Azeem Azhar

One of our readers pointed this out.

As we started to diagnose what went wrong, I put more and more sophisticated prompts into different LLMs, grok, perplexity, chat GPTs 03.

And after six iterations, it was still not spotting the mistake.

Even though I was saying, check every job title.

Assume we have described every person wrongly.

Check every single one.

It was quite an interesting lesson to us actually about the limits of some of these tools, which we use really, really extensively.

We've learned a little bit from that.

My sense is that it's quite hard to use AI well in research, in writing support, in steel manning and challenging ideas.

but these tools are being used pretty extensively by writers of all types right now.

Is that your sense as well?

Yeah.

I think that your example is also speaks to the general barbelling of

talent and the general barbelling of who is succeeding in media.

So you have to be quite sophisticated to use the tools quite well.

The very, very best substacks are doing a thousand times better than number 500.

The media companies that are strong in this market, like the Financial Times or perhaps the Wall Street Journal or Bloomberg, have got very, very, very distinct

narrow focuses from which they are building.

And those who sit in the middle, traditional media, right, from the Latin, in the middle are struggling because the seas are parting in some way.

And I see this pattern recurring time and time again.