David's Self Reflective AI Question?

AI
1

Hive memory? I’m hoping the group will be my Google Gemini for recent Focused/MPU Podcasts.

In a recent(ish) episode of one Focused or MPU, @MacSparky mentioned he asks an LLM to “ask five thoughtful questions” about his writing and/or journaling. I like using LLMs for better self reflection but can’t remember anything else David said.

Does anyone remember what he said or which episode?

2

Possibly 805? The Most MPU Possible? I listed to this one recently (among a couple others) and recall David talking about it in one of them. This seems most likely to me upon a cursory look at the notes: Mac Power Users #805: The Most MPU Possible - Relay

3

Thanks I downloaded and ran it through MacWhisperer. No such luck.

Focused 234 - Intentional Friction has a quick mention: “ask me five to ten reflective questions on this”. Somewhere I thought he said more than this.

4

Oh. I listened to that last week too. That checks out. But yeah, I have a similar ghost memory. I think he talks about what he does with ChatGPT’s responses as well.

5

Definitely MPU, as I remember it too and I don’t listen to Focused. Sorry that’s as much help as I can give :grinning:

2 Likes
6

He does talk about it in episode 227 of Focused at about the 12 minute mark.

1 Like
7

I don’t know the Episode, but David said he was using the Apple Private Cloud LLM via shortcuts as he was pushing his journal to the LLM and asking it to ask him questions about the entry.

Understandably, David didn’t wish his private journal entries to be used to train someone’s model, so the Privacy of the Apple Cloud version was preferable.

1 Like
8

Yes. That matches my memory. I suspect it’s a segment about the betas.

9

Good shout, I forgot that was a ‘26 feature.

10

Well played. The comment was: “Ask me, you know, five reflective questions on this, and just see what happens”.

I guess it might be time to ask @MacSparky to give us an example with entirely fake information.

@geoffaire Apple private cloud is one approach for privacy. The other approach that I often use, run models locally. The current qwen3-30b-a3b-instruct-2507-mlx is very good.

1 Like