333

It feels like yesterday when the title was 222. Actually, it was 111 weeks ago, on August 18, 2024, when the 222nd Recurrent Patterns issue went out.

I am amazed now, as I was amazed then. I wrote 333 weekly articles, but most importantly, you were reading them. I thank you all for that!

Perhaps it is excusable to reflect on the last 111 posts. And I think it would be fun to ask AI what it makes of all the posts.

I pointed AI at the website and commanded it (I never ask) to go back to post 222 and analyze my articles.

Here are the tidbits:
- I wrote a lot about AI. Out of 110 posts, 99 mention AI. — There is not much argument here. The last two years were AI madness, which overshadowed any other tech news for better or worse
- It is obvious that I am skeptical and prefer strategy over hype — Yes, even AI can see that!

True to the name of this publication, several recurrent patterns keep repeating:

• LLMs are Large Language Models and not Knowledge Models

• Technology is a tool and not a strategy or moat

• Me, mocking vendors’ disclaimers

• Little tolerance for hype

I’d say that’s accurate as well.

The next section was about the people companies I wrote about:

• OpenAI/Sam Altman — rudderless, no strategy

• Google — monopoly

• Microsoft — graveyard of ideas

• AWS — Alexa and evenhanded coverage of its activities

• Meta — digital pimp

• Tesla/Musk — unkept self-driving promises

I must admit, when you summarize over 100 posts like that, it feels very accurate, and the patterns emerge.

When it comes to trends over the last 2+ years — at the beginning ‘mostly mockery of product launches’ and in 2026, focus on money: ‘circular AI financing, OpenAI’s losses, IPOs, pricing and ads.’

I’d say it sums up the last 2 years pretty well.

But then things got more interesting. According to AI, my posts contain numerous predictions.

Where were my calls the most accurate?

• Tesla — missed deadlines for unsupervised self-driving

• Google’s antitrust case: it dragged on, and Chrome wasn’t sold.

• AI companies’ pricing power (better say lack of): OpenAI cut some model prices by up to 80% in July.

• AI browsers: they didn’t unseat Chrome

• Microsoft: Copilot and bundling played out. Microsoft and regulation calls are almost all hits.

Clearly these were easy wins. Anybody who’s betting that any product Mr. Musk promises will be late is a guaranteed winner. Google knows how to play the monopoly game. Any startup trying to win the browser war smokes bad stuff. And Microsoft? Nothing has changed since the early days of MS DOS.

But then things got heated.

Things I missed:

• My claim that LLMs are circus monkeys. The proof is that AI now scores high on Frontier Math (and I would add the solved the Navier-Stokes Millennium Prize Problem and it is at the PhD level. That, I would argue, is highly questionable. I would bring your attention to this post, Teaching an old monkey new tricks, where it was later revealed that Epoch AI, when creating these tests, attached this footnote: ‘We gratefully acknowledge OpenAI for their support in creating the benchmark.’ When you train your model on a test where you have the inside track … your pet monkey can pass the test.

• Self-driving is a myth: ‘true for Tesla, but Waymo gives about 500k paid rides a week.’ Agree. Waymo is racking up miles and miles. It must be my personal bias when driving through downtown Vancouver. I have no idea how self-driving gets through that or the roundabouts in Europe.

• Meta’s AI glasses: ‘they are selling.’ Yes, it got so popular that Meta had to introduce AI glasses without the camera. It is also such an amazing piece of technology that Mr. Zuckerberg, in every public appearance, is spotted with them — not. #chokingonirony

And the (AI) jury is still out on OpenAI shrinking to a minor player, LLMs never becoming reliable and the AI industry shakeout. We will see about that; I am firm on these.

I hope you find this trip down memory lane interesting. I did. I would add that the AI answers proved my point. The LLM is good at going through a large corpus of content and can reasonably accurately provide a summary and assessment. Its weakness, at least in my opinion, is assessing the accuracy of the content. Why? It can only provide this information based on the training data. When the data is inaccurate, the assessment will reflect that. Which ironically aligns well with my take on this.

I got asked why I originally picked 222 for the celebratory issue and not 200. No reason; One day I realized I had already passed the 200 mark and the next ‘nice number’ is 222, so I went with that. And because this is a recurrent patterns publication, I couldn’t revert to 300!! Hence the 333. I used the term ‘nice number,’ and the other day I was reminded about numerology. I was curious if 333 has any significance. I learned that it ‘signifies harmony across different facets of your life, indicating that your thoughts and actions are aligned,’ and ‘The number 3 represents joy, optimism, and self-expression. Tripling it in 333 amplifies this energy into a bold push for forward movement.’ There you go — my thoughts and actions are aligned. Who knew!

Thank you all for reading, and for your feedback and conversations. I guess we have no other choice than meeting again at issue 444 — ‘viewed as a sign that you are surrounded and supported by guardian angels or the universe’ and ‘It encourages disciplined effort, methodical planning, and persistence rather than wishful thinking alone.’ — if that’s not a recurrent pattern, I don’t know what it is.

Cheers,

Vaclav