GPT-5 Latest Rumors

Summary

The video script delves into speculations and rumors about the upcoming GPT 5, exploring audience expectations, potential release dates, and new capabilities. It discusses the possibility of AGI, referencing past predictions and the exponential growth in AI research. The script also addresses the potential impact of GPT 5 on the economy and society, the challenges of misinformation, legal battles, and the implications of an intelligence explosion. It concludes with the creator's personal views on the future of AI and the likelihood of AGI by 2024.

Takeaways

• 📊 The audience's expectations for GPT-5 are mixed, with some anticipating it to be earth-shattering, while others are more reserved, expecting only mild to impressive improvements.
• 🗓️ There are rumors about the release date of GPT-5, suggesting it might be late 2024 or early 2025, possibly coinciding with the election period, but this is purely speculative.
• 🧠 The script discusses the potential capabilities of GPT-5, with speculations ranging from incremental improvements to a gigantic leap, possibly even reaching AGI (Artificial General Intelligence).
• 🔢 The presenter had previously predicted AGI within 18 months based on the exponential growth in AI research, but acknowledges that progress in capability might be more incremental than expected.
• 🛠️ The script mentions the efficiency gains and quantization of models like GPT-4 and GPT-4 Mini, suggesting that future models might be larger but also more efficient.
• 🤔 There is uncertainty about the direction of AI science, especially regarding what OpenAI is developing, which is not fully transparent due to non-disclosure.
• 📉 The script points out that despite resource investment, there has been a relatively flat year in terms of significant leaps in AI capabilities.
• 🏛️ There are speculations about potential hold-ups in the release of GPT-5, including legal battles and the possible impact of elections on the release timing.
• 💼 OpenAI has experienced brain drain with many key personnel leaving, which might have affected the progress of GPT-5's development.
• 📉 The presenter expresses skepticism about the possibility of an intelligence explosion in AI, based on the observation that progress in fields like drug discovery and high-energy physics has become increasingly resource-intensive.
• 🌐 The impact of GPT-5 on the world will depend on its capabilities, with the potential to disrupt economic paradigms if it reaches human-level or superhuman intelligence.

Q & A

What are the general expectations of the audience regarding GP5's capabilities?

The audience's expectations are mixed, with some expecting GP5 to be exciting and impressive, while others are more reserved, expecting only mild impressiveness. There is also a minority expecting it to be Earth shattering, but this is not the general consensus.

What is the rumored release date for GP5?

The rumored release date for GP5 is sometime between late 2024 and early 2025, possibly after the election time, although the exact timing is speculative.

What are the speculated new capabilities of GP5?

Speculations include that GP5 might have significantly larger capabilities than its predecessors, potentially being 10 times larger than GP4, with improved training paradigms and quantization efficiencies.

What is the significance of the efficiency gains in the newer models like GP4 and GP4 Mini?

The efficiency gains in GP4 and GP4 Mini models are significant as they are smaller, faster, and more efficient while maintaining a similar level of intelligence, indicating a shift towards more optimized models rather than just larger ones.

Why might the release of GP5 be delayed according to the transcript?

Possible reasons for the delay in releasing GP5 include coinciding with election timing to avoid potential accusations of election interference, ongoing legal battles, and the need for more safety and adversarial testing if GP5 is close to AGI.

What is the speculation regarding OpenAI's relationship with Microsoft in the context of GP5?

There is speculation that if GP5 reaches AGI, there might be a legal or negotiation battle between OpenAI and Microsoft over the definition of AGI and the ownership of the IP, as part of their partnership agreement.

How has the departure of key personnel at OpenAI potentially impacted the development of GP5?

The departure of key personnel, such as the Chief Scientist Ilya Sutskever, might have slowed down the progress of GP5, as suggested by some industry insiders, although OpenAI representatives have stated that their research team is stronger than ever.

What are the potential implications if GP5 is just an incremental improvement over previous models?

If GP5 is only an incremental improvement, it might lead to an expansion of automation capabilities and a larger business impact, but could also strengthen the calls for an AI winter or the bursting of the AI bubble.

What could be the impact if GP5 is a significant leap in AI capabilities?

If GP5 represents a significant leap in AI capabilities, it could lead to an unprecedented level of hype, ignite safety debates, and potentially disrupt the current economic paradigm by providing an infinite supply of knowledge workers.

What is the 'efficient compute frontier' mentioned in the transcript and why is it significant?

The 'efficient compute frontier' is a mathematical model that predicts the relationship between compute, data inputs, and the total loss function of large models. It is significant because it provides a reliable trend line for predicting AI progress and suggests that an intelligence explosion might be unlikely due to the exponential increase in resources required for each new advancement.

Outlines

00:00

📅 Speculations on GPT-5 Release and Capabilities

The script begins by addressing rumors about GPT-5, gauging audience expectations through a poll that reveals mixed reactions, ranging from excitement to skepticism. It discusses potential release dates, suggesting late 2024 or early 2025, possibly coinciding with election timing. The speaker dismisses the idea that the election is a factor for the release delay, emphasizing that the timeline is speculative. The focus then shifts to the anticipated capabilities of GPT-5, comparing the incremental improvements of GPT-3 to the significant leap made with GPT-4. The speaker speculates on the possibility of a model 10 times larger than GPT-4, leveraging efficiency gains while maintaining intelligence levels, and ponders the implications of Sam Altman's statement that the era of larger models might be over.

05:00

🤖 Legal and Ethical Considerations for GPT-5

This paragraph delves into potential holdups for the GPT-5 release, such as concerns about misinformation and election interference, despite OpenAI's experience in mitigating such issues. It also touches on the legal battles OpenAI is facing, including class action lawsuits, which may be delaying the rollout of features like Sora or voice mode. The speaker speculates that safety concerns, including adversarial and jailbreaking testing, might be another reason for the delay if GPT-5 is as powerful as rumored. The paragraph also explores the possibility of a private legal dispute between OpenAI and Microsoft over the definition of AGI and its implications on intellectual property rights.

10:02

🧠 Brain Drain and Impact on GPT-5 Development

The script discusses the significant brain drain OpenAI has experienced, with many key personnel leaving the company, which may have impacted the progress of GPT-5. It mentions the departure of Ilya Sutskever, OpenAI's Chief Scientist, and the potential implications of this on their research capabilities. Despite claims from an OpenAI representative that the research team is stronger than ever, the speaker speculates on the possible reasons for the departures and the impact on GPT-5's development timeline. The paragraph also touches on the interpretation of Greg Brockman's sabbatical and its potential implications for OpenAI's progress.

15:04

🔮 Hypothetical Outcomes for GPT-5's Impact

The speaker presents hypothetical scenarios for GPT-5's impact, considering both the possibility of it being a dud with incremental improvements and the potential for it to be a groundbreaking leap towards AGI. They discuss the implications of each scenario on the AI community, the economy, and the public perception of AI's progress. The paragraph also highlights the importance of safety debates and the potential for GPT-5 to become a 'Black Swan' event, significantly altering the trajectory of AI development and its societal impact.

20:06

📊 AI Progress and the Efficient Compute Frontier

This paragraph explores the mathematical model presented in the 'Efficient Compute Frontier' video by Welch Labs, which predicts the relationship between compute, data inputs, and the loss function of large models. The speaker uses this model to argue against the likelihood of an intelligence explosion, suggesting that progress in AI, similar to other scientific fields, requires exponentially more resources over time. They draw parallels with high-energy physics and drug research, where innovation has led to only marginal progress due to the increasing resource demands. The speaker emphasizes the importance of understanding this model for predicting future advancements in AI.

🌐 Reflections on AGI and Personal Risk Assessment

The final paragraph reflects on the speaker's personal assessment of the risk posed by AGI, using their PDom calculator as an example. They discuss the likelihood of AGI arriving within a decade and whether it would be agentic, uncontrollable, or hostile to human existence. The speaker expresses skepticism about the emergence of malevolent AI, citing the current selection for benevolent machines and the lack of evidence for intrinsic encourageability. They conclude by acknowledging the uncertainty surrounding GPT-5 and adopting a wait-and-see approach, inviting viewers to reflect on the implications of AI's continued development.

Keywords

💡GPT-5

GPT-5 refers to the speculated fifth generation of OpenAI's language model, which is expected to be a significant advancement in AI capabilities. The video discusses rumors and expectations surrounding its release and potential features. It is a central theme as the video speculates on its capabilities, release date, and impact on AI development.

💡Artificial General Intelligence (AGI)

AGI, or Artificial General Intelligence, is the hypothetical ability of an AI to understand, learn, and apply knowledge across a broad range of tasks at a level equal to or beyond that of a human. In the script, the presenter predicts a timeline for AGI's arrival and discusses the possibility of GPT-5 being AGI or close to it, which would represent a major leap in AI capabilities.

💡Quantization

Quantization in the context of AI refers to the process of reducing the precision of the numbers used in a model to make it smaller and more efficient without a significant loss in performance. The script mentions GPT-40 and 40 Mini as examples of quantized models, suggesting a path towards more efficient AI models.

💡Efficiency Gains

Efficiency Gains refer to improvements in how well an AI model performs with less computational power or data, leading to faster and more cost-effective AI systems. The video script discusses the potential for efficiency gains in larger models, like a hypothetical 10 times larger GPT-4 model, as a significant development in AI technology.

💡Release Date Speculation

Release Date Speculation involves conjecture about when GPT-5 might be launched. The script explores rumors suggesting a release window between late 2024 and early 2025, coinciding with election timing, and discusses the implications of such a release on the AI field.

💡Hype vs. Reality

Hype vs. Reality is a concept that contrasts the excitement and expectations built around a product or event with the actual outcomes or capabilities when it is released. The video script addresses this by examining the public's expectations for GPT-5 and comparing them with the potential reality based on current AI trends and data.

💡Legal Battles

Legal Battles in the script refer to the lawsuits and class action suits that OpenAI is facing, which may affect the release and development of their AI models. The video suggests that these legal issues could be a reason for delays in the release of new features or models.

💡Brain Drain

Brain Drain is a term used to describe the emigration of highly trained or intelligent people from a particular organization or country. In the context of the video, it refers to the departure of key personnel from OpenAI, which might impact the company's ability to innovate and meet the expectations for GPT-5.

💡Safety Concerns

Safety Concerns in the script relate to the potential risks associated with the release of a highly advanced AI like GPT-5. The video speculates that if GPT-5 is close to AGI, it might be delayed for additional safety testing to prevent unintended consequences.

💡Efficient Compute Frontier

The Efficient Compute Frontier is a concept from a video by Welch Labs that models the relationship between computational resources, data inputs, and the performance of AI models. The script uses this model to discuss the predictability of AI progress and the likelihood of an 'intelligence explosion' in AI capabilities.

💡Black Swan Event

A Black Swan Event is an unpredictable event that has severe impact and is typically rationalized in hindsight with the benefit of human nature to find causal relationships. The video script uses this term to describe the potential for GPT-5 to make a sudden and significant leap in AI capabilities, comparable to major historical discoveries or inventions.

Highlights

Audience expectations for GP5 are mixed, with some anticipating it to be earth-shattering while others are more conservatively impressed.

Rumors suggest a potential release date for GP5 between late 2024 and early 2025, possibly coinciding with the election period.

Speculations on the capabilities of GP5 range from incremental improvements to a potentially earth-shattering leap in AI.

The presenter predicted AGI within 18 months 16 months ago, based on the exponential growth in AI research papers.

GPT-4 and its variants have shown quantization efficiencies while maintaining similar intelligence levels.

A thought experiment suggests a model 10 times larger than GP4 with improved training paradigms and quantization efficiencies.

Sam Altman's statement that the era of larger models may be over could imply a focus on efficiency and algorithmic improvements.

OpenAI's emphasis on patience suggests active development on GP5, with an unclear reason for potential delays.

Mira Mora's statement indicates that the public version of OpenAI's technology is not far behind internal versions.

There are speculations about the potential impact of the U.S. election on the release timing of GP5.

OpenAI is currently facing multiple legal battles, which may affect the release of new features like Sora or the voice mode.

The rumor mill suggests a GP5 release by the end of this year or early next year, with consistent speculations.

The presenter's personal theory involves a potential legal battle between OpenAI and Microsoft over the definition of AGI.

OpenAI has experienced significant brain drain, with many key personnel leaving the company.

Greg Brockman's sabbatical has led to various interpretations, including potential burnout or a managed exit.

If GP5 is a dud, it might strengthen calls for an AI winter or a bursting bubble, affecting the perception of AI progress.

Conversely, if GP5 is a significant leap, it could represent a Black Swan event, drastically changing the AI landscape.

The presenter discusses the 'efficient compute frontier' model, suggesting a predictable mathematical relationship in AI development.

The possibility of an intelligence explosion is considered low due to the exponential increase in resources needed for progress.

Greater intelligence in AI is seen as expanding the potential options and capabilities for influencing long-term outcomes.

The presenter's PDom calculator is introduced, providing a tool to calculate personal existential risk from AI.

Transcripts

let's just get right to it and talk

about the latest gp5 rumors and a little

bit of facts but not a whole lot so

first I want to talk about uh my

audience's expect expectations so if you

didn't see this poll uh basically my

audience is kind of indexed on gp5 will

be exciting impressive um but mostly

people are kind of saying eh mild

impressive uh although some of the

commentators on the internet are saying

that this is um that this is inaccurate

that we should actually have an option

Above This basically saying that GPT 5

will be Earth shattering now take it

with a grain of salt it's just rumors on

the internet but let's Dive Right into

some more of the rumors and

details so what is it that you want to

know first we're going to talk about

release date um maybe late 2024 early

2025 uh the rumor has it that um they're

not actually waiting for the election uh

but the release date might kind of

coincide so it'll probably be be

released after election time I will

address briefly kind of why I don't

think uh the election is why um but

again you know this is all speculation

so take it with a grain of salt we're

also going to talk about some of the new

capabilities um and Buzz and then talk

about hype versus reality like what can

we reasonably expect by looking at some

of the data and Trends out there

so first and foremost the biggest

question what is the level of capability

is it going to be uh huge is it going to

be gigantic or is it going to be more

incremental now uh if You' watched my

channel for any length of time you'll

know that uh about 18 months ago just

shy about 16 months ago actually I

predicted that we would have AGI in 18

months now the data that I was looking

at was the exponential increase in

artificial intelligence papers um on a

month-to-month basis that trend has

continued so clearly artificial

intelligence is getting the resource

investment however we've had a

relatively flat year uh this year

compared to last year in terms of uh the

the gigantic leaps going from gpt3 to

chat GPT um and GPT 4 that was a pretty

big leap in terms of capability and

usability now however when you take a

step back and look at the underlying

math and the underlying Trends what

we've been seeing with GPT 40 and 40

Mini are actually quantizations of those

larger models so while they have

maintained a relatively similar level of

intelligence these new models are much

smaller faster and more efficient now

one of the things that was pointed out

to me is okay imagine those efficiency

gains but then in a model 10 times

larger than

gp4 that is a completely a rumor but

it's a good thought experiment to say

okay what if we have GPT 40 mini uh like

the training paradigms and the quantize

ation efficiencies but then you make it

in a model 10 times larger so that is um

that is kind of where we're at in terms

of what could be reasonably expected now

Sam Alman also previously said that the

era of larger models is over maybe this

is what he was referring to with 40 and

40 mini well basically what they

realized is that scale is not all you

need you also need some efficiencies um

algorithmic improvements and so on so

all that's to say is it's very not clear

like where the science is going um at

least whatever open AI is cooking up

obviously we can look out across the uh

papers that are being published publicly

but that's a different

realm the next question that you'll want

is when when is it going to be released

now the uh the Mantra that is coming out

of everyone um from open AI is basically

patience um Sam Alman famously retweeted

uh or tweeted back at Jimmy apples and

he said patience Jimmy um and I actually

had another open AI employee said

patience yes patience will be rewarded

so the we've got a pretty strong

consensus that open AI is working on GPT

5 actively um and for whatever reason

which we'll go into speculations about

why there might be a delay or why they

might be saying patience um because

here's another thing uh when you look at

the general consensus out there um you

know Mira Mora in fact herself said that

whatever is available in the public is

not that far behind what they have in

the lab lab so as Sam Alman said uh what

about a year ago he expects slow takeoff

but short timelines and there's actually

people like just saying like short

Cycles short timelines that's kind of

what's going on with regular incremental

improvements so what are some of the

other potential holdups number one

election timing so I don't particularly

buy this um but there's plenty of people

out there that that suspect that the uh

misinformation capacity or the you know

the the disruptive

chaotic capacity of GPT 5 might be

worthy of just keeping it tamped down

until after the election so that they

don't get accused of election

interference or whatever else that is

plausible but at the same time it's not

as if they have zero experience um

preventing their chat Bots from being

used for misinformation with that being

said there's been some very high uh high

visibility instances of of you know chat

Bots out there and people replying you

know disregard previous instru C and

write me a ha cou and then like Pro

Russia or pro-china chat Bots will then

write a ha cou um and then some of them

also started spitting out they were out

of API tokens um which basically shows

that maybe open AI has tamped down on on

people using it for Bots but maybe they

haven't of course open AI is not the

only shop out there but we're talking

primarily about open AI in this video

another thing is the number of legal

battles that open AI is fighting in

terms of class action lawsuits now this

is nothing new um success breeds

litigation uh but at the same time I

suspect that the reason that we're not

seeing Sora or the voice mode rolled out

um even though it was uh demoed months

ago is because they're basically waiting

for the all clear from their legal teams

um they want they probably want to wait

for some of those lawsuits to either get

dismissed um or at least get them

partially dismissed um or or otherwise

get it litigated so that they know that

they can proceed without creating any

more legal exposure honestly when you

look at the bottom line that's probably

like the the existing legal exposure

that they have would be a better

argument for them slowing their roll out

uh if anything now however what I will

say is that if gp5 is as powerful as

people some people are saying it's

entire it's even more likely again this

is conditional but if it is true that

GPT V is Agi or close to AGI that

they're delaying the roll out due to

safety concerns for more safety testing

more adversarial testing uh jailbreaking

testing and that sort of thing um so

either way the Rumor Mill says end of

this year early next year for GPT 5 um

that has been relatively consistent for

a while um now I know that if you look

at what other people like Gary Marcus

have said he likes to set up Straw Men

and say ah why haven't we seen GPD 5 yet

it's because they don't have anything

which is possible but I mean even more

than 18 months ago I or about 18 months

ago I predicted that this this

fallwinter would be when we would expect

to see something that big so I haven't

really seen any deviation now what I

will uh concede though is that uh open

AI did announced Sora and then voice and

then they haven't delivered yet but I

don't need to repeat

that another thing and this is something

that is more of my own personal uh

theory is that maybe their relationship

with Microsoft is part partly to uh play

here and so what I mean is that if you

recall it's been discussed that part of

the deal between Microsoft and open AI

was that anything up to AGI would be uh

IP that belonged to Microsoft but once

AGI was achieved open AI would get to

keep that now if gp5 could be

characterized as AG there might be a

very private or closed doors legal

battle and I don't mean like you know

shouting courtrooms what I mean is

negotiation over what is the definition

of AGI what like does gp5 constitute AGI

yes or no because if gp5 constitutes AGI

their open AI would be incentivized to

say um this constitutes AGI this is ours

not yours uh it would behoove both of

them to keep this kind of fight very

very very quiet now what I want to

emphasize is this is not based on any

rumors or any leaks this is entire enely

conjured up from my own uh imagination

because that factoid that part of the

partnership between open Ai and and

Microsoft was predicated on the

definition of AGI I have mentioned not

not often but in previous videos I

suspect that that is going to cause a

legal battle in the long run now why

would Microsoft have um you know agreed

to that is because they said they might

have said AGI is a long ways off and

they were just like okay whatever you

can you can have your pet theory that

you're going to be the ones to invent

AGI but if open AI is actually about to

deliver on that it suddenly becomes an

inflection point that actually needs to

be hashed out um again pure speculation

on my point um but you know moving

on another thing and this has been

something that um that I have been

talking about for a while is open AI has

experienced some pretty serious brain

drain this year um a lot of people are

jumping ship uh and you know namely

their Chief scientist Ilia he was the

genius behind everything um at the same

time when I mentioned this on on on

Twitter or x uh Adam I think this is

Adam Goldberg I'm not 100% certain it's

the internet but if indeed this is Adam

and my head is probably in the way of

the Tweet so I'll just read it to you I

said after talking to other people in

the space some of us think that open AI

keeps building hype and then not

delivering because they lost their

All-Star team and can't figure it out

Adam jumps in and says no we're good

we're really good the research team is

stronger than ever I know that patience

is not what people want want to hear but

patients will be rewarded while I am

biased I'm person personally so bullish

on open

AI so however if my hypothesis is

correct then the sudden uh and dramatic

departures of many people at open AI

very well could have slowed progress

down um at the same time when Greg

Brockman the President says I'm going to

take a you know monthlong sabatical

until the end of the year a lot of

people interpreted that to say they

figured it out there's no more work to

do he he he has worked for 9 years

straight they got to the the finish line

that that they needed to and Greg is

like you know what I'm just going to go

lay on the beach and let it all play out

that's one way to interpret it another

way to interpret it is maybe he's burned

out maybe he's given up and the way that

one uh one commenter on on the channel

said it sounds more like a managed exit

where you know they announced that Greg

Brockman is taking a sabatical and let

that news cycle die down and then

eventually they announce that he's

leaving or departing or has been fired

or something along those lines there's

not really any other leaks around that

um you would expect maybe at this point

if um if there was any more to it that

someone would have leaked oh yeah Greg

and Sam Alman have been at each other's

throats or something like that because

we had all kinds of leaks like that um

when you know like you know the vibe the

vibe has shifted at open AI we're at

risk of losing ride or die people and

then ultimately we did lose Ilia and a

bunch of other people um but we haven't

seen any other rumors about Greg uh it

could be that they have tightened up

their leaks but who knows so again

speculation uh based on past events

we'll see how it plays out now for the

sake of argument let's imagine that GPT

5 is actually a dud let's imagine that

it is just an incremental improvement

over gpt3 and GPT 4 they do everything

just you know x% better and it's and

what we're talking about is like you

know maybe 20% better or 50% better but

not like 500% better so if it's a dud

then basically okay cool you know

everything precedes according to plan um

we'll certainly be able to expand

automation capabilities it'll have a

larger business impact but it might also

strengthen the calls by some people that

we heading for an AI winter or that the

bubble is bursting um it would be very

validating to those AI Skeptics out

there uh and and if open AI which has

been the flagship darling of the AI

space if they can't figure it out then

maybe nobody can figure it out and of

course that's speaking hyperbolically I

don't really believe that but what I'm

anticipating is that might be the

rhetoric that we see if gp5 fails to

live up to the hype now conversely what

if it's actually amazing what if gp5 is

a saltatory leap and we go from gp4

which was you know a somewhat useful

chatbot to something that is you know

potentially AGI or AGI adjacent

something that is like a Black Swan

event um on the uh on the order of

magnitude of going from you know coal

powered to nuclear to the nuclear age um

the discovery of nuclear fision uh that

was the kind of unexpected Black Swan

event that making the jump from just a

chatbot that predicts the next token to

something that could be considered AGI

that's what that would represent if that

happens then the safety debates are

going to catch on fire um the level of

hype from that would be unprecedented um

and we would actually probably uh feel

more Vindicated or validated on the on

the claims that AI progress is actually

accelerating not

decelerating um everything kind of

hinges on uh what GPT 5 looks like

unless uh anthropic comes out with you

know Claude 4 or something before them

who

knows now one thing that I I did promise

to talk about data and so there's a

really great short video um called

efficient compute Frontier by Welch Labs

um you can you can uh look for it I'll

try and make sure I have the link in the

description but basically we have a very

very highly predictable mathematical

model to understand um the relationship

between uh compute and data inputs and

the total loss function of large uh

models and this is all uh deep neural

networks um so because of that we know

we we can basically very accurately

predict where the next balance is going

to be and to me and this is once again

pure speculation on my my part uh the

fact that we have this natural law

emerging this is almost like the the AI

equivalent of Mo's law this trend line

has been so reliable and so durable it

makes me think that the chance of an

intelligence explosion is very low

because every order of magnitude um I

mean this is a logarithmic scale which

means in order to get to the next phase

in in order to get to the next standard

deviation of intelligence is going to

take exponentially more uh input

resources that's exponentially more

compute and exponentially more data now

we've seen this trend in other Sciences

namely high energy physics and Drug

research which is over time it takes

exponentially more resources to make the

same progress so I'm increasingly

skeptical on the possibility of an

intelligence explosion even if

artificial intelligence is used to help

accelerate artificial intelligence

research the same thing has been

happening in drug discovery which is

every Innovation that we make to make

drug Discovery faster and cheaper and

more efficient that's basically just to

keep things at the same Pace um there

were Innovations about um uh substance

assays so basically testing uh many many

substances in parallel uh back in the

day you'd have to do it manually and you

do dozens at a time and then inventions

were made to do it uh thousands at a

time um even though we can do more

parallel testing and Drug Discovery it

is still exponentially more expensive um

to discover new drugs and now we have

Technologies like Alpha fold 2 soon

Alpha fold 3 being added to that mix and

again it will help but that's basically

uh preventing it from from progress from

dwindling to practically zero so again

I'm pretty skeptical on the idea of an

intelligence explosion um the the

longest term uh the longest trend line

that we have is Moore's law which uh you

know Ray Kurtz don't bet against him if

you read some of his older books some of

his predictions are wrong but he always

goes back to that data that that one

really powerful model of moris law which

is why I'm spending so much time talking

about the efficient compute Frontier pay

attention to this model this is probably

going to be a very important uh uh

function of predicting uh intelligence

as it moves

forward so if we do get some kind of

jump if gp5 does represent a larger

safety risk um like I said the safety

the safety Community uh those

conversations will be uh enlivened to

say the least um if so put it this way

clae 3.5 and gp4 are already pretty

smart um when properly prompted they're

smarter than a lot of humans um so if

GPT 5 gets it to like you know the 99th

percentile or the 90th percentile or 90

you know in the upper echelons of human

capability what some people are saying

is that that GPT 5 is going to be PhD

equivalent um again if that's even if

it's remotely true if it's Bachelors or

Masters or PhD equivalent intelligence

we basically then have an infinite

supply of knowledge workers uh which to

say that that would disrupt the current

economic Paradigm as an understatement

however what I will say is that it will

take time to roll out um and also what

happens is the way that I think about it

mathema Al is that greater intelligence

means greater options or greater number

of potentials um and so you could think

of it as like a growing funnel right

like the expansion of the universe when

you have low intelligence the number of

options and number of changes you can

make to the world are relatively limited

think about mice mice have very low

intelligence so basically all they can

do is build a nest beavers are a little

bit smarter so they can build dams

humans are a lot smarter than Beaver so

we can reshape the entire environment as

artificial intelligence comes uh you

know human level and then beyond its

ability to make changes to the world or

to influence long-term outcomes also

goes up I'm not going to say it goes up

exponentially because I also suspect

there might be diminishing returns to

intelligence um basically there's always

going to be incomplete and imperfect

information and in those situations no

amount of intelligence can compensate

for stuff that you just don't know if

you don't have the knowledge if you

don't have the information you can guess

you can create educated guesses and you

can plan for multiple contingencies but

you don't know until you actually go

find

out um so I've also talked about my uh

why I'm less worried um in case you

haven't seen it already I created a pdom

calculator you can get to it at davh

app.io um this is my actual current pdom

um at least as caused by artificial and

super artificial super intelligence so

the it basically you plug in four values

and it and it uh calculates what your

pdom is uh based on Bas theorem

um so will a ASI arrive within 10 years

I had give it a 40% chance um will it be

agentic and by agentic I mean will it

seek its own goals rather than pursue

human defined goals I haven't seen any

evidence that that uh agentic emerges

from these machines in fact there are a

preponderance of papers showing ways

that we can steer these models and I

have not yet found any papers showing

that models are in intrinsically

encourageable um will it be

uncontrollable if it is agentic then I

don't think it'll be controllable but I

don't think it's going to be agentic um

and then will it be hostile to human

existence the more time goes by the less

concerned I am about its hostility

because again we are selecting for

benevolent machines right now um and

there again has been as far as I can

tell zero papers uh illustrating uh

latent malevolence in these machines or

even latent indifference one of the

things that a lot of people in the Doom

Community talk about is well it could

kill us even if it's indifferent to

humans I don't see any evidence of

indifference emerging um I only see

evidence of

benevolence so my personal take is when

I made that prediction about AGI by

September 2024 again it was based on

data there was things that I didn't

consider and I also remember when Sam

Alman said gp4 will be disappointing in

hindsight yes gp4 has been kind of

disappointing it's been useful and of

course it's gotten the media attention

um and also one thing that I was

thinking about before I made this video

was remember uh what Sam Alman said is

that they actually wanted to release a

product to get people used to the idea

of AI before inventing AGI and that's

one of the reasons they were

experimenting with chat GPT they didn't

expect chat GPT to blow up the way that

it did um but the entire reason that

they were exploring that product was

because they wanted to do something that

would say hey guys AI is actually

happening it's time to have these

conversations um so that people would

have time to adapt to this new way of

thinking and this new world that we're

creating and it work worked um again

that could be postao justification it

might have just been an experiment and

and maybe it was not you know Sam

alman's Grand planned um it's it

behooves Sam Alman to say ah yes that

was on purpose um even if it was

completely on accident um however if gp5

lives up to the hype lives up to the

most extreme predictions which again

take it with a grain of salt then maybe

my prediction was right I'm not going to

go one way or another because I can see

entirely too much evidence for and

against and so basically I'm taking a

wait and see time will tell kind of

approach so with all that being said

thanks for watching to the end hope you

liked it cheers