Luck

Publishing Note: My editor says that this article is somewhat disjointed and should have focused on a single theme.  And I get that.  On the other hand, life is somewhat chaotic, and issues and topics in one area inevitably spill over into different areas.  I would appreciate your opinion.   I could a) continue to write interconnectedly on the full stack of topics from the personal level and feature level up to national policy, b) break up my posts by topics and do posts that are each a little more focused, c) quit writing about some issues, or d) something else!  Let me know your thoughts.

Luck

Seth Godin published a note recently — there are lotteries happening all around us, every day, we are all at the mercy of luck.  This resonated with me — I have been incredibly fortunate in my career.  Yes, I have worked hard and tried to work smart, but there has been a significant element of luck.

I was going to be an electrical engineer, but I stumbled upon the Computer Science minor option at Ohio State, which I had not even known existed. My college years coincided with the birth of the PC industry, an incredible economic opportunity.

In the early days of the PC wave, I took a job at Microsoft; I could just as well have taken a job at WordPerfect, Lotus, Ashton-Tate, or any of the many other early large software companies, most of which failed to thrive.  

I had two job offers at Microsoft, one in languages and one in networking.  Either would have been great, but I chose networking for somewhat random reasons, and that set me up well for the internet wave.

I met great people at Microsoft, who have been mentors, business partners, and friends to this day.  

I undoubtedly have some survivorship bias.  But even in a world dominated by luck, I stand by these principles:

• Take lots of small chances with contained downside.  Have lunch or coffee with lots of people; try a new task or skill; always dig in and learn something new.  The only thing I have to lose is a little time; the upside is unknowable.
• Align my bets with those of other great people.  Work with the most intelligent and highest-integrity people I can find.  
• When I see a rare chance where my effort can bias the outcome, throw myself into it.
• Realize life is a lottery, and be gracious and helpful to everyone I meet.  

What a time to be alive

GPT-5 is out and the reviews from people I follow are good — Ethan Mollick, Latent.space, Zvi Moshowitz.  

I’ve always felt that my late grandfather lived through a lot of change.  He was born in the 1890s on a small farm in Ohio, where he grew up with horses and carriages. We still have the sleigh bells his family used on their horses.  During his lifetime, he witnessed the introduction of automobiles, flight, radio, television, nuclear power, the moon landing, and computers.  All driven by advances in basic sciences, energy, and industrialization.  And through it all, he remained engaged, upbeat, optimistic, and interested.  

It feels like my lifetime is giving his a run for the money.  I’ve lived through semiconductors, PCs, the internet, mobile networks, the cloud, AI, space commercialization, CRISPR, mRNA, and more I forget.  It seems the pace is not slowing down.

It is a great time to learn about new things and not be too encumbered by the past.  As a friend said recently, while discussing Claude Code, software development is fun again! I am experiencing the same sense of wonder and empowerment that I felt when I first got a PC.  

Runtime AI follow-up

I was fortunate to have some people read my last post on runtime AI, and I got some great comments on this post, as well as some snarky comments.  

One flip comment was, “This is a great approach if you hate money.”  I do not hate money.  This may be an expensive approach today.

But what do we believe will happen to the cost per inference over the next 5 years? In the 2023-2024 timeframe, we saw a decrease of ~150 times in the cost per inference with subsequent ChatGPT GPT-4 releases.  ChatGPT costs have increased since then, primarily due to the models' remarkable improvements.  Dramatically lower-cost models have now emerged — the Chinese open-source models, and GPT-oss, just last week — which have costs per token of just a fraction of ChatGPT's costs.  I can’t say what the curve looks like exactly, but it is inevitable that the cost per inference will drop dramatically in the long run.

And even at today’s prices, I am certainly willing to pay something to avoid dealing with the complexities of HTML, JavaScript, CSS, React, Webpack, and so on. This tradeoff is only going to become more favorable towards AI.  Sam McKelvie’s comment, to push the AI processing to the edge, makes this even more compelling. Inference on my phone or PC is “free”; I’ve already paid for that capacity fully.  I am very interested in exploring this idea with Sam.

At some point, the cost of inference is so low, you’d have to be a masochist to keep writing the code yourself.  

Another class of comments focused on the shortcomings of a fully AI app — “Too slow, too hard to debug, wasteful of energy, security issues.”  And all these points are valid.  I am reminded, though, of past platform transitions.  PCs were too slow, were too limited, were toys — you could never do real business computing on them — until you could.   HTML apps were too primitive, too slow, lacked expressiveness, and were too flaky for real apps — until they weren’t.  

Moore’s law grinds inexorably.  It is not a time to be cautious and cling to the old ways of building software.  It is time to imagine how dramatically things could change if we had pervasive, cheap AI.

The US and luck

The US, as a nation, has been fortunate.  And it has encouraged this luck by making many modest investments in research.  Semiconductors, the internet, CRISPR, space launches — all due to early government investment.  It is hard to fathom why we would cut these investments — we should be increasing them and expanding into new areas.  As David Patterson points out in the article, the benefits are overwhelming:

All told, American taxpayers invested just under $100 million in the labs I helped lead. Accounting for inflation, technologies that came out of them went on to generate over $1 trillion in product sales. That is a 10,000-to-1 return on investment to the public and surely at least 1,000-to-1 return directly back to the government in taxes.

10,000-to-1 return on investment!! Where else can you find deals like that? It is irresponsible, bordering on malfeasance, to cut these investments. I agree with Patterson; as a taxpayer, I am livid.

Organizational Lifetimes

I have some empathy for the idea that we should sunset some government activities.  I was reminded recently of a quote from Steve Jobs in his somewhat famous graduation speech at Stanford — 

…death is very likely the single best invention of life.  It is life’s change agent.  It clears out the old to make way for the new.

We all have an effective working life of 40 to 50 years, during which we accumulate knowledge and power, utilize that knowledge and power to accomplish things, and then pass on what we can to the next generation.   And this all works out pretty well for human society.

It is odd, then, that we allow our organizations to have unbounded lifespans — companies, schools, governments, churches, unions, political parties, and so on.  Organizations accumulate power, become bureaucratic, and shift their focus from serving the community to maintaining their power and control.  They become a drag on innovation, disruption, chance, and luck.

One wonders if we wouldn’t be better off with forced finite lifetimes for all these organizations.  Let the next generation reinvent them and reconstitute them as needed.  The principles underlying organizations may be long-lived, but the specific form of the organization need not last forever; in doing so, it is likely an impediment to progress.

So I understand the desire to dismantle existing federal bureaucracies. The failure is not in dismantling these bureaucracies; rather, it lies in the failure to put in place a plan for their reconstitution. There is much to criticize about our current research system, our education system, and their results. But we don’t need to eliminate the Department of Education or the NSF — we need 2.0 versions of them, reimagined and revitalized.

Shorts

Stock Compensation: I was lucky to join Microsoft in the late 80s; the company was committed to meaningful sharing of stock ownership with employees.  This encouraged cooperation throughout the company — everyone shared in the company’s results.  (It helped that the company was doing well — with everyone doing well with the stock, there was less value to infighting).  

I’ve been at large companies where stock ownership was not distributed. The companies invent metrics for people to meet, and people try to game the metrics; there is a bit of zero-sum thinking about promotions and compensation. This is not a pleasant or productive environment.

There is a new bill in Congress that would reward companies that give stock to rank-and-file employees — I couldn’t agree more with this.  There is no good reason for employees not to be shareholders in the companies at which they work.

Electrification: I’ve read several articles that attempt to disentangle renewables and electrification.  It is worth untangling these.  Electrification has benefits all by itself — separating demand from supply with a neutral intermediary, which allows more dynamic switching of supply.   A great article here on the status of electrification.

And a good article on the nearly unlimited amount of solar power — we are not limited in the amount of energy available to us; our limits are due to other resources and costs.

Pocket Endoscope: From Recomendo, so cheap, I feel stupid not buying one.