Open AAI just got done hosting some sort

of town hall in which they brought in a

bunch of the developers and Sam

Jippidity Alman fielded a bunch of

questions from a bunch of developers and

tried to kind of give uh you know open

AAI or his perspective on what's going

to happen and there was honestly some

pretty good questions out there and some

kind of I I guess interesting uh

responses from Sam and not only that but

at the end of all of this or towards the

end there was kind of like a really you

know off-putting question but an even

more off-putting answer like I couldn't

even believe he would say it just makes

it makes uh makes things feel more

uncertain you know because here's the

deal is that typically if I'm going to

listen to the doom and gloom of AI you

got to go to Daario okay CEO of

Anthropic okay every single time you see

him talking about the doom and gloom

he's making the painful face and going

you know

I mean it's going to they're

going to create an oligarchy and I'm

just not sure how to prevent it I guess

everyone creat create an AI is going to

be a trillionaire and then we're going

to have the permanent underclass. I

mean, it's just one of the costs of

progress, you know, like every time

you're like, "Dude, yo, Dario, why why

you got to bring me down like this,

okay? Why are you talking about a

country of geniuses? Okay, it hurts my

feels." Anywh who, the town hall lasted

a little bit longer than an hour and so

I kind of picked out some of my favorite

parts and I'll go over them and we'll

yap about it for a second. Then we'll do

the ending one that's kind of I don't

know just I still I guess feel weird

about it. I kind of feel weird that

nobody feels weird about it. I don't I

don't know. Maybe I you know I get I get

so easily influenced. So this guy asks

like, "Hey, models are really expensive,

dude. You going you going to hook a bro

up? You going to make him cheaper?"

>> I think we should be able to deliver GPT

5.2x high level intelligence

by the end of 2027 for

Do you want to give a better guess? I

can give one otherwise.

Anyone want to give a guess? I would say

at least 100x less.

>> Uh but that's kind of a crazy statement,

right? So 5.2x high within the next uh

what is that 22 months is going to be

100x cheaper, which by the way is

actually pretty consistent from Samuel

Jippy over here that he actually does

say that AI is going to reduce the cost

10x every single year. So in 22 months,

23 months, we should see 5.2 2 being

100x cheaper, which actually would make

it usable by a lot of people. Now, I

have no idea how they're going to make

it 100x cheaper. It's just going to be

100x cheaper. I have a sneaking

suspicion that there's going to be one

man who has to figure this out, too.

Just carrying the weight of inference on

his shoulders. Before we get to, you

know, the the the crazy part of this

whole thing, there was one more. Hey,

guess what everybody? We had a celebrity

in the mist. Look at this one right

here.

>> I want to ask about something a little

bit different, though. more on the

technical side. I

>> uh by the way, this is the O dev

YouTuber YC founder.

>> One of the fears I have as the models

and the tools we used to build with get

better is that we might get stuck with

the way we have things working. Now,

>> Theo goes on and explains the problem a

little bit more and Sam of course gives

effectively an answer, non-answer.

Honestly, this is a very good question

and this is kind of like the big fear,

especially with ads kind of creeping in.

One can imagine that maybe the LLM just

always does give kind of a certain

answer once we get past the kind of

hard-coded big displayed ads. But

nonetheless, this happens all the time,

right? Like if you ask to build an

application for the web, the chance of

you getting React is just really, really

high. Like certain patterns just exist

and statistically speaking, that's the

pattern you should use. And if these

magical statistical machines do

anything, they're going to give you the

answer that you need for that question,

which just might be the same

technologies over and over again, which

makes it actually kind of interesting

because how does one make a better

system if one doesn't even know how to

make a system to begin with? And even

more so, how does one of these models

know which of the systems to pick when

80% of them keep using the same item?

Like, wouldn't that be the one you'd

want to choose, the one with the most

results? Great question. Very tough one.

There's I feel like there might be a

little bit of manual manipulation coming

in. Again, ads. Little bit worried about

that one. All right, let's actually get

to the big one. This is the question

that I just feel like I I I guess I

didn't see anyone talking about. It's a

really interesting question, but more so

it's the response. I'm going I'm going

to let the full question be played and

then part of his response be said and

we'll stop it at kind OF LIKE THE WHAT

PART. question is where does security

fall in this 126 roadmap and um broadly

how do you think about some of these

issues

>> security broadly or biocurity

specifically

>> um either preferably biocurity

>> there are many ways AI can go wrong in

2026 certainly one of them that we are

quite nervous about is bio uh the the

models are quite good at bio and right

now most of our and by like our not just

open eyes the world strategy is to try

to restrict who gets access to them and

you know put a bunch of classifiers to

not help people make novel pathogens. AI

is going to be a real problem uh for

bioteterrorism. Uh AI is going to be a

real problem for cyber security. AI is

also a solution to those things. It's a

solution to a lot of other problems as

well. I think we need like a societywide

effort provide the infrastructure for

this resilience not labs that we trust

to sort of always block what they're

supposed to block and you know there

will be many good models in the world.

We've been talking to a lot of bio

researchers companies about what it

takes to be able to deal with novel

pathogens. I think there are a lot of

people interested in the problem and a

lot of people reporting that AI actually

seems helpful at this but it won't be a

technological it won't be an entirely

technological solution. You will need

the world to think about these things uh

differently than we have been. So I am

very nervous about where things are but

I don't see a path other than the sort

of resiliencebased approach and it does

seem like AI can really help us do that

fast. If something goes really wrong

like visibly really wrong for AI uh this

year I think bio would be a reasonable

bet for what that could be and then as

we get into next year and the following

year you can imagine lots of other

things going really wrong too.

>> What like are we going to get a co 2.0?

Is that is that what he's dropping right

here? That we're about to have some

horrifying moment in time in 2026, 2027?

Like, this doesn't sound good. Also, by

the way, I hate this answer. AI is going

to be a real problem uh for

bioteterrorism. Uh AI is going to be a

real problem for cyber security. AI is

also a solution to those things. It's a

solution to a lot of other problems as

well. Dude, there's something about the

fact that you can create something that

could potentially have implications on

millions of people and then also be

like, "Hey, you know what? You know what

though? I know it's the problem, but

it's also the solution." It's like,

dude, isn't that kind of the thing they

always warned me about in history class?

Isn't that what the bad guy always does?

Creates the problem and then sells you

the solution. I I am Am am I wrong here?

Are we see are we seeing ourselves the

active creation of of of of the

classical historical bad guy? Anyway, so

that is like a realities that he's

dropping. And here's the thing is I'm

not even sure how AI solves that. If

somebody gets a hold of these models, if

things get cheaper, if technology gets

much much better, one could imagine that

producing large scale models that we

have today in a couple years will be

significantly cheaper due to

improvements in technology or whatever

nonsense actually ends up happening. And

then boom, all of a sudden they can just

have their own model that does things

that are terrible and then what? How do

you prevent that? How does AI prevent

that? It doesn't. Somebody just went

off, brain drained your super great

model, tossed it into a smaller one, and

bought a bang b Like what? This isn't

good. I don't know. I just feel like I

had to to talk about this one. I I know

this one's like the least happy of all

my videos. I just don't even know what

to do with it. I just feel like how come

no one's talking about this point? It

kind of feels uh rather interesting that

uh Mr. Jeypy over here thinks that 2026

or 2027 if something does go really

wrong, it's going to be bioweapons.

Dang, that's uh that's not a W. Not you

know what, today not a W. Anyways, the

name is I hope you enjoyed this nice

town hall.

You know,

you guys, you know, I always, you know,

I always appreciate you. You know what

I'm talking about. I hope you guys, you

know, I hope you feel encouraged. I hope

you're out there, uh, learning, actually

taking the time to get better at your

craft. Uh, you know, maybe not letting

Claudebot take over all of your private

messages and then accidentally exposing

it for the world to just come in for

free. You know, I hope you guys are, you

know, not doing that. Having a good day,

you know, again, don't worry that bad

Sam Alman's not going to get you. Don't

you worry.

Hey, you're probably wondering why am I

in San Francisco? Well, I'm here for a

big event and I'm going to stream the

whole thing. It's going to be live on my

channel for the next 5 days. So, if

you're watching this video, it's

probably live right now.