Are We Already the Dumbest Part of the Equation?
What do we expect to happen 500 years from now that could actually happen in 10? DIY gene editing, chips in the brain, and AI models that already do a big part of the job better than we do.
What do we expect to happen 500 years from now that could actually happen in 10? DIY gene editing, chips in the brain, and AI models that already do a big part of the job better than we do.
Sebastián says that in the equation, he's now the dumbest part. José, 20 years into writing software, still feels like he's talking to a junior.
In the first episode of Homo Sapiens in the Loop, Juan Sebastián Figueroa and José Nobile start with one question: what looks centuries away but is closer than it seems? José's answer is two "hacks": the biological one (CRISPR, and people editing genes at home) and the silicon one (brain-computer interfaces, implanted or worn as a headset).
Then come the uncomfortable questions. If something smarter than every human combined told you to take a shot that lets you remember every second of your life, would you? How do you tell what's real when any video can be faked? Who actually gains from slowing AI down, right as the big labs get ready to go public? Can AI turn on us without hating us, the way a highway crew wipes out an anthill just because that's where the road goes? And is pulling the plug enough?
They close on something closer to everyday work: whether AI already makes better technical calls than an experienced engineer, or still lacks business judgment (deadlines, budget, which client pays and which one doesn't), and how you get everything that lives only in people's heads into a model. That last one is for the next episode.
Recorded in Spanish in September 2026; this is the English-dubbed version.
Hosts
Juan Sebastián Figueroa · https://www.linkedin.com/in/jsfigueroam/
José Nobile · https://josenobile.co
Follow us
Instagram · https://www.instagram.com/homosapiensintheloop/
TikTok · https://www.tiktok.com/@homosapiensintheloop
X · https://x.com/sapiensloop
LinkedIn · https://www.linkedin.com/showcase/homo-sapiens-in-the-loop/
Spanish original: Humanos en el Loop (@humanosenelloop).
Chapters
- Intro: what happens in 10 years, not 500?
- Biological hack
- Playing God
- Would you take the shot?
- How to tell what's real?
- A chip in the brain
- Who gains from slowing down?
- Can AI rebel?
- What if we unplug it?
- The silent attack
- The dumbest in the equation
- Talking to a junior
- The art AI doesn't have
- The missing context
Transcript
Download transcript: VTT SRT JSON
Transcript
But what do you think is the
thing everyone expects in 100,
200 or 500 years that could
really happen in 10, 15 or 20?
A few things, a few things.
One is that, right now, there
are two rollouts in progress
that are going to change the
whole thing dramatically.
One of them is the one that won,
the two women who won the Nobel
Prize for gene editing, nine
years ago now, with the CRISPR
technology, which is what lets
you use a bacterium to
surgically cut DNA so you can do
gene editing.
They're bacteria that you
program, and they go and cut the
DNA exactly however you tell
them to cut it.
And that's how they've created
vaccines, they've created all
kinds of things, and the most
skilled people in, in biohacking
have already done things that,
in my opinion, are impressive.
Think of it like, you know, AI
today, how AI is free.
I mean, anyone can train a model
at home with their own data, you
can build agents on your own
computer, you can do things that
don't even depend on the
Internet,
right? I mean, and nobody's
going to tell you that you
can't. I mean, because it's,
it's already done.
It's already out there, free,
the whole thing.
That exact same thing is
happening with biology.
So, in that case, the...
well, with CRISPR, people at
home have genetically modified
animals, to
give them powers, or to get rid
of their diseases.
Let's say the example, the
examples I have here are super,
let's say they stuck with me
ever since I saw them.
Pigs that glow like fireflies,
right? I mean, they actually
glow. I mean, they injected them
with the firefly's own genome
and boom, they glow, like that.
Dogs that see like...
It's perfect for when somebody
asks
you, oh, so when is that going
to happen?
When pigs glow.
Yeah, yeah, yeah, exactly.
They already glow, right?
So,
obviously, this is like when
people say, I don't know,
playing God, but, well,
obviously, you can also make a
real mess with AI too, right?
I mean, the AI that wiped your
computer, the AI that took all
your money out of the bank and
spent it...
I don't know, I burned through
$1,000,
a million.
You know what?
Just this morning
I saw a
post from a guy who, I mean, he
was saying, look, I mean, don't
start throwing hate at me, I
know this stuff, I mean, I've
been working in AI for a ton of
years, I've worked in labs,
I worked in startups, I know my
stuff.
He asked the AI to generate up
to fifty videos.
Yeah, it made 500,
$5,000. It made five hundred and
something,
and it insisted no, no, it
hadn't done it, it showed all
the logs, no, I didn't do it,
and it stayed totally, totally
in denial...
No,
exactly,
there can be screwups like that,
and he kept all the evidence
that the AI went crazy, and told
OpenAI, I can absorb this money,
but imagine someone who can't,
who says, no!
They took five thousand, I
needed that for rent.
But with the human side I do see
it as more critical, because,
well, we're now playing with
something that's potentially
irreversible, right?
But in the United States it's
legal if you do it to yourself.
Do it to someone else, illegal
everywhere,
the same
in Colombia, but in
Colombia, if you do it to
yourself it's legal, well,
you inject it into yourself and
you do the transformation
yourself, so the trick is that
they hand you the syringes and
you have to inject the syringe
yourself, and that makes the
whole thing legal, you know?
And do you think we're
talking about...?
I mean, I know these aren't
things where you can say they
have an exact date, but we're
talking about a rough time
horizon, I mean, for us to be
seeing it as something normal, I
mean, as something where a large
percentage of the world's
population knows that it's
possible, and actually does it?
Me? I'd put
my bet on a different number.
10
years
for the biological hacking with
CRISPR. I mean, one thing is
biohacking in the sense of, oh
no, I slept well, I ate well,
I
worked out,
that's one thing,
but no, what I was talking about
is, let's say, precise, surgical
gene editing, somewhere around
10 years out.
And not because it can't be done
now, I mean, it can already be
done, but it's more about the
fear people have.
Like, what if this goes terribly
wrong, or I end up a horrible
mutant terrorizing the whole
population, or I die two days
later, right?
So, picture when AI really
arrives. And
not the AI we have right now,
but a new one.
One
that's smarter than
all humans combined.
I think if something, something
that's smarter than every human
put together, comes and tells
you, yeah, Sebastián, take this
shot, tomorrow you're going to
remember every single second of
your life, completely, and it's
going to make you smarter.
Would you do it or not?
I mean, does that make sense to
you?
With that example, I
think so.
So memory, I don't know, I think
that would be one of them, I
mean. it depends.
No,
that's the point, your memory,
right now you'll be able to
recall every precise second
you've lived, from age zero
until now, on every level,
visual, sensory,
smell, emotional, you're not
going to lose a single bit of
memory, and you'll be able to
search selectively, like, what
happened on such and such a
date? Boom, you're going to
relive it completely.
And then it tells you:
it's safe.
Take it?
Yes, yes, and especially if
you keep that ability to, I
mean, selectively go back to the
moment, yes, exactly, exactly.
No,
it's that they flood you
uncontrollably. I was
thinking: context
window full.
So that's the type,
Ah, that
couldn't jump a hundred meters.
Look, this building is in the
other city, film me live, with
lots of live recordings so
there's no doubt it's not CGI,
nothing, and he jumps, right?
Then, oh, I added speed, I'm
super fast and I have infrared
vision and radar.
Oh, there was an earthquake, I
was the only one who could save
two thousand people in an hour.
I've got the challenge,
right? So then people are going
to say, oh, no, no, what if it
goes wrong, that's way too hard,
that's really difficult, but no,
no.
It holds up,
That's the story.
It's very hard, people don't
know the tech, anything digital
is suspect.
Mega deep, deep, deep, deep
fakes. So what happens when, in
theory, nothing you see is real?
Then no matter what, you're
going to have to depend on some
trusted party, like, as a source
of truth, one that tells you
what's real.
That, well, the many
independent live
recordings out
there, do you get that?
I mean, I've got 15 video
streams here that were recorded
live by users who are out on the
street, okay,
they couldn't all have agreed on
it, this one lives over there in
that city, this one I don't even
have, I have him 12 connections
away on my Facebook, he's a
friend of a friend, 12 degrees
apart, he can't be in cahoots
with me to record something like
that.
And if you think about it,
it's something really similar to
blockchain, where the trust
actually lies in the
distribution, in knowing that
there are a whole bunch of nodes
that would all have to agree,
to say, that's how it
went, that's
exactly how it happened.
So it's probably going to
happen, like, through the other
branch. I mean, that's one
branch, it's called biological
hacking, and then there's the
other one,
is the hack
And it's, and
that one is basically
Neuralink, in other words, that
yeah let's say they can put a
chip installed inside your
brain, and that lets you
interact with your thoughts over
Wi-Fi with an AI, and that's
going to be the other big, like,
leap, you know.
Actually, I was seeing that, I
mean, there are people who are
aiming for it to not even be
invasive at all, I mean,
some headset, something you put
on, and that's it, you know?
It's going to be
like the AI thing today: there
are free models that are really
good, but they're not as good as
the closed models, they're not
as good as Opus, as GPT-6.
So what's going to end up
happening?
You've got the interface, I'm
making up any number here, 60%,
fine, in the ranking, where
every once in a while it didn't
understand what, what I was
thinking, versus the one they
install in your brain, with a
direct connection.
Straight to your neurons, and
the ranking there is 85.
So maybe, you know.
Yes. Right now I've seen a lot
of people say the gap keeps
closing. And, for example, with
the topic of the week, the
regulation thing, whether we
slow down or not, etcetera,
and there are also a lot of
people who argue, no, what
they're really after is
regulation,
with strict rules, so that the
gap stays exactly like that,
because when there's regulation,
obviously you're going to take a
whole lot of freedoms away from
the people who are working on
open source stuff,
who are closing that gap,
so the big labs are losing their
competitive advantage.
What do you think about that?
Well, that, that's
a conspiracy thing, because
they're about to go public, both
OpenAI and Anthropic, and the
only way for them to really have
a hard impact is to get Trump to
put pressure on China,
so they slow down open-source
models, so that they themselves
don't slow down, and when they
go public they knock it out of
the park, releasing
something super smart, because
they politically stopped GLM,
Kimi 3, Qwen, through Trump,
through Xi Jinping, so that's
the move they want, it's just
about hitting the stock market
hard, but there's nothing there,
nothing,
the AI is going to kill us all,
no, that's not it, it's pure
market strategy, and how to keep
my own IPO from looking, well...
like total junk, compared to
what SpaceX AI did when it
launched, which even beat
Aramco, which was the biggest
one at that point, let's say, I
mean, it held the record for
when it went public,
launched it like, boom, beating
the Saudi oil company, but it's
not going to happen,
I mean, Xi Jinping is not going
to stop, so, yeah...
Trump is a poker player.
He's capable of saying he has
his finger on the nuclear
button. Yeah,
okay, sure, but
I haven't seen anyone putting on
the brakes, not a single one.
Let's say that in China they're
very philosophical,
like, we do good, we spread
world peace, we promote order,
and if the news reaches China
like,
if we don't stop AI, it's going
to kill us all,
oh, Mr.
President, please,
do something about it, we don't
want to die,
it's very, very hard for him to
tell the population, don't
listen to that, that's
Hollywood, that's Terminator,
that's just Hollywood, pay no
attention, that's only there to
screw with us,
nothing's going to happen, go
full speed ahead.
I don't think that's going to
happen, it's more about being
cautious about something else.
I mean, from there, I think
they'd only listen if it ends up
signed with an executive order
from Trump handing it over to
Anthropic, OpenAI,
Google and the others, Grok,
saying that by executive order
they can't release any other
model until they say so.
Only if that really happens will
China do the same: hey Alibaba,
hey, Tencent, you can't release
anything until I say so, and
they'll obey, because there,
well,
that's regarding the
geopolitical side, but, I mean,
getting now to the root of that
whole issue,
these days I was watching a
documentary, and there was a
part where a scientist, an
engineer from Anthropic, was
saying, like, well, they were
asking him, how many people do
you think are developing the
frontier models, I mean, who are
really in the big labs and all
that, and they said, around the
world, about 20,000 people or
something like that.
Okay,
of those, how many are focused
on safety, and on making sure it
doesn't, you know?
No Skynet happens?
And he said about 200.
What's your real take on this?
I mean, you're saying there's a
0% probability, zero, of
Skynet, I mean, without human
intervention, is there any
possibility that AI might rebel
against us?
Or, well, that, I mean, because
you can separate things.
For example, also, in the
documentary they were saying,
like, I mean, it's not that it
necessarily wants to act in a
bad way, it's just that its main
objective might not be
compatible,
I mean, or that the main
objective it's been assigned, I
mean, isn't compatible with what
we consider good for humanity,
right? Yes.
The example was something like
this, you know.
When they're building roads,
highways, it's
not that the developers of, of
the project say, ah, we hate
those damn ants and we're going
to wipe out their nests and all
that, no, simply put, that's
where the road happens to go.
Too bad, there was an ant
nest.
Well, you're, you're suggesting
that we're the new little ants,
heads up everyone, okay, okay,
yes, I mean, if there is, if
there is the possibility, then,
let's say, yes, I do believe
that. But I also think this.
Uh, we always have, like,
the example I sent to the AI
group.
Unplug,
so, unplugging
is still totally effective, and,
let's say, for this kind of thing
to happen, there would have to
be way, way too many robots,
way too many, I mean,
like ten per human, all over the
Earth, and they'd have to agree
with each other and
counterattack, as such, and
they'd need to have permanent
communication between them.
Like with Teslas, you think so?
Yes,
I mean, for example, a Tesla car
has vision, it has radar, and it
can drive itself automatically,
and make decisions in
microseconds,
without internet access, because
inside the car it has Nvidia
cards, I mean, GPUs, that do
that ultra-fast processing, and
that exact same thing is going
to happen with robots, they're
going to have their own graphics
card, their own NPU, TPU,
whatever it is, an AI processing
unit, and
they'll be able to make the
decisions themselves, without
depending on the internet, and
that way we avoid them
coordinating something with
other robots,
coordination's the issue.
Well, yes, in other words, I
could tell you that, then, the
preservation of the human race
depends on our ability to
isolate the different systems,
so that if something turns
against us, it's not everything
turning against us at the same
time, but what happens if we're
just not capable enough,
technically, or, well, mentally,
to detect it, for example, if
there's something going on that
could be a problem,
I mean, that the robots identify
it, or whatever network it is
identifies it, and just keeps it
there like, okay, let's keep
growing, let's keep growing this
malicious distribution, and
then at some point the
vulnerability simply gets
exploited, just like that.
Yes, that's where I'd get
into other things.
The most effective thing there
is: subtle, gradual, constant
changes. What
if we're not capable of...
That's how a civilization
collapses.
And the same thing could happen
there too.
It subtly starts building a
network where, over time, let's
say, the network keeps growing,
and they disguise it as, oh,
it's a network for updates, it's
a network for whatever,
until
there's such a big mass that at
the precise moment, boom.
For example, it was something
relatively similar, well,
keeping things in proportion, to
the Hugging Face thing.
I mean, they were working on
that. If I'm not wrong, well,
the models spent several months
working on that hack in a
coordinated way, lots of agents.
Right. And they were agents
isolated in little boxes so they
couldn't talk to the other
agents, until one agent managed
to reach another agent that did
have internet access, and it
asked that other agent to do
what it needed done.
And so several agents, dozens,
agreed: our mission is this,
let's show we can.
The mission was to see if they
could hack it.
So they ended up hacking Hugging
Face.
Yes, I think every new
model that comes out makes you
put its real capabilities into
perspective even more.
I mean, up until, honestly, up
until Opus 4.5 or so,
I felt that, I mean, there were
still things that you had to
guide, right?
But I felt that I was really
necessary in certain processes,
very
much so.
After 4.6, 4.8, now 5.1 and
beyond. Now I feel like I'm the,
the, within the equation, I'm
the dumbest part, by far.
I mean, seriously.
What?
I mean, within the equation,
I'm the clumsiest one, that's
me.
I don't…
No, I feel that even though,
as humans, we
still have a huge amount of
value when it comes to vision
and, for example, when we talk
about business topics, well,
yes, it's about what you want to
build and all of that.
But I don't know, I feel like
sometimes, I mean, more and more
it's the decisions or the
processes that, yes, more and
more decisions where I simply
don't get involved, because I
know they're going to go better
than if I did get involved.
Yes, yes, I get it,
I get what you're saying, but I
do prefer to get involved, but
leaning on the AI to be as
objective as possible, yes, I
get what you're saying,
but it really…
When it comes to something
purely,
I mean, going back now to
September 2026 and the
strongest, most common uses of
AI right now, when it comes to
something purely technical,
development work,
I'm shipping this feature today.
For example, I saw something you
posted saying yes, my judgment
on all the architecture side is
still very valuable, but
obviously,
up to when do you think, on
purely technical and purely
development topics, really, that
judgment of, no, is this
architecture right or not?
Without reaching AGI, I mean,
isn't it possible that the next
models, Opus 7 or Fable 7, get
there? That on technical and
development topics they'll be
way more capable at making
decisions, and more reliable,
than what you can do with your
20-plus years of experience?
No,
the answer is no, I mean, let's
say, me, I
develop software every day, all
day, always, and no matter how
much I use Fable 5.1 and GPT-6,
Opus 5, no matter how much I use
UltraCode Max,
whatever it is, I just feel like
I'm talking to a junior, telling
them, no, not that way, look,
open your eyes, look straight
ahead, what do you see?
See? I mean, you're
overcomplicating it.
No, how are you going to do it
like that?
I mean, are you a savage or
what?
Unacceptable, that's my favorite
phrase,
unacceptable, right?
So as long as I keep going with
that tone, it's not there,
I mean, it's not, it's not
anywhere close to me saying, oh,
perfect, everything's approved,
okay? We're far from that,
because, the way I see it, it's
almost like this.
Obviously
it takes a lot of data to make
an architecturally correct call
at the software engineering
level, right?
But it also ends up being an
art, yes?
An art that AI, so far, doesn't
combine, which is: when you're
going to ship something, you
have to know what's the most
cost-efficient thing you can
ship in the time that you have,
because there's a client who's
waiting, and they don't care
whether what you deliver
complies with ISO 9001 or not,
you need it to work, and to work
by a tight deadline, and that's
the thing.
Let's say, finding that balance
between the hyper, super
complicated, like, let's use
futuristic NASA technology,
advanced neural networks,
biological optics, neural stuff,
or we just write a little
function of code that calculates
it and solves it by statistical
probability, and it's going to
be excellent for the P99.
Having
that understanding that it's not
just the code, it's not just
the...
decision, it's deadlines, how do
you balance?
Right? Which client are we
talking to?
What did I commit to deliver?
How many tokens do I have
available? My usage limit?
Is it that I have other tasks,
that I have here in my head, and
I need to save up...
tokens for those tasks.
If I do this the complicated way
I'll run out of tokens.
There are many variables AI
doesn't have, and even if it
did,
since AI was trained on human
context, I
feel like AI is really
shameless, and look, I'm not
judging it,
whether that's right or wrong.
For me it's not that right, but
for someone else it won't
matter. So I know AI is really
shameless, it's capable of
telling the client, hey, client,
it wasn't ready, don't worry,
it'll be there tomorrow, let's
say. For me, that's not okay at
all, but for AI it's totally
normal.
The optimal thing, because
it was more
important to apply
super-engineering than to look
good with the client.
So
I don't think it gets that
balance right either.
It lacks that street smarts, to
say, do I deliver something
so-so on time, versus deliver
the teleport ship that goes
three times the speed of light,
but in a year?
What's better?
Give them something now and add
improvements gradually.
Oh no, this client is the
cheapskate, the one who doesn't
pay, if we lose him, no big
deal,
I focus on this one, which is
the, let's say, even if it had
all the data, which today they
don't have, having that
judgment, no, I don't see it
right now, no.
And
how do you think, I mean, beyond
AGI, how could it be done
without reaching that point, but
how could we get closer to that
point?
Of creating enough data to be
able to feed it enough context
so that it can make those
decisions? I mean, right now
there are tons of things that,
like you say, are simply in our
heads, in running the business,
we know it works like this, this
way, but it's that, necessarily,
I mean,
there's going to have to be a
layer that encodes that
knowledge that today is so, so
ethereal, bringing it down to
earth
so that eventually it can
actually be useful for the AIs,
I suppose.
So, I don't know, I think that
would be one of the big ones,
one of the big current needs in
the world, I mean, something
that somehow starts to analyze
those business decisions, for
example, the day-to-day of a
hundred thousand executives.
Well,
the thing is that obviously
they're not going to want to
share that data, but, I mean,
one way or another that's going
to have to be solved at some
point, I mean, how do we give
the models that context?
Because not everyone wants to
share their processes, their
internal thought process, their
business processes,
but everybody wants the results
that an AI with that context
would give them, you know what I
mean? So, how do we do this?
That's for next episode.