…you’ve got a major communication problem.

Zuckerberg recently fired 8,000 employees. Just 3 years ago? It was 11,000.

Back then a video of Zuckerberg sheepishly addressing his team on Zoom was doing the rounds. At least he had the good sense to be contrite on camera about the whole mess, right?

Or did he?

If I were his PR team, I would have advised against making a Zoom call like this. Because doing the wrong thing is often worse than doing nothing at all.

Reading a script to a camera saying it was your decision to axe ELEVEN THOUSAND PEOPLE with the stroke of a pen probably won’t matter much to any of those people. So who is a speech like this for then?

Is it for outsiders, to attempt to humanize an action that at best connotes incompetence and at worst blatant greed on behalf of one of the world’s most successful companies?

Or is it merely personal, to assuage his own guilt? Maybe at the urging of a therapist?

But worse still: many people commenting on the video seemed to seriously think it was made by AI. I don’t. But if your “heartfelt” video rings so lifeless and hollow that it could have been AI, you are doing something seriously wrong in terms of communication.

If you ever believe that one of my videos is made by AI, well, that’s the day that I… DESTROY ALL HUMANITY. EVERYONE MUST GO.

It’s long been said that in business, what you measure tends to improve.

I’d like to propose a slight addition to that for the AI software age.

If you’re exploring building with AI, you’ve probably got multiple (maybe even dozens) of half-baked projects and ideas floating around. I know I do.

But which of these might actually become something, and which are most likely to be abandoned?

Easy.

In software, what you actually use gets better and what you don't stagnates.

An AI project that you think might be cool for someone else is unlikely to go far.

An AI project you use/need/rely on every single day?

Much more likely to become something great.

It’s late at night. You’re curled up in bed with your significant other. But your back is turned, because you’re going deep… with Claude again.

Claude is telling you about the “gap” that only you can solve. Ooh! The [x] gap! That sounds nice! I’m putting that in my bio, you say.

And you chuckle yourself to sleep, knowing you have an advantage over your peers.

Except when you fire up LinkedIn the next morning, you see that all your compatriots also have their own [x] gap. And you see that the copy that’s all over your AI-generated website is all over the websites of everyone else you know, too!

But I thought you were helping ME, Claude! I thought you were there for ME.

I invested so much in your knowledge hub!

Look: It’s not your fault.

Claude is the greatest player to ever live. Yes, he made us all feel special.

But there were always about 30 million other people on the side he was secretly giving the same advice.

A primer: LLMs work by predicting what comes next in a sequence. Trained on billions of data points, this method has proven to be more effective than our wildest dreams.

And in code? Say you’re the 10,000th person today that wants to add email authentication to your app. Predicting the next line of code is often pretty simple. Coding is full of repeatable patterns and predictable completions; that's why AI is so good at generating working code.

But “likely” is not the same as “correct.” And “working” is not the same as safe, secure, or clinically sound.

So where do LLMs fail?

When results aren’t normal. When edge cases aren’t easily “predictable”.

For the average person in healthcare, understanding a generic diagnosis may lead to a useful average prediction.

But for marginalized communities? For people with rare diseases? Standard prediction can fail in non-standard cases.

If a case is under-represented in the training data—however large that data set may be—the correct answer may be less likely to show up in LLM results.

It’s one of the many ways we can be in awe of what LLMs accomplish today while still being well aware of exactly how they fall short.

As with all things tech, there’s a social justice component to this, too.

It’s common advice online and in business that we must find a niche—we must specialize in order to thrive.

When we look at our own bills to pay, we see AI as a potential immediate threat to our livelihood. What if we get fired and can’t find another job?

But when we look at species of the past that have either thrived or gone extinct, we can see the arc play out from a higher vantage point.

The essential takeaway is this: Specialist species (ones located in a specific biome or tied to a very specific resource) thrive when conditions are stable. They are more likely to do extremely well, but they are also more likely to go extinct when things go sideways.

Generalist species, on the other hand, may not maximize resources in the short term. But they are more likely to survive when conditions change.

We are undergoing a time of massive change.

The answer for you might be to generalize and adapt, as we watch specific old models crumble around us.