OpenAI's Hype Over Delivery Dilemma
Key Points
- The AI community is caught between the hype surrounding new large language model features—like OpenAI’s Advanced Voice Mode and Sora—and the slower, limited roll‑outs of those features to the broader public.
- OpenAI deliberately fuels hype to maintain its market‑leader image, which helps secure Microsoft’s enterprise deals and justifies its heavy investment, even though many announced capabilities remain in closed beta or delayed.
- This hype‑first strategy isn’t unique to OpenAI; several other LLM providers also prioritize buzz to attract attention, while those that avoid hype do so because their incentives differ.
- The rumored “Strawberry” upgrade—promising enhanced reasoning and autonomous internet navigation—is likely being leaked as part of a hype battle rather than an imminent product launch.
- Recent benchmark advantages of competing models like LLaMA and Anthropic’s Claude Sonnet are prompting developers to consider switching away from ChatGPT, highlighting the gap between OpenAI’s hype and its current performance.
Full Transcript
# OpenAI's Hype Over Delivery Dilemma **Source:** [https://www.youtube.com/watch?v=r-NdPRBPa5E](https://www.youtube.com/watch?v=r-NdPRBPa5E) **Duration:** 00:08:43 ## Summary - The AI community is caught between the hype surrounding new large language model features—like OpenAI’s Advanced Voice Mode and Sora—and the slower, limited roll‑outs of those features to the broader public. - OpenAI deliberately fuels hype to maintain its market‑leader image, which helps secure Microsoft’s enterprise deals and justifies its heavy investment, even though many announced capabilities remain in closed beta or delayed. - This hype‑first strategy isn’t unique to OpenAI; several other LLM providers also prioritize buzz to attract attention, while those that avoid hype do so because their incentives differ. - The rumored “Strawberry” upgrade—promising enhanced reasoning and autonomous internet navigation—is likely being leaked as part of a hype battle rather than an imminent product launch. - Recent benchmark advantages of competing models like LLaMA and Anthropic’s Claude Sonnet are prompting developers to consider switching away from ChatGPT, highlighting the gap between OpenAI’s hype and its current performance. ## Sections - [00:00:00](https://www.youtube.com/watch?v=r-NdPRBPa5E&t=0s) **Hype vs Reality in LLM Rollouts** - The speaker critiques how OpenAI and other model makers prioritize promotional hype over timely, widespread product releases—using advanced voice mode and Sora video generation as examples—to maintain market dominance despite frustrating customers. ## Full Transcript
large language models and the updates
that they get are hard enough to
understand without having a constant
tension between the hype that these llm
makers bring to the table and the actual
product releases that they offer I was
thinking about this because advanced
voice mode which is something that has
been widely hyped that has been released
in a closed beta as of mid August 2024
is still not widely available even
though it was discussed this spring
shown this spring said to be super cool
this spring by open AI Sora also by open
AI is kind of in the same boat we still
don't have widely available video
generation from open AI even though they
announced it even though they had a
whole page dedicated to it what I find
fascinating is that open AI is
deliberately adopting a hype approach
here that makes sense from a game theory
perspective but is super frustrating to
customers
and they're not the only ones there are
a lot of other model makers out there
who are adopting a hype first approach
and the ones that aren't it makes sense
for them given their incentives let me
walk through that sort of comparison
quickly let's start with open AI they
need attention to maintain Market
leadership they need to be shown and
seen to be the leaders in AI to keep
their number one position in usage and
that matters to them because even though
they are very well funded by Microsoft
they still need to show that they're
number one for Microsoft to defend
propose Drive Enterprise deals based on
the open AI model set with very large
companies which is key to Microsoft's
overall monetization strategy and the
way Microsoft is thinking about their
open AI investment so they have to get
attention and that means they have to be
constantly seen as moving in the
direction of significantly improved AI
even if actual shipments to scaled out
user Footprints lag way behind
and that's what we're seeing that's why
Sora is not really widely available yet
that's why advanced voice mode is not
widely available yet and that's why in
the most recent hype example I don't
think strawberry is going to be widely
available for a while what is strawberry
you might ask it is rumored or leaked to
be
the next iteration in reasoning and
autonomous internet navigation from chat
GPT the reason why they decided to leak
it seems pretty clear to me it's a hype
battle and I noticed that the strawberry
leaks really gain speed and momentum
after report started to drop that a lot
of folks with open API pipelines by open
I mean easy to switch out of large
language models like if you're deploying
an application you want to be able to
switch an llm on the back end it's super
easy well once llama released last month
and once uh the latest uh version from
anthropic Claude Sonet
released there was a persistent push to
start to shift those API pipelines over
to other models not chat GPT because
chat GPT was widely perceived as being
lower on a significant range of
benchmarks even their 40 model versus
Sonet versus uh
llama
and as those reports began to circulate
as it began to become apparent that
people who build this space were moving
away from open AI toward a more advanced
model suddenly leaks began to multiply
from open AI that hey we're working on
something new it's called strawberry
it's really cool and then yesterday
August
12th it turns out that they've been
releasing something in the wild in 40
for weeks and not telling anyone about
it and they just sort of had a cryptic
announcement to say hey we've got a new
and improved 40 model in the wild it's
been out there for a few weeks I hope
you've been liking it that is not a
release note that does not help someone
who is trying to understand the wide
latent space that you have in an llm
capability set to actually use that
space to do useful work it just doesn't
work and we need release notes even if
they're hard because we need guidance as
users to start to figure out where to go
next because the chat window doesn't
really tell us anything I've talked
about that in previous previous videos
the the chat window just says say
something and we have to understand
enough of the llm to prompt
appropriately and if I don't know that a
model is upgraded and this is not just
an open AI Problem by the way this is a
larger industry problem with llms right
now if I don't know what the capability
set in the upgrade is it is hard for me
to know how to change my usual prompting
strategy to get more out of the upgrade
so I don't necessarily perceive any
value because the value is in the
response to my prompt and my prompt will
need to change if the latent space in
the model has shifted if the capability
space has
adjusted so all of that to say yes
there's rumors about a new release
called strawberry I wanted to
contextualize it in the larger sort of
hype cycle and I wanted to call out as I
close this video the difference between
this approach and the way meta is
handling things because I think meta
exemplifies sort of the opposite take
meta doesn't have the same set of
incentives they are not trying to monit
their model Mark Zuckerberg has been
extremely clear about that and their
only goal is to build an ecosystem which
means their real value is if a real
model is released widely so that people
can build on it and that is why meta is
shipping widely and letting researchers
letting developers build against their
models that's why llama when it was
released was actually released not just
announced because meta doesn't have the
same need to maintain
hype they just need to build an
ecosystem that really
works now I'm not here to say who's
going to win this race if we step back I
think one of the big concerns I have is
that all of the players who see what is
happening have decided that this is an
important enough moment in history that
they are willing to invest potentially
billions of dollars over and that's not
my take by the way that's actually from
like Google memos that have been
released that say that they're looking
at it from a game theory perspective and
saying we'd rather
overinvestment deprecate super fast how
do you win when the llm is out of date
in 90 days and you've put so much into
it and now people are talking about
switching I remember it was just earlier
this year when 40 was incredible and we
were so excited for it and now sonnet is
out and that's better in certain ways
and llama's out and that's better in
other ways and it's just going to keep
happening this is not like the age of
railroads where you could build train
train track to a village and it was your
train track and it was a durable
investment and you could actually get
the return on investment by monetizing
it over an extended period of time we
need to be in a place where you can get
that kind of return on investment and
right now what big companies are betting
on is that they are going to get to that
place later and they are willing to
spend now on a insane pace of
acceleration in l M intelligence until
they get to a spot where they can
establish a dominant Market position and
start to monetize and so they're willing
to do all of this throwaway work that
essentially amounts to better cheaper
intelligence for everybody which is
great for consumers great for
professionals who are trying to level up
their
work and they're willing to delay
monetization that's going to have wide
implications on the earnings reports of
major companies over the next few years
they are going to be willing to take
hits on their earnings
that are substantial that are material
that are in the billions and billions of
dollars tens of billions of dollars in
order to
show that they can win at this game
because they think winning at the game
of artificial intelligence is that
important okay so if you want to
understand like why I contextualize
strawberry the way I do that's my take I
think rumors like the strawberry rumor
out of open AI need to be understood
inside like a game theory frame inside
an arms race frame so that you actually
can see what each player is trying to do
and not just look at the capabilities
because so often the capabilities Trail
way behind all right that's my take what
do you think about strawberry