Beyond Compression: AI for Deep Thinking
Key Points
- Most people use AI mainly for compressing information—turning notes, long documents, or articles into concise summaries—rather than for deeper cognitive engagement.
- The brain processes compressed content differently, so relying on AI-generated summaries can limit the formation of new mental connections and the transformative learning that comes from prolonged, focused study.
- The problem isn’t AI itself but how we use it; for complex, high‑impact tasks we should allocate more “brain time” and use AI to reduce cognitive load, allowing us to think more deeply about the core subject.
- AI can support deep work by handling routine compression tasks while we focus on expanding our understanding, as illustrated by the speaker’s challenge of writing a book on AI.
- Mastering prompting is valuable but akin to learning to ride a bicycle—useful for quick tasks—where learning to “drive” AI as an intellectual partner offers greater long‑term productivity.
Sections
- Beyond Compression: AI Partnership - The speaker argues that using AI merely to compress information short‑circuits deep learning, and proposes leveraging AI to offload routine summarization so we can spend more brain time thoughtfully engaging with substantive content.
- Prompting as a Bicycle Skill - The speaker likens mastering AI prompting to riding a bike—an essential, efficient skill—and explains how a conversational voice AI served as an interactive partner that kept ideas flowing, rather than merely transcribing them.
- Choosing AI Models as Cognitive Partners - The speaker outlines how they evaluate and switch between AI models like Claude Opus 4 and Gemini 2.5 Pro to deeply critique outputs, clarify their core ideas, map their conceptual terrain, and manage partnership dynamics, noting a Substack guide they authored for model selection and their $200‑per‑month OpenAI subscription.
- AI as Thought Amplifier - The speaker argues that thoughtfully selected AI tools, such as Opus 4, can deepen understanding and improve cognition when used deliberately rather than as a shortcut for mental effort.
Full Transcript
# Beyond Compression: AI for Deep Thinking **Source:** [https://www.youtube.com/watch?v=p63MKDEsuFc](https://www.youtube.com/watch?v=p63MKDEsuFc) **Duration:** 00:12:59 ## Summary - Most people use AI mainly for compressing information—turning notes, long documents, or articles into concise summaries—rather than for deeper cognitive engagement. - The brain processes compressed content differently, so relying on AI-generated summaries can limit the formation of new mental connections and the transformative learning that comes from prolonged, focused study. - The problem isn’t AI itself but how we use it; for complex, high‑impact tasks we should allocate more “brain time” and use AI to reduce cognitive load, allowing us to think more deeply about the core subject. - AI can support deep work by handling routine compression tasks while we focus on expanding our understanding, as illustrated by the speaker’s challenge of writing a book on AI. - Mastering prompting is valuable but akin to learning to ride a bicycle—useful for quick tasks—where learning to “drive” AI as an intellectual partner offers greater long‑term productivity. ## Sections - [00:00:00](https://www.youtube.com/watch?v=p63MKDEsuFc&t=0s) **Beyond Compression: AI Partnership** - The speaker argues that using AI merely to compress information short‑circuits deep learning, and proposes leveraging AI to offload routine summarization so we can spend more brain time thoughtfully engaging with substantive content. - [00:03:29](https://www.youtube.com/watch?v=p63MKDEsuFc&t=209s) **Prompting as a Bicycle Skill** - The speaker likens mastering AI prompting to riding a bike—an essential, efficient skill—and explains how a conversational voice AI served as an interactive partner that kept ideas flowing, rather than merely transcribing them. - [00:07:02](https://www.youtube.com/watch?v=p63MKDEsuFc&t=422s) **Choosing AI Models as Cognitive Partners** - The speaker outlines how they evaluate and switch between AI models like Claude Opus 4 and Gemini 2.5 Pro to deeply critique outputs, clarify their core ideas, map their conceptual terrain, and manage partnership dynamics, noting a Substack guide they authored for model selection and their $200‑per‑month OpenAI subscription. - [00:10:33](https://www.youtube.com/watch?v=p63MKDEsuFc&t=633s) **AI as Thought Amplifier** - The speaker argues that thoughtfully selected AI tools, such as Opus 4, can deepen understanding and improve cognition when used deliberately rather than as a shortcut for mental effort. ## Full Transcript
I want to suggest that most of us are
using AI a little bit backwards. Stay
with me. This is worth it. We are using
AI primarily for information
compression. So take my meeting notes
and turn them into really neat outputs.
Take this large product requirements
doc, turn it into this for engineers and
this for my executive stakeholders. Take
this 100page PDF and turn it into
something succinct. take Nate's super
long substack and make it into something
I can digest. I kid you not, people do
that and it's kind of
hilarious. The point is we're using it
to compress information. But something
that I heard that's been really sticking
with me is the idea that the brain
doesn't process compressed information
in the same way. And one of the things
that we need to learn to think about is
when do we want to tolerate less brain
time on a
subject versus when do we want to
actually optimize our partnership with
AI? So we spend more brain time
marinating in what really matters.
So a lot of the learning that you get
when you read a large book, a deep book
on a big subject, it comes from your
brain forming new connections as it
spends extended time in the subject. If
you get and you can a very short
one-pager, you will get a prey, a
summary, an executive briefing on the
book. you are unlikely to have the kind
of lifechanging experience that you had
if you really dipped into it. Now there
are people who will take that insight
and say wow so AI is the problem. I
actually
think the way we use it is the problem.
Not that it's bad to use it for
compression. I too use it for adjusting
my meeting notes. I too use it for
sending different messages to different
stakeholders for routine business
communication. But if you need to do
really deep
thinking, you need to spend time on a
subject, you need to optimize your
cognitive workload to enable you to do
your best thinking. And AI actually can
really help with that. And I'll give you
a specific
example. I am contemplating writing a
book on AI. Writing a book is a hard
task. It's much harder, exponentially
harder than a single article.
it needs to be useful especially with AI
some of the challenges in having
something that stands the test of time
etc and I find that when I'm
contemplating writing a book the thing
that is most difficult for me it's not
the information compression it's not
that I need it to go out and tell me all
the news about AI I already get plenty
of news trust me it's that I need to
work with AI in a way that helps my
brain expand and kick around the subject
matter. And we don't really optimize for
that when we only talk about prompting
because prompting really optimizes for
one-way
communication. Learning how to prompt
well is a skill, but it's sort of like
learning to ride a bicycle versus
learning to drive a car. Both are
helpful. The car is going to take you
farther if you learn how to do it well.
And I think increasingly prompting is
like the bicycle skill. It's incredibly
efficient if you master it. Everybody
should know how to do it. It should be a
universal skill. I do think we're going
to need it for a while. I think it
provides a durable edge if you can ride
well, if you can prompt well. And I love
it. But if you can learn to actually
cognitively partner beyond an individual
prompt with AI, that's like driving a
car. That's like actually going farther.
And so in my case with the book, just to
circle back, I spent 25
minutes in advanced voice mode talking
with Open AI. Uh, and it was I think
they use like a variant of their 40
model in advanced voice mode. And you
might be waiting for me to say I had the
most profound insights. Like the AI
really helped me. It's it really wasn't
that. In fact, I sometimes told it this
is really vanilla. I don't love this.
What it was that was distinct and
special was it was there when I needed
to talk out loud. It would let me talk
out loud for a while. It actually
listened. It actually took notes and it
actually responded with just enough
interest, engagement, and riffing to
keep my brain flowing so I could keep
the idea coming because I had an idea
that I wanted to talk out. I knew I
couldn't just dictate into a
transcription device for 25 minutes and
get it. But if I had someone talk back
to me in the right cadence, I'd probably
be able to fish it out of my brain
because we're conversational people. And
so a conversational AI was the right
choice. In this case, advanced voice
mode really has gotten better since they
shipped uh that update that makes it,
you know, more aware of when it
interrupts you. uh it uses more
discontinuities verbally like uh and um
it just sounds more natural and it
allows you to forget it's there and just
focus on the subject which is what I
needed to do. That's just step one. I
did the verbal piece and then the second
piece was I really needed to wrestle
with what the terrain of the idea looked
like and I knew 40 wasn't strong enough
to do it. And so instead of trying to
force it inside that conversation, I
literally pulled the transcript out into
a Google doc and I stuck it into 03 as a
raw transcript and I said, "This is a
transcript. This is my perspective on
the transcript. This was the intent I
had with this idea. Really, this is
where I iteratively arrived." Uh because
I find that a lot of time when you're
talking out loud, you're sort of naming
the work as you go. And that's part of
the cognitive journey that we've kept
inside our heads for 200,000 years. And
now we have a partner to do it with. And
it's true that like sometimes if you did
paired work together with humans, you
could kind of get to this. But I
actually think it's a distinct dynamic
with AI. I have not worked and I have I
worked in offices, right? I've worked in
person with teams. I love that dynamic.
I don't want it to go away when we work
well together with with good human
colleagues. But it's not the same as a
good AI colleague. An AI colleague that
you can brainstorm with in that way
where they listen and just riff with
you. It's like this crossover between
the way a therapist listens to you and
the way a colleague listens to you. And
you'd never expect a human to do that,
but it's super helpful for your
thinking. So we do that. We name the
work. We get to 03 through the Google
doc. 03. And honestly, it's not just 03.
I don't want to just w Open AI here,
right? Like you can go and do this in
Opus uh the claude model opus 4. You
could do this in Gemini 2.5 Pro. You'd
get similar results. The idea is you
want to think deeply and critique the
model results that you get from the 40
conversation. basically take my intent,
take what I was able to articulate, take
the riffing that advanced voice mode did
and help me get to a crystal clear
understanding of the heart of the
idea. That's what you want in the next
step. It's basically help me to define
the coordinates of the terrain that I'm
in now that I've named the work. And
once we do that, the sky becomes the
limit. We can open up to understanding
what are the partnership dynamics that
work from here. So, as a specific
example, I thought 03 did a good job
getting to the heart of the thesis for
the book. I thought the outline felt a
little bit heavy. I need to go into a
different model. I'll probably choose
Opus 4 for this to start to refine what
that looks like a little bit more. I
need to pick a different cognitive
partner. And if you're wondering, how do
I pick models? I put a whole thing up on
Substack on how I sort of pick models
for different tasks.
Um, to me, I think the piece that I keep
coming back to, stepping back,
reflecting on all of this is that I find
the most value. Like, this is real
expense, right? Like I I'm
pro-subscriber to open AAI, right? Like
I pay my 200 bucks a month. Not
everybody has to. I don't think you have
to to get a ton of value out of it. I
saw the article that said, "Is free so
good that you don't need plus?" I think
for a lot of people it is. But for me,
whatever whatever your investment level
is in AI, the value you get is so much
greater if you use it in this way as a
way of getting your brain time on
subject. Optimizing for time on subject
versus optimizing
for just compressing and repurposing
information. Most of the use cases I see
from companies if they are not super
fluent in AI end up being around
compressing and repurposing information
which is fine. There are some cost
savings there but helping your brain
work better has a lot more upside over
time and I think we talk about it a
whole lot
less and I wish we would talk about it
more. I wish we would not get so far
into the compression trap that we don't
think about this idea that AI can help
us map and expand the cognitive
territory that we work in by partnering
with us in a way that is, you know,
somewhat reminiscent of what we would do
with a human, but also distinct. I I
also, you know, just as I said, I
wouldn't ask a human to do the kind of
listening that I asked advanced voice
mode to do. I also wouldn't really ask a
human to work on sharpening my thesis
for an outline the way I would ask 03 to
do it. I wouldn't ask a human to go and
like necessarily do the research and do
the thinking um and the for lack of a
better term cognitive shaping on the
thesis with me. It's not just doing it
on its own that I would ask of Opus 4.
Opus 4 is great for shaping ideas and
really thinking through concepts. And so
I'm finding in a sense that my brain is
able to think better by spending more
time with AI models if they're carefully
modulated. It's not just throwing
prompts in there. There's no secret
prompt to this. It's deciding the kind
of task I
need and then making sure that
I approach AI and ask myself what
is the model that I can choose that will
most help me
to optimize for getting my brain deeply
into the subject matter so that I become
someone who truly understands it. That
in and of itself is opposed to it's
antithetical to this idea that AI lets
you go cheap on your brain, right? It
lets you skip the brain power. It lets
you not do the work. It lets you read
the one pager, not the full book. I
don't think it has to. I think it can
actually help you to think better. I
find it
does, but it's how you use it. And it's
not that to say at all that like again I
started this by saying I use the
compression stuff too. I think we all
do. we all will. It helps us save time.
It's about knowing when to sort of pull
the thinking button and say, "I want to
use this to actually expand my time
thinking about this because this is
really important. This is work that
needs and deserves the cream of my
brain, and I need to orient my approach
to AI so that I can spend more time
marinating on this in a way that works
for my brain." And part of why there's
no magic prompt for this is our brains
are wired pretty uniquely. The way I
choose to like walk around and talk and
like talk my ideas out loud may not work
for everybody. There are people for whom
they need to write it out by hand, take
a picture of it, get a transcription,
and maybe that's the way they start. Um,
I think your mileage is going to vary
because your brain varies. But the
concept of optimizing for having your
brain in the subject is something that
allows us to view AI as an expander, not
just as a compression tool. It's
something that optimizes for what we can
create, not just how we can cut cost by
compressing information, even if that's
useful. So, thinking about this was
helpful for me. I hope it kind of turns
the wheels for you, maybe helps you
think about AI a bit differently and try
out advanced voice mode. It's fun.