — Lucent, Nortel, and Cisco were well-known practitioners of vendor financing.]

Anyway, back to OpenAI‘s perceived financial doldrums. 

Earlier this year, OpenAI claimed it had about 100 million users — the fastest to 100 million before Facebook’s Twitter rival, Threads, lapped it. Most software-as-a-service experts believe that anywhere between 3%-to-8% conversion to “paid” customers is a reasonable conversion rate for freemium software services. 

If you assume OpenAI is converting at a mere 1% of its total users, that is still a million users paying $20 a month — or about $240 million yearly. This dovetails with information from a Reuters report that quoted an OpenAI investor presentation. OpenAI itself projected revenues of $200 million in 2023 and $1 billion in 2024. No matter how you dice this rumor, it is pretty clear OpenAI is not going bankrupt anytime soon. 

***

Nevertheless, the story did catch fire and become part of the conversation. It is the latest in a drumbeat of stories about a faltering OpenAI. There have been complaints about the quality degradation, especially of its API, on OpenAI’s message boards. However, a research report from a group of analysts sparked real media interest.

Researchers from the University of Berkeley and Stanford University took two-time frames — March 2023 and June 2023, to look at the performance and behavior of GPT 3.5 and GPT 4.o and concluded that “the performance and behavior of both GPT-3.5 and GPT-4 varied significantly across these two releases and that their performance on some tasks have gotten substantially worse over time, while they have improved on other problems.” 

The researchers published their results — which are very nuanced — in a paper earlier this summer. The media machine went apocalyptic and ran with the “OpenAI is degrading” narrative. The media attention took mere two seasons to go from flattering to faltering. 

This is a familiar story — a technology-related media machine has a certain predictable playbook. I have seen it repeat itself many times over. Aaron Zamost, a one-time PR guy from Square (nee The Block) and Google, wrote what I think is a seminal piece about the media hype cycle from the “pr person’s” perspective that is more colorful than my summary. It is worth reading.

I am writing from a media guy-turned-investor perspective. I have written about many companies — most of the time with measured enthusiasm. If you have spent three decades in the world of startups, as I have, you are also likely to be more wrong than right. 

A new startup comes along, touting a brave new idea that it claims will change the world. Sometimes the hype is real —Google, Facebook, Airbnb, Instagram, Square, Tesla, and Uber are good testimonials. The media buys into the idea. The origin story is just too good to pass up. The founders are young and idealistic. The news coverage is generally positive and starts to ramp up, much like Hollywood press writers about new starlets and rock press writers about newly discovered musical genius. 

The startup becomes the hottest thing in the valley, goes on an expansion spree, and is labeled the greatest (new) company in the process. And then, just as predictably, doubts about the product prop up. There is talk of management and internal turmoil. This is the arrival of the FUD phase — Fear, Uncertainty, and Doubt. It won’t be long before the company is called the worst company in the world. (It is true in some cases, such as Facebook and Juul.) 

The media machine moves on to a new hero, a new idiot, and a new villain. Rinse and repeat. 

***

It shouldn’t surprise anyone that time and patience are as much a virtue in Silicon Valley as they are in the real world. Sure, with plentiful capital around these days, it might not take much time for an idea to become a startup. There are no shortcuts for a startup to become a viable, self-sustaining business. No one can avoid the vagaries and friction of a very human world. The challenges of managing people. The indecision over purchasing decisions. Real or unreal fears about the future guide how the world goes, and technology companies are no different. The long arc of time and a set of fortuitous decisions define whether a startup will become Facebook or a Friendster. 

It is too soon for anyone — including OpenAI CEO Sam Altman — to know what the future holds for the company. It is remarkable that in less than ten months, it has caused everyone, including Google and Apple, to reevaluate their future. Large Language Models (LLMs) made the rounds in academic circles. Now, you can hear late-night television shows make LLM jokes. 

Media machine, however, needs to feed the Internet monster. Simple narratives such as OpenAI (or any other company) are degrading or bankrupt, delivering the dopamine hit and fractional attention for even more fractional advertising dollars.

---

Five Good Reads

• How to understand and get the real truth from a privacy policy?: The Markup has published this primer, which is a must-read. I think it is “bookmark worthy.”
• The Secret Life of the 500+ Cables that Run the Internet: if there is one thing I miss from my time as a journalist, it is writing about this stuff. This is such a fantastic read for a network nerd like me. 
• How did Thomas Edison trick the press into thinking he invented the light bulb?
• What exactly are the economics of AI?
• Catching up with the weird world of LLMs: This is as elaborate and in-depth as you can get on large language models. 

---

Etc: 

• A walk by the water and fog inspired some photography featured in a short visual essay: Fine Fogust Finds. And what do you know, my seagull neighbors from 2020 have returned with a new baby. 
• I just celebrated the 10th anniversary of becoming an American citizen. How lucky am I to celebrate two independence days — July 4th and August 15th? Happy independence day to all my family and friends from India. 

---

August 15, 2023. San Francisco