gctwnl

Part #6 of a series of lab notes on an experiment: how good is AI-generated code really? The code was not very good, but Claude's reaction to the audit wasn't all that bad.

By the way, there is caching (cache TTL is hort but during sessions it is active). E.g. Settings Status Config Usage Stats

  Session
  
  Total cost:            $15.59
  Total duration (API):  18m 3s
  Total duration (wall): 2h 13m 16s
  Total code changes:    232 lines added, 80 lines removed
  Usage by model:
       claude-opus-4-6:  1.5k input, 41.8k output, 26.4m cache read, 212.6k cache write ($15.59)

What is that, exactly?

Why was this flagged? The article is a serious attempt at observation and analysis.

That is interesting. $200/month for >$10k at API-pricing? That's way more than the 2.5x-12.5x the article observes.

Wait, what?

Yes, th enumber is based on subsidised subscriptions and he article makes that clear.

There are no revenue estimates in the article). Only as serious as possible estimates of cost per task.

The data is given and not made up. The data based on actual use and on analysis of public numbers.

The no-caching is explained in the linked conversation. The numbers are from actual use.

Short 3 minute excerpt of a talk on fundamental properties of IT as well as fundamental properties of humans that influence design decision making in complex (IT) landscapes. Only slightly tongue-in-cheek.

Why, if the author asks it to summarise a single webpage and gives the link should ChatGPT go out and load 5 more (one is the same page again, the others short overview pages, so won't have influenced the result much)

And why all this talk about trying to engineer a prompt so that in the end the result is good? Should an actual usable system not just handle "Please summarise [url/PDF]"? That is, I suspect, what people expect to be able to do.