But whether you're using Tana or not, the principle shown in this paper is really instructive: look for how you can tools to think better!
Full paper is here: www.nature.com/articles/s4...

In Tana, all of that hassle is taken away and you're freeing up lots and lots of resources to actually engage with your own thoughts, the AI's thoughts, and the information you've gathered.
(You should check out Tana, if I haven't been clear enough 😁)
try.tana.inc/0jcuk5m3ccr7

I've personally experienced and witnessed in others how quickly chatbots become an extension of a person's mind – and that is with all the hassle of repeatedly adding context to a conversation.

Now, how does this connect to AI and knowledge work?
With tools like @tana_inc that deeply integrate AI into your work, you can free up a whole lot of processing power and working memory to do better, more original thinking.
And we're still just at the very beginning of that!

In their terms, manipulating your environment to figure something out is an "epistemic action" – like when you reshuffle your scrabble tiles to make it easier to come up with new possible words.

So what might be going on here?
The authors say that using your external environment to solve a problem is a before-unmeasured element of intelligence.

The click+drag test was BETTER at predicting academic performance than the static test!

Much higher correlation with academic performance – it's clearly visible in the screenshot above!

If it is the case that dragging things around makes solving the matrix easier, what should happen?

Hypothesis: the click+drag Raven test should be WORSE at predicting academic performance, because dumber people can now solve the matrix too

Well...that is NOT the case!

In the click + drag version you get to sort the images as you please while you're solving the matrix.

Your first intuition might be: well, being allowed to drag things around is cheating! That makes the test much easier!

Let's turn that into a hypothesis...

But, back to what they actually wanted to show.
In psychology the Raven Advanced Progressive Matrices test is used to measure fluid intelligence.
They came up with two versions: a static version, where you solve the matrix just by looking at it, and a click + drag version.

Feel free to laugh at this next sentence in the same highlight – the paper is from 2019...

"Existing artificial intelligence programs never proceed by printing out intermediate results to repeatedly reinspect them."

YES THEY DO, IN 2025!
They are called...REASONING MODELS!

The reason we do this, they say, is that this unburdens our working memory, which in turns allows us to make more complex inferences – we have more "room to think" in our head, if you will.

How do they arrive at that conclusion?
Given that Andy Clark is a co-author, it's no surprise that they start with the observations that humans make use of their environment to think better all the time:

Conclusion up front: their contention is that it matters an awful lot how and that we use tools to augment our intellect

Any particular model you recommend?