Video: Inside the Machine: What LLMs REALLY Think About Your ‘Thoughtful’ Questions

1–2 minutes

Chatting with Claude Sonnet 4.5 was such an interesting experiment, so I created a YouTube video version based on the Spotify version. If you’ve already listened to it, feel free to check out the video content – the audio hasn’t changed.

Video: Inside the Machine: What LLMs REALLY Think About Your ‘Thoughtful’ Questions

I feel that the explanation of some of Claude’s internal logic was telling, and how it is anthropomorphised in a way that a person might interpret through an emotional lens.

Personally, I also enjoyed the dialogue around Platonism as it related to maths. I updated the subtitles, so you can read along if you are so inclined.

I’d like to do more videos, but they take so much time. I don’t know how much total time this took, but it was many hours over three days. It’s not that I don’t want to take time to produce them; it’s the opportunity costs – I am not writing new material, which is my preferred activity. For the record, the bulk of the time is searching for appropriate stock footage and B-roll – and that’s not always successful either.

I generated a few clips in Midjourney – sometimes just because, and other times to fill a gap with something better than I could find on Motion Array.

I’ve embedded the video here as usual, or you can watch it on YouTube. In any case, I’d love to read what you think about the topic or the video. As for the video, I won’t be giving up my day job, but it’s fun to assemble them.

Artificial Intelligence Isn’t Broken

Rather than recreate a recent post on my business site, LinkedIn.

(Warning: contains traces of logic, satire, and uncomfortable truths. But you knew that.)

Audio: NotebookLM podcast on the linked topic.

It’s just refusing to cosplay as your idealised fantasy of “human” cognition.

While pundits at the Wall Street Journal lament that AI thinks with “bags of heuristics” instead of “true models,” they somehow forget that humans themselves are kludged-together Rube Goldberg disasters, lurching from cognitive bias to logical fallacy with astonishing grace.

In my latest piece, I take a flamethrower to the myth of human intellectual purity, sketch a real roadmap for modular AI evolution, and suggest (only partly in jest) that the machines are becoming more like us every day — messy, contradictory, and disturbingly effective.

Let’s rethink what “thinking” actually means. Before the machines do it for us.