GPT-5 Is Coming In 2024? Sam Altman Announces Updates at Y Combinator W24

Corbin ai
14 Jan 202412:37

Summary

TLDRThe video discusses the implications of the upcoming release of GPT-5, a new AI model from OpenAI. It starts by showing a clip from a recent AI conference where an expert explains GPT-5 will have improved reasoning abilities compared to GPT-4. The host then analyzes tweets from another AI expert that contain key takeaways from the conference. The first takeaway is that startups need to focus on contextual optimization rather than just behavioral optimization. This means startups should avoid building on top of GPT-5's capabilities, and instead provide unique value and solve niche problems. The second takeaway is GPT-5's API will likely get better and cheaper over time, which is positive but may also increase competition. The third warns startups against simply repackaging GPT as their own product. The host explains most people are unaware of advances in AI, so GPT-5 will be a wake-up call. He then outlines two potential business paths: services for non-coders, and software for coders. For non-coders, the focus should be building services that integrate GPT-5, since most businesses won't understand it. For coders, they should build niche software that solves focused problems, not just repackage the API. Overall, the key message is GPT-5 will be transformative, so startups need to provide unique value not easily replicated by the technology itself.

Takeaways

  • GPT-5 will likely launch in 2024 and bring major improvements in AI reasoning ability.
  • GPT-5 may operate like an advanced personal assistant that understands user data and preferences.
  • Startups should focus on solving niche problems rather than competing directly with large AI companies.
  • Non-coding startups can offer AI integration services to help businesses adopt new tech.
  • Coding startups can build niche SaaS products that solve specific problems.
  • AI website builders will likely make web design agencies obsolete in the future.
  • Most people are currently unaware of rapid AI progress and will be surprised by GPT-5.
  • API costs are decreasing which will increase competition but improve startup viability.
  • Services that support human needs not easily replicated by AI have an advantage.
  • Being an early mover provides opportunity, but the space moves quickly.

Q & A

  • What year is GPT-5 expected to launch?

    -GPT-5 is expected to launch in 2024.

  • How might GPT-5 operate as an AI system?

    -GPT-5 may operate like an advanced personal assistant that can understand user data, preferences, calendar info etc.

  • What is the suggested strategy for AI startups to avoid competition with large companies?

    -Startups should focus on solving niche problems rather than competing directly in areas targeted by large AI companies.

  • What services could non-coding startups offer?

    -Non-coding startups could offer AI integration services to help businesses adopt new technologies like GPT-5.

  • What is the suggested approach for coding startups?

    -Coding startups can build niche SaaS products that solve specific problems vs. replicating large AI models.

  • How will GPT-5 impact web design agencies?

    -AI website builders will likely make traditional web design agencies obsolete in the future.

  • What is the current awareness of GPT-5 capabilities?

    -Most people are currently unaware of the rapid pace of AI progress and will be surprised by GPT-5 abilities.

  • How are API costs changing?

    -API costs are decreasing which will increase competition but improve startup viability.

  • What types of services will have an advantage?

    -Services that support human needs not easily replicated by AI have an advantage.

  • Why is being an early mover important?

    -Being an early mover provides opportunity, but the AI space moves quickly so timing is key.

Outlines

Introduction and Overview

The paragraph introduces the video, stating it will provide information revealed at a recent AI conference regarding GPT-5, a new reasoning-focused AI model slated for release in 2024. It previews key topics like how GPT-5 will impact startups and what it means for the future of AI.

Insights from the AI Conference

This paragraph summarizes key insights shared at the conference. GPT-5 will focus on improving reasoning ability and reliability compared to GPT-4. Customizability, personalization, and connectivity to user data are other areas of focus. The improvements could make GPT-5 integral like an advanced operating system.

Other Relevant Twitter Insights

Three main Twitter insights are highlighted: Startups need to provide more contextual optimization versus just behavior optimization to avoid direct competition with GPT-5. The GPT-5 API will improve and get cheaper, forcing startups to provide unique value. Startups should avoid just rewrapping GPT-5 and instead focus on unique reasoning abilities.

The Significance of GPT-5

This emphasizes GPT-5's significance, stating it will 'wake up' many people currently unaware of AI's progress. Staying informed is key, as GPT-5 represents a new phase of the AI revolution that will drive major changes.

Starting an AI Business

The paragraph contrasts coding and non-coding pathways for starting AI businesses. Non-coding focuses on services, while coding enables software-as-a-service. Key advice includes solving a micro-niche problem first and targeting markets resilient to displacement by future AI.

Conclusion

The concluding paragraph recaps the key points about GPT-5 and starting an AI business. It encourages checking out accompanying playlists and videos for more details.

Mindmap

Keywords

💡GPT-5

GPT-5 is the next version of OpenAI's Generative Pre-trained Transformer language model that is expected to be released in 2024. The video discusses some of the expected improvements in GPT-5 like better reasoning abilities, increased reliability and consistency, more customizability, and the ability to connect with personal data sources.

💡reasoning ability

One of the most important expected areas of progress for GPT-5 is improving its reasoning abilities. Currently GPT models have very limited reasoning capacity but the speaker notes this is a key focus for future iterations. Better reasoning would allow GPT-5 to provide more logical, nuanced responses.

💡reliability

In addition to reasoning, reliability of responses is called out as an important area of improvement for GPT-5. The goal is for it to provide the best possible response consistently, instead of varying widely across repeated queries.

💡customizability

The speaker notes people want AI systems like GPT tailored to their specific needs and preferences. GPT-5 is expected to enable much more customization in areas like the style of responses, underlying assumptions built into the model, and integration with personal data sources.

💡startups

With the advancements expected in GPT-5, startups need to avoid directly competing with or replicating its capabilities. Effective startup ideas would either contextualize the technology for specific use cases or focus on markets less vulnerable to automation.

💡services

For startups without coding skills, the speaker recommends focusing on building services to help businesses integrate GPT-5 into their operations. Coding startups could offer GPT-5 software services, targeting niche problems not easily replicated by the core model.

💡micro niche

When using GPT-5 to build software services, the advice is to start by solving micro niche problems vs. attempting to recreate the full model. Meet focused customer needs first, then expand capabilities over time.

💡automation

The speaker cautions startups to consider which markets may be taken over by AI automation in the future. Certain direct services like web design agencies face significant displacement risk from evolving generative AI.

💡API

Much of the discussion focuses on startups strategically leveraging GPT-5 via its API rather than attempting to compete with it directly. Continued API access is vital for many new innovations built atop AI models.

💡change

A repeated theme is the transformative nature of advances like GPT-5, which may drive changes across industries and society compared to past technological shifts like the internet. Monitoring these changes closely is key.

Highlights

GPT-5 will focus on improving reasoning ability, reliability, customizability, and personalization.

Startups need to provide contextual optimization rather than just behavior optimization to differentiate from large AI companies.

Don't build startups that just rewrap GPT-4 APIs, as they will get disrupted when GPT-5 releases.

Focus on solving micro niche problems rather than trying to reinvent the wheel.

Build services catered to markets resilient to AI disruption.

Non-coders should focus on building integrator services using automation platforms.

Coders should build niche SaaS products that leverage AI APIs.

Solve micro problems first, additional value can be layered on later.

Serve markets that won't disappear with future AI advances.

Don't build services like web design agencies that will be automated.

Most businesses are unaware of AI advances, so huge opportunity exists.

Many missed the GPT-3.5 and GPT-4 hype, GPT-5 will wake them up.

Stay on top of AI advances, as they are like a new internet.

Check out concept to software videos for tangible startup ideas.

We're still early, but don't wait until 2030 to jump on the AI train.

Transcripts

play00:00

most important areas of progress will be

play00:02

around reasoning ability right so in a

play00:04

recent conference we got a little bit of

play00:07

information about gbt 5 that will be

play00:09

coming out and possibly coming out this

play00:11

year in today's video we're going to

play00:12

learn everything that happened yesterday

play00:14

at this event we're going to learn what

play00:15

this means for people trying to do

play00:16

startups we're going to learn what gbt 5

play00:19

even is and overall just get a premise

play00:21

and idea of what is going to happen in

play00:23

2024 when it comes to AI welcome back

play00:25

y'all so I learned about this because of

play00:27

the fact that I saw trending here and I

play00:28

saw GTB T5 a lot of times this kind of

play00:31

stuff is just clickbait so I'm just like

play00:32

you know what the heck could this be I

play00:34

clicked it and this is very real I made

play00:36

a tweet on it and basically this is

play00:38

going to be my video on it this is going

play00:39

to give my perspective on what this

play00:41

means I'm going to give Insight of if

play00:43

you want to start a company in 2024 when

play00:45

it comes to AI like what to do in this

play00:48

context as this changes a lot this

play00:50

stuff's moving extremely fast like I

play00:54

don't even know what the next 5 years

play00:55

are going to look like this is going to

play00:56

be pretty crazy stuff y'all this is

play00:57

pretty transformative stuff but to start

play00:59

off in today's to video let's just go

play01:01

watch a video from this conference to

play01:03

give us more context of what's incurring

play01:05

here side note make sure to follow me on

play01:07

Twitter it is in the description down

play01:09

below but let's go and watch this clip

play01:10

when you look at the next two years what

play01:12

what do you think some of the the key

play01:14

Milestones will be multimodality will

play01:17

definitely be important we started

play01:19

speech in speech out speech in speech

play01:21

out images eventually video clearly

play01:24

people really want that we launched

play01:26

images and audio and it had a much

play01:29

stronger respon than we expected we'll

play01:31

be able to push that much further but

play01:33

maybe the most important areas of

play01:36

progress will be around reasoning

play01:37

ability right now gp4 can reason in only

play01:40

extremely limited ways and also

play01:43

reliability you know if you if you ask

play01:45

GPT for most questions 10,000 times one

play01:48

of those 10,000 is probably pretty good

play01:50

but it doesn't always know which one and

play01:52

you'd like to get the best response of

play01:54

10,000 each time so that'll be that one

play01:56

thing I want to point out when he starts

play01:58

diving into this conversation ation this

play02:00

is basically he's going to describe how

play02:03

gbt should be applied in future context

play02:06

and let's translate this he's basically

play02:09

saying don't mess with me in this Niche

play02:11

that I'm about to say right now which is

play02:13

a pretty big

play02:15

Niche that that that increase in

play02:17

reliability will be important

play02:19

customizability and personalization will

play02:21

also be very important people want a

play02:24

very different very different things out

play02:25

of gp4 different styles you know

play02:28

different sets of assumptions

play02:30

we'll make all that possible and then

play02:31

also the ability to have it use your own

play02:33

data so the ability to know about you

play02:36

your email your calendar how you like

play02:38

appointments

play02:39

booked that's

play02:41

huge that is like basically kind of

play02:44

entertaining an idea of an operating

play02:46

system in a way in the sense of that

play02:49

movie Her Like I referenced in the other

play02:51

video but imagine basically a operating

play02:54

system that knew all the data knew

play02:56

everything about you and communicated

play02:57

with you like a human I mean they're

play03:00

playing they're basically playing like

play03:01

we want the big part of the pie here

play03:04

connected to other outside data sources

play03:05

all of that those will be some of the

play03:07

most important areas of improvement okay

play03:10

so from this event we got a bunch of

play03:12

other information as you see from our

play03:14

Twitter post here from Jason you can go

play03:16

aead and check him out on Twitter let's

play03:18

go ahead and go over the ones that seem

play03:20

relevant to me and cool stuff we should

play03:22

ask so coming up to this Twitter post

play03:24

this was actually the original one that

play03:26

I found that I was like okay is this

play03:27

real and basic basically this person

play03:30

gave us a ton of information when it

play03:32

comes to this I even said my own little

play03:34

tweet you can go and like it if you want

play03:35

to like it but if I scroll down here I

play03:37

believe he dived a little bit deeper

play03:39

okay so yeah three main points here so

play03:41

the first one is that startups need to

play03:43

contextualize optimization more than

play03:45

Behavior optimization basically what

play03:47

this is saying is a lot of startups are

play03:50

going to be chopped at the knees if the

play03:54

market they're trying to serve is going

play03:56

to be the same Market Microsoft trying

play03:58

is trying to serve or sorry open AI a

play04:01

nonprofit uh is trying to serve so let

play04:04

me go ahead and just describe that a

play04:05

little bit more General so the idea here

play04:07

is this kind of okay actually I can give

play04:09

a real example here as we know in

play04:11

November we had Dev day what happened

play04:13

during Dev day that we knew basically

play04:15

chopped a lot of startups at the needs

play04:17

the gbt store so it's that kind of idea

play04:21

people that TR create startups that are

play04:23

basically they call them rapping are you

play04:25

if you're rapping uh basically chat gbt

play04:27

so in this context I have Chad gbt 4 and

play04:31

then I have mega AI all Mega AI is Chad

play04:34

gbt 4 but just rebranded as Mega AI

play04:37

these kind of companies are going to get

play04:39

absolutely clobbered and we already saw

play04:41

that with Dev day so this is an example

play04:44

of what's incurring here in the context

play04:46

of hey when we are going to go to gbt 5

play04:50

and when this incurs you don't want to

play04:52

be in a position where basically the

play04:54

functionality and the value you provide

play04:57

gbt 5 can do it as well because at that

play05:00

point open AI just says goodbye Avid Zin

play05:04

check out number two second one this is

play05:07

very interesting y'all so from the event

play05:09

they found that obviously the API will

play05:11

get better I hope so outages can be

play05:14

annoying during development but also

play05:16

cheaper this is pretty crazy y'all when

play05:19

uh this whole thing started uh Chad gbt

play05:22

open AI like the prices are pretty

play05:23

expensive but then Dev day incurs and we

play05:26

got a slash by

play05:28

50% this was very advantageous and

play05:31

really good cuz it's just like holy

play05:32

smokes like that really helps with cost

play05:34

and processing of data but it was a step

play05:37

in the direction that I wasn't expecting

play05:38

I was actually expecting stuff to get

play05:40

more expensive in the context of API

play05:41

usage but they're actually going

play05:42

opposite obviously because the Market's

play05:44

getting more competitive with Google and

play05:46

these other companies like Claude and

play05:47

the thropic saying that though it

play05:49

becoming cheaper is really positive I'm

play05:51

really excited about that but also what

play05:54

that does is that puts us in a position

play05:57

where basically only the best IDE ideas

play05:59

only the best value points and

play06:02

businesses are really going to stand out

play06:03

because end of the day I don't want to

play06:05

race to the bottom when it comes to you

play06:08

know certain certain services and values

play06:10

you provide in the context of API that

play06:12

might have been a little confusing I

play06:13

also want to point out uh near the

play06:14

latter half of this video I'm going to

play06:15

go on a whiteboard and basically draw

play06:17

out two different Paths of different AI

play06:19

businesses you could start in this

play06:21

basically in this new context and number

play06:23

three kind of addressed what I said

play06:25

earlier basically you don't want to

play06:26

create maybe a reskin or a of gbt 4 but

play06:31

this rap is better at reasoning because

play06:33

end of the day the model is going to

play06:34

improve and then your company is going

play06:36

to go kapoot that being said that kind

play06:39

of covers most of gbt 5 in regards to

play06:42

like actual relevant information we got

play06:44

from this event but we want to keep

play06:46

something in mind here y'all a lot of

play06:47

people are asleep right now if you're

play06:49

watching this video you're already part

play06:51

of like the tenth of a tenth of a tenth

play06:53

of a population that even realizes this

play06:55

stuff's incurring a lot of people are

play06:57

asleep at the wheel people that were

play06:58

asleep for gbt okay I understand if you

play07:00

if you were asleep for gbt 3.5 it is

play07:02

what it is I mean the alarm the alarm

play07:04

kind of went like 3.5 is very impressive

play07:06

in the beginning and then obviously the

play07:08

four was just like okay now we got we're

play07:09

leaps ahead of it but if you were asleep

play07:11

for

play07:12

3.5 and you were asleep for four the

play07:15

people that were asleep during these two

play07:17

models are 100% going to be woken up by

play07:20

gbt 5 this is going to be a knock at the

play07:22

door like hey by the way yeah this

play07:24

wasn't a fad this is here to stay so

play07:26

knowing that obviously it's it's going

play07:28

to be in our best interest to make sure

play07:30

we stay on top of this kind of stuff

play07:31

because the fact that this isn't a joke

play07:33

this is like the our our version of the

play07:35

new internet uh because this is going to

play07:37

change a lot of stuff that being said

play07:39

let's get to that latter half of this

play07:41

video where I go on whiteboard and show

play07:42

you two different routes you can take

play07:44

all right so we got this person right

play07:46

here I don't know who this is but uh

play07:48

they got a really nice sweater um from

play07:50

here though we got two routes so we got

play07:54

a coding route and we got a non-coding

play07:56

route I think that's the best way to

play07:57

structure this to be honest with you as

play07:59

both a is API but one's non-coding one's

play08:00

coding so if I'm and I can't really

play08:04

actually I can let me do this y'all I'm

play08:06

going to do coding and then we'll do uh

play08:12

non-coding okay I'm gon make this as

play08:15

fast as possible because to be honest

play08:17

with y'all YouTube for some reason only

play08:19

likes to push my shorter videos maybe

play08:22

when I get a larger audience it wants to

play08:23

push my longer videos who knows that's

play08:25

just how you YouTube algorithm works I

play08:27

dive into a lot of the stuff I'm about

play08:28

to describe right now a ton in all the

play08:30

playlists you can see it from my channel

play08:31

but let's just put it this way if you

play08:33

have no coding experience and you don't

play08:35

want to take the leap to coding your

play08:37

best route here in the context of

play08:39

starting a business is going to be

play08:40

Services plain and simple Services

play08:43

that's going to be your best route not

play08:45

creating an AI image that goes on a

play08:47

T-shirt and then selling the t-shirt on

play08:49

Etsy it's going to be services this is

play08:50

going to be your best route here as your

play08:53

underlying business itself can be

play08:54

scalable using AI automation but the

play08:56

actual service itself has value because

play08:59

basically majority of businesses don't

play09:02

even know what the heck AI is or asleep

play09:04

at the wheel and when this thing hits

play09:06

the fan a lot of them are going to be

play09:07

like how the heck do I integrate this so

play09:10

non-coding is going to be Services

play09:11

coding is going to be services but

play09:14

software is a

play09:16

service if you are coding so basically

play09:19

for non-coding your advantage or your

play09:22

point your direction is going to be

play09:23

building pre-built services that are

play09:25

kind of like I'm ready to integrate this

play09:27

into your back end XYZ

play09:29

XYZ using automation platforms like make

play09:32

zapier using open a API other

play09:35

alternative apis as well if you choose

play09:36

to do so I prefer open AI that being

play09:39

said for coding we're going to be just

play09:40

doing services so your next question

play09:41

might be okay but in coding context we

play09:44

are just accessing the API directly

play09:46

therefore how do we create a software

play09:48

that isn't just a rewrap of the API as

play09:51

we've seen a that is not the way to go

play09:52

with this new news and honestly that has

play09:54

never been the way to go the way you

play09:56

want to approach this is very simple

play09:58

people overthink things you're not going

play10:00

to we're not trying to reent the wheel

play10:01

here I'm going to give you two pieces of

play10:03

advice first piece of advice is you want

play10:06

to basically solve a Micro Niche problem

play10:10

solve a micro problem Corbin that's not

play10:13

going to make enough money Corbin that's

play10:14

not a good idea Corbin trust me when you

play10:18

start micro and you solve a Micro Niche

play10:20

problem

play10:21

software you will realize through the

play10:23

Journey and through the process there

play10:24

will be a ton of other stuff you add on

play10:26

later on as the pain points become more

play10:28

relevant to you also want to point out

play10:29

if you if you reached to this point of

play10:30

this if you reached to this point of the

play10:32

video and you're like you know what I

play10:33

want more tangible ideas go on and check

play10:35

out my playlist here called from concept

play10:37

to software I give there's two videos on

play10:39

it one's like a Google Maps example and

play10:42

one's a uh three different ones like

play10:44

twitch Gmail like I actually give like

play10:46

applicable examples so you can see that

play10:47

being said micro then the value points

play10:50

will come later second thing who are we

play10:54

serving whoever you

play10:56

serve play a little mining experiment of

play10:59

yourself gbt 5 comes AGI comes is the

play11:04

underlying Market you're

play11:05

serving going to disappear or be taken

play11:08

over by AI let me give you a good a

play11:10

really good example of this website

play11:12

design do not start a website design

play11:15

agency in 2024 you're going to get

play11:18

absolutely chopped at the knees when

play11:20

this because AI is going to be able to

play11:21

build websites give it a couple more

play11:23

months give it a couple more years I

play11:24

know y'all see those little Twitter

play11:26

things where it's like build a website

play11:27

in 60 seconds that's cap but I'm saying

play11:30

genuinely there will be a point in the

play11:31

future where you'll be able to put on a

play11:32

prompt and it'll be able to make the

play11:33

entire website for most cases too in a

play11:35

lot of businesses they don't even need

play11:36

anything complex and need a landing page

play11:38

contact home whatever may be don't do

play11:40

that Orient it Point your services

play11:44

towards a demographic and a population

play11:47

that's going to be resilient through

play11:48

this AI Revolution so that just about

play11:50

does it if you made it to this point in

play11:51

the video that means everything I kind

play11:53

of described there when starting a

play11:54

business seemed pretty relevant to you

play11:56

so check out my channel if you haven't

play11:57

already as I show both Pathways and give

play12:00

you Insight like real world Insight on

play12:02

how to basically approach this as we're

play12:04

still really early to the market here

play12:06

but if you're watching this in 2030 I'm

play12:08

sorry you missed the train maybe the

play12:10

next train is AR VR who knows without

play12:13

further Ado y'all make sure to like you

play12:15

feel like you learned something check

play12:16

out the playlist at the end and I'll see

play12:17

you in the next video so that just about

play12:18

does it in this video we learned a

play12:20

little about gbt 5 and its implications

play12:22

holy smokes this stuff is moving really

play12:24

fast in this industry check out that

play12:26

playlist right there if it interests you

play12:28

you YouTu told me you would like that I

play12:30

don't know and then hit that button

play12:31

right there it it does something I think

play12:33

it does like an explosion effect not too

play12:35

sure