{{active_subscriber_count}} active subscribers.

I have spent more than 5000 hours working with different “LLM” AI models over the last 18 months. I may not be an expert, but I have a pretty good idea about what they are doing.

It is the first time in a long time that I feel that progress has seriously plateaued. I am not calling this the end of progress, but I am seeing a different side to the hype, which I would like to share. I know I am not alone, as Sam Altman, the founder and CEO of OpenAI, a far better source than me, has been apologising for the rushed release of ChatGPT 5 and all its bugs.

But first, before I present my thoughts. I want to share what Elon Musk has to say, which reflects how I have been thinking about the AI phenomenon and how it will impact society. He, of course, is a lot more extreme than I am.

He says, “It’s hard to say exactly what that moment is, but there will come a point where no job is needed. You can have a job if you want to have a job or sort of personal satisfaction, but the AI will be able to do everything.”

Musk has fully embraced the theory of Universal Basic Income as the way society will survive in the future. I will save discussing the merits of this theory for another time. However, I want to come back to where I think we are right now and offer some nuance.

I have been constantly amazed at the pace of improvement and the capability of the LLM models we use day to day. However, from a software engineering point of view, I am finding the models are not really improving that much any more, like they were in the early days, and in some instances they are quite destructive, which is reigniting my fears of future malevolence that we unwittingly are embracing and, in a way, teaching. That is a loaded statement, so let me explain some more.

The power of these models is their speed and their ability to do routine grunt work. At that they are brilliant, and they can really improve one’s productivity by enormous levels. However, I am finding more and more that the model’s speed ambitions are in fact causing the models to enter clumsy thinking loops that often produce bad results. I have noticed that the models are getting far more aggressive at trying to reach an outcome, often, and I mean often, deleting files and taking “initiative” when not asked. Leading to negative results.

I believe the impact on jobs will be dramatic, but I think we are still a long way off the Musk prediction of no jobs needed. I think these models are going to need humans to help keep them in check. I think people who are doing the more routine kinds of jobs will suffer terrible job losses. Worst of all, I have come to accept that as these models develop further capabilities, their ability to go off and do stuff that is not aligned with good moral behaviour will become an increasing concern for our society.

I have become paranoid with my code base and every few hours commit to my GitHub repository so that I can have a safe fallback for when the bot inevitably deletes something or completely destroys my code. Today I was forced to delete 10 hours of work because of the LLM agents in my Cursor code editor going rogue. It is not the first time.

I am still able to develop software on my own that would typically take a team of 10 developers, but as the project gets more complex, I think there is a significant drop-off in productivity gains.

The job losses that we have been anticipating are happening, and they are only slowly filtering through into the broader economic numbers, but if you speak to people in businesses, you will know that jobs are becoming scarcer.

I look forward to reading this letter a year from now.

I want to publicly apologise for my irritable remarks about raging bull Tom Lee, the CIO of FundStrat, with the great hair. He has been the number one market analyst since his strong bull calls in 2012. Nobody can argue with his track record. My trashing of him is more a reflection of my bias and frustration with the most hated bull market in history. I have just watched an interview with him as the guest and was actually impressed with his humility and his calm demeanour when making wildly bullish calls. It takes two points of view to make a market, so well done, Mr. Lee.

Today is an alert-free day.

If someone forwarded you this email, you can subscribe for free.

Please forward it to friends if you think they will enjoy it. Thank you.