GPT‑4 Mini Nears AGI, Costs Skyrocket
Key Points
- OpenAI has quietly released the new “03” model to a very limited pool of vetted researchers for safety testing, with a public “mini” version slated for January and a full rollout planned for the following year.
- Early testers say the 03 model is edging toward artificial general intelligence, prompting OpenAI to develop unprecedented alignment and red‑team safety measures before broader deployment.
- Running the most capable version of 03 on the ARC AGI benchmark costs roughly $1,000 per task (and $100 for the mini), prices most users are unwilling or unable to pay for routine answers.
- By 2025, raw intelligence will no longer be the primary constraint; societies will match tasks to the appropriate level of AI, reserving the most powerful models only for problems that outstrip human expertise.
- The current hype cycle of ever‑more intelligent language models will give way to pragmatic, application‑focused use, with a clearer understanding of each system’s true capabilities.
Full Transcript
# GPT‑4 Mini Nears AGI, Costs Skyrocket **Source:** [https://www.youtube.com/watch?v=klDJJFB6HAE](https://www.youtube.com/watch?v=klDJJFB6HAE) **Duration:** 00:04:04 ## Summary - OpenAI has quietly released the new “03” model to a very limited pool of vetted researchers for safety testing, with a public “mini” version slated for January and a full rollout planned for the following year. - Early testers say the 03 model is edging toward artificial general intelligence, prompting OpenAI to develop unprecedented alignment and red‑team safety measures before broader deployment. - Running the most capable version of 03 on the ARC AGI benchmark costs roughly $1,000 per task (and $100 for the mini), prices most users are unwilling or unable to pay for routine answers. - By 2025, raw intelligence will no longer be the primary constraint; societies will match tasks to the appropriate level of AI, reserving the most powerful models only for problems that outstrip human expertise. - The current hype cycle of ever‑more intelligent language models will give way to pragmatic, application‑focused use, with a clearer understanding of each system’s true capabilities. ## Sections - [00:00:00](https://www.youtube.com/watch?v=klDJJFB6HAE&t=0s) **OpenAI's Limited GPT‑03 Release** - OpenAI has handed the highly capable GPT‑03 model to a tiny, vetted researcher pool for safety red‑team work, emphasizing its near‑AGI performance, upcoming public rollout, and the significant economic and cost considerations of its broader adoption. ## Full Transcript
at 10:00 a.m. today December 20th open
AI closed their 12 days of open aai ship
Miss by releasing the 03 model to a very
very very tiny group of researchers you
are not going to get it I am not going
to get it this is for safety researchers
only and you have to apply and you have
to be cleared the reason for that is
that they are going to be releasing this
to the public in the next year they say
in January for the Mini version and then
after that for the the full version of
03 but they need to complete red teaming
they need to complete safety work first
and that safety work has only become
more important as the model has become
more capable in fact open AI said they
had to Pioneer a new kind of alignment
testing and safety work just for the 03
model because it's so capable and that
brings me to sort of the big headline
here
03 according to the people who have
played with it and tested it is
approaching general intelligence
and we should start to think about where
we want to use it and apply it if it is
indeed going to be an autonomous system
capable of doing most economically
valuable work more effectively than
humans which is the the rough rule that
uh open AI keeps for what artificial
general intelligence
means the trick is it's not
cheap so even as we think about what to
do with it we have to think about what
we're willing to spend on it and I think
that for those of us who are listening
to this and saying well there goes my
job it's not exactly that clear I looked
at a graph displaying the costs of
running heavy compute tasks appropriate
for
03 in the arc AGI test
suite and I'll link The Arc AGI test
report for 03 here it is over
,000 per task and I want to ask you are
you you ready to pay $1,000 for an
answer from chat
GPT I not very many of us
are are you ready
to are you ready to pay $100 for an
answer from 03 mini that's the cost that
03 mini was running not very many of us
are and so the reason I call that out is
because we are going to live in a world
in 2025
where
intelligence is not a bottleneck
anymore and instead we have to think
about how to apply the correct level of
intelligence for the task that we want
to accomplish and 99% of the time it is
going to be less intelligence than the
maximum available and we will bring in
03 as a society when we really need to
solve a hard problem like if we want to
solve something that a PhD can't
effectively solve that's fine
we would bring in 03 for that but in
most cases for most of our work we
wouldn't need it and we would have
plenty of cheap capable intelligent
models and so we've been in this Rush
since
2022 of getting more and more
intelligent models and sort of being in
awe of how well they do at mimicking
human language and intelligence and
calling out their shortcomings that's
been the story the story in the years to
come is going to be we don't actually
know all the capabilities of these
systems sometimes their intelligence
outstrips us we use them for specific
applications and for most of us
intelligence is just like the air we
breathe it's no longer a bottleneck we
can have all the intelligence we need at
our fingertips it's just about getting
the right intelligence into place that
is a massive change uh and it's going to
take a long time for us to figure out
what it means and 2025 is the year we're
going to start to do that so 03 is here
and we're all going to find out what
that means uh when it starts to come out
next year cheers