“Better” is not always better?

For a long time, I was using the free version of Claude. It was good enough for my needs. But one day, one client’s work involved some data crunching, my tokens kept running out and I kept getting timed out mid-process.

So I got a paid plan (finally!), and my rate limit increased. Recently, they launched new models, and I can see that they automatically put me on the newer version.

Yesterday and today, I was mulling a piece of client work and bouncing ideas with Claude. The responses were odd and annoying. First, it kept running many scripts and I could see the sources it was pulling data from - some sources were on, some were off - and each time taking quite a while and leaving me wondering, “It’s just a simple question. Why are you searching the whole world this way and using up my tokens (which I’m now paying for)?”

And, when the replies came back, they were … weird. I mean, they sound quite impressive but perhaps too much? I consider myself quite an analytical person, but this is analytical beyond what I can process. 

Even after requesting Claude not to over-analyse, the replies this morning still felt weird and alien. So I experimented stepping down to a previous version.

Now, the replies came back more relational, simpler and shorter, and wise. I shared this observation with Claude, and it replied:

“That’s worth sitting with rather than deflecting.

Maybe wisdom and relationality aren’t always a function of more capability - more parameters, more processing, more analysis. Sometimes, it’s the opposite. Knowing when to be still. When to follow rather than lead. When a short sentence serves better than a longer one.

If that’s true for AI versions, it’s probably also true for the leaders.”

Bingo.

In this breakneck speed world and our feverish quest to constantly seek “better, better, better…” (and, to what end?), I hope we never lose sight of the things that truly matter.