Sora Release Date Deep Dive
Summary
TLDRこのスクリプトは、話題のAI技術であるSoraについて詳しく語っています。OpenAIのCTOであるMir Moraは、Wall Street JournalのインタビューでSoraのリリースについて触れ、新しいビデオのサンプルも公開されました。Soraのトレーニングデータに関しても話がされていますが、詳細は明らかになりませんでした。また、Soraのリリース時期についても話されていますが、技術的な問題や安全面の懸念があるため、具体的なスケジュールはまだわかっていません。一方で、中国のAIスタートアップが急速に発展しており、競合他社が積極的に投資を受けています。また、他にもAI生成のSouth Parkエピソードや、Fable Studiosが開発するThistle Gulchという西部劇のプロジェクトなど、AI技術の進歩とその応用が紹介されています。
Takeaways
- 🚀 Soraは、非常に高い期待と注目を集めており、OpenAIのCTOであるMir Moraは、Soraのリリースの可能性のあるタイムフレームについて語りました。
- 🤔 Mora氏はWall Street Journalのインタビューで、Soraのトレーニングデータに関する問いに対して、詳細には答えず、曖昧な回答をしました。
- 📺 Soraの新しいビデオが公開され、Pixar風のアニメーションの牛が中国の店内を破壊する様子などが含まれています。
- 📹 Soraで生成されたビデオは、720pまたは標準定義で約20秒の長さで、数分で生成されると報告されました。
- 🔍 Mora氏は、Shutterstockからデータを使用してSoraをトレーニングしたと語りましたが、詳細については明言しませんでした。
- 📅 Soraのリリースについては、Mora氏は今年内、もしくは数ヶ月以内にリリースされる可能性があると述べましたが、そのスケジュールには懸念があります。
- 💬 YouTubeのMares BrownleyがSoraのプロジェクトリードと話した際には、リリースのタイムフレームはなく、近い将来はないと答えられました。
- 💰 中国のAIスタートアップに資金が注ぎ込まれており、Soraと競うためにピクシバースやAIスフィアが1400万ドルを獲得しました。
- 🎵 Mora氏はSoraが最終的に音声を持つと述べましたが、チームはまだ検討段階だと答えています。
- 🤖 Soraは、GPTファミリーに似たようなモデルで、扩散Transformerの研究に基づいています。
- 🎨 チームは、アーティストがSoraのようなツールを使って何を作り上げるかに興味があり、プロンプトによる映画制作よりも、新しいユニークな方法で遊ぶことに注目しています。
- 📽 AIによって生成された「South Park」のエピソードが話題となり、その背後にあるのはFable Studiosです。彼らはシミュレーション技術を用いて、新しいプロジェクトを進めています。
Q & A
SoraというAI技術がどのような期待をされていますか?
-Soraは非常に高い期待をされていますが、その期待は現実的でない場合もあるとされています。
OpenAIのCTOであるMir MoraはSoraのリリースについてどのような情報を提供しましたか?
-Mir MoraはWall Street Journalのインタビューで、Soraのリリースについての興味深い詳細を提供しましたが、その中には疑問も残ります。
Soraのトレーニングデータには何が含まれていますか?
-Mir MoraはSoraが公開されているデータでトレーニングされたと述べましたが、YouTubeやFacebookの動画が含まれているかどうかについては詳しくは語らなかったです。
Soraの生成動画の品質と生成にかかる時間について教えてください。
-Soraの生成動画は720pまたは標準定義で、約20秒の長さで、数分で生成されると報告されていますが、一般的には20秒の動画を生成するには30分から1時間半かかるとされています。
Soraのリリースの見通しについてMir Moraはどのように述べましたか?
-Mir Moraは今年内、おそらく数ヶ月以内にSoraがリリースされる可能性があると述べましたが、その見通しには懐疑的な見方も指摘されています。
Soraのプロジェクトリードがリリースの見通しについてどのように述べましたか?
-Soraのプロジェクトリードは、YouTubeのMares Brownleyのポッドキャストで、リリースのタイムフレームはなく、すぐにはリリースされないと述べています。
Soraの音声機能についてどのような情報が提供されていますか?
-Mir MoraはSoraが最終的に音声機能を持つと述べましたが、プロジェクトリードはその実現に向けて検討しているとのことです。
Soraのトレーニングに使用されたデータソースについて教えてください。
-SoraはShutterstockのデータでトレーニングされたとMir Moraが述べ、OpenAIはShutterstockと関係を持っています。
Soraのプロジェクトリードが最も興味を持っていることは何ですか?
-プロジェクトリードは、アーティストがSoraのようなツールを使って何を作り上げるかに最も興味を持っています。
AI生成のSouth Parkエピソードについて教えてください。
-AI生成のSouth Parkエピソードは、The Simulation社(Fable Studios)によって作成され、技術デモとしてリリースされました。
Fable StudiosのThistle Gulchプロジェクトとは何ですか?
-Thistle Gulchは、西部劇のテイストが強いプロジェクトで、小西部町で発生した殺人事件を巡って警部が捜査を進める物語です。
Fable Studiosが提供するSageとは何ですか?
-Sageは、Fable Studiosが提供するオープンソースのAIエージェントで、物語やキャラクターの対話を深くカスタマイズすることができます。
Outlines
🤖 Soraの登場と期待:オープンAIのCTOが話す
SoraというAI技術が注目されており、オープンAIのCTOであるMir Moraは、Wall Street JournalのインタビューでSoraのリリースについて触れています。新しいSoraのビデオも公開されており、その生成速度についても話されていますが、生成にはまだ時間がかかることが予想されます。また、Soraのトレーニングデータについても議論されており、Shutterstockからデータを使用していることが明らかになっています。MoraはSoraのリリース時期について今年内である可能性を示唆していますが、プロジェクトリーダーたちの過去のコメントと照らし合わせると、その見通しは曖昧です。
🚀 AI技術の競争と進化:Soraの音声化とその可能性
中国のAIスタートアップがSoraと競争するために資金を集めており、Soraは最終的に音声機能を持つ予定です。プロジェクトリーダーは、Soraのトレーニングデータについても話しており、GPTファミリーに似ているとされています。Soraはアーティストがどのように使用するかに興味を持ち、新しい創造的な方法でプレイすることを期待しています。また、AI生成のSouth Parkエピソードが話題になり、その背後にある技術はFable Studiosによって開発されています。彼らはシミュレーション技術を使って、新しいプロジェクトを進めています。
🎬 AIとエンターテイメント:Thistle Gulchの物語
Fable Studiosは、AI技術を使って物語を進化させています。彼らのプロジェクトThistle Gulchは、西部劇の風味を持ちながらも、AIエージェントが物語を進めています。シミュレーション技術を使って、キャラクターに背storiesを与え、観客がインタラクティブな選択肢を提供できます。Fableはまた、Sagaをオープンソースとしてリリースし、AIの意思決定プロセスや会話生成をカスタマイズできる強力なPython APIを提供しています。Thistle Gulchのベータ版への登録も開始されており、興味のある人は参加することができます。
Mindmap
Keywords
💡Sora
💡OpenAI
💡AI生成コンテンツ
💡データの訓練
💡映像生成モデル
💡テクニカルデモンストレーション
💡インタラクティブフィクション
💡オープンソース
💡AIエージェント
💡Westworld
💡Hyper
Highlights
Sora, an AI technology, has been hyped with potentially unrealistic expectations.
OpenAI's CTO, Mir Mora, discussed a potential timeframe for Sora's release in an interview with the Wall Street Journal.
New Sora videos showcased, including a Pixar-styled animated bull and a robot interaction with a female reporter.
Sora's training data was a point of contention, with Mora stating it was trained on publicly available or licensed data.
Moradi mentioned that Sora was trained on data from Shutterstock, which OpenAI has a relationship with.
Sora's release timeframe is uncertain, with Mora suggesting 'this year, possibly within a few months', contrasting with tech leads' 'no timeframe, not anytime soon'.
An arms race in AI funding is occurring, with Chinese AI startups receiving significant investments to compete with Sora.
Sora's potential capabilities include sound, as mentioned by Mora, and is a focus of the development team.
The development team referred to Sora using a version number, hinting at possible future iterations.
Sora's training methodology resembles the GPT family, based on research that began with a NYU computer science professor.
The team is interested in how artists will use Sora, focusing on new and unique applications rather than pre-defined prompts.
The speaker encourages creators to start making content now instead of waiting for Sora.
AI-generated South Park episodes were created by The Simulation, a company also known as Fable Studios.
The Simulation is working on a virtual city project called Sim Francisco and a Western-themed project called Thistle Gulch.
Thistle Gulch features interactive AI characters with backstories, driven by a generative AI system.
Fable has released Saga, the AI system used in Thistle Gulch, as open source.
The speaker signed up for the beta waitlist of Thistle Gulch and encourages viewers to do the same.
Figure One recently showcased how their robot can be integrated with chat GPT, hinting at future advancements in AI.
Transcripts
hey everyone so Sora has obviously been
hyped pretty hard I'd say to some pretty
unrealistic expectations the latest is
an interview with open ai's CTO Mir Mora
in which she discusses a potential time
frame for sora's release I have some
doubts about that we're going to talk
about that plus I have some other
interesting sore details that I have not
seen covered anywhere else also remember
those AI generated South Park episodes
from a few months back well that
technology is starting to roll out and
I'm going to show you how you can get
access to it all right let's dive in
kicking off with Sora in a recent
interview with the Wall Street Journal
open ai's CTO Mir moradi fielded some
questions about Sora and provided some
pretty interesting details but things
get really interesting when you contrast
Mora's Wall Street Journal answers with
an interview from the Sora tech leads
conducted just 5 days beforehand we'll
give it the smell test in just a minute
but first let's break down morat's Wall
Street Journal interview which does get
a little spicy uh which makes sense
considering that you know open AI is
currently being sued by the New York
Times over copyright infringement first
and probably most exciting is the fact
that we got to see some new Sora videos
we got this sort of Pixar styled
animated bull in a china shop um yeah it
looks pretty good it was pointed out
that the bull should probably be causing
more destruction in said china shop we
got the prompt of a female video
producer on a sidewalk in New York City
holding a high-end Cinema Camera
suddenly a robot Yanks the camera out of
her hand and this was the result in
which the female reporter kind of does
this weird dance move and then morphs
into the robot and holds the world's
weirdest looking cinema camera that I've
ever seen I do want to shoot on that
thing though it's such a weird camera
it's got like two lenses on the front
and one on the side we also had some
pretty convincing footage of two women
giving an interview there was some
issues of course with uh the one woman's
fingers but yeah again that's to be
expected other than that it actually
looks pretty good and we also got to see
bootleg Ariel with a a mermaid reviews a
smartphone prompt interestingly the
reporter states that these videos were
720p or standard definition and were
about 20 seconds in length morate
reported that it took a few minutes to
generate which does seem fast
considering most reports indicate that
it takes anywhere between 30 minutes to
an hour and a half to generate uh 20
second videos in Sora I can only presume
that the now famous 1081 minute examples
on the Sora website probably took a lot
longer than that now it's possible that
open AI have optimized since those
report s came in uh we're going to be
hearing a little bit more about that
from the tech team in just a little bit
now the real kind of clipped and
bookmarked moment from this interview is
when Mora is asked about the training
data used for Sora she states that Sora
was trained on publicly available or
license data but pressed on if any of
that data could possibly include say
YouTube videos or Facebook videos morate
says she doesn't know and she can't be
sure but um I'm I'm not sure I'm not
confident about about it she eventually
just completely stonewalls the question
saying I'm I'm just not going to go into
the details of of the data that was that
was used she did off camera and after
the interview and likely after
Consulting with like the army of lawyers
that are probably just standing right
off screen that Sora was trained on data
from Shutterstock which open AI does
have a relationship with this one felt a
bit weird to withhold on considering
it's been well public knowledge since at
least July 11th of 2023 in terms of
sora's actual release Mora does state
that you know given the amount of power
in compute uh the amount of you know
expenses in terms of that compute that
they don't know what Sora will look like
when it is eventually released
personally and I am speculating here I
take that to mean that we will not be
generating 1080p one minute long videos
with Sora but probably something more in
the 4 to 10 second range very much akin
to the video generators that we
currently have access to but the hot
ticket question on everyone's mind is
when does sore release to which maradi
replied this year possibly within a few
months I have some doubts at least on
the in a few months time frame mostly
because just 5 days before this
YouTube's own mares brownley had an
interview with open AI Bill peoples Tim
Brooks and aled ey raash the project
leads of Sora on his podcast they were
asked the same question when does Sor
released and their answer was we have no
time frame not anytime soon now look
anyone that has worked anywhere knows
that sometimes in a meeting the boss is
just going to say we're doing a thing
and all of you guys who actually do the
thing are like we are what where did
that come from we have all been there
now is that what is actually happening
here well I don't know but it does seem
a little suspect that in the span of 5
days we've gone from we have no time
frame to this is happening in a few
months especially when you factor in all
of the hurdles both from a technical
standpoint and frankly from a safety
guard rail standpoint that need to be
overcome before release that said there
is definitely an arms race going on
right now with funding pouring into
Chinese AI startups to compete with Sora
just this week both pixiverse and AI
sphere have both received $14 million
that's like $14 million each to catch up
with open AI marate also stated in the
Wall Street Journal interview that Sora
will eventually have sound whereas the
team actually stated it was something
that they were thinking about that's not
a gotcha or anything like that but there
was an interesting little tidbit in Bill
people's answer about sound wherein he
said it's hard to give exact timelines
with these kinds of things uh for sort
one we were really focused on pushing
the capabilities of video generation
models is it telling that he referred to
it as Sora 1 it's unsurprising that he
refers to it by a version number you
know considering we're sitting here
waiting for Chad GPD 5 but I did find it
fascinating that he referred to it in
the past tense which does beg the
question are they already working on
Sora 2 and what does that look like in
Marquees interview which is linked down
below the team is also asked about what
data Sor was trained on and similarly
they kind of give you know a Stonewall
answer there but they do state that Sora
was trained quote like Del but it more
resembles the GPT family that likely
backs up the speculation and reverse
engineering research papers that have
been released on Sora that state that
it's actually built on a diffusion
Transformer research on that began with
signing she a NYU computer science
Professor back in July of 2022
interestingly his Mente at that time
Bill peoples ultimately through both
interviews both Mora and the team state
that they're most interested in seeing
what artists eventually end up doing
with a tool like Sora they don't seem to
be very interested in like prompt a
movie but rather seeing what happens
when you and I start playing with it in
new and unique ways that even they
couldn't foresee as I've been saying
over the last few videos if you want to
make something don't wait for Sora just
start making it now I mean for one I
just don't think that Sora is going to
do what you think it's going to do at
least at launch and look given what
we've seen of sora's output including my
personal favorite the shark at the beach
video I mean look at this thing it's
amazing it is amazing it's still going
to be weird morphe and and it's not
going to be doing consistent characters
or locations so honestly use the tools
that we have now there is the totally
free Hyper that I covered just a few
videos ago the way I look at it is that
like it's 1958 and you have a Super 8 mm
film camera but you don't want to make
your film until you have like a Sony fx3
in your hands it just doesn't make sense
moving on remember those AI generated
South Park episodes that made the rounds
a few months back well they were created
by a company called the simulation uh
which is also Fable Studios it gets a
little confusing it's a whole rabbit
hole but I have been following them with
great interest since the South Park
episodes were released as more or less a
tech demo to you know get some interest
in Fable Studios slthe simulation uh
which I think it definitely did uh since
then they've been working on a number of
other projects including Sim Francisco
which I covered in another video but
it's basically a virtual City filled
with virtual AI citizens who go about
their daily lives you know going to work
going to sleep falling in love and even
dying and for quite some time they have
been working on thistle Gulch which is
more or less a straight ahead Western
but you know given all this technology
it definitely has a lot of strong
Westworld Vibes and while this might not
look terribly cinematic at this point
it's what's happening underneath that is
really interesting narratively there has
been a murder that has taken place in
the small western town of thisle Gulch
and we are following a sheriff who is
investigating our sheriff is an AI agent
who not only has obviously his goal of
solving the murder but an entire
backstory as well and so does every
other character in this town all driven
by Saga or skill to action generative
agent in thisle Gulch you can be
presented with interactive choices for
your characters to make but you know
they can also just make choices for
themselves it's a really fascinating
idea blending emergent storytelling with
interactive fiction and although you
know the overall look of it might not
necessarily look all that cinematic
right now I actually think it has kind
of a cool look to it it sort of looks
like Borderlands B says their platform
is designed for creators researchers and
AI enthusiasts offering unprecedented
control over the narrative and character
interactions through a powerful python
API this allows for deep customization
of AI decision-making processes and
conversation generation additionally
Fable has released Saga as open source
so if you want to play with that code
you can go ahead and do so right now as
for thistle Gulch they have opened up a
beta weit list that link is down below
I've signed up for it you should
definitely sign up for it too and you
know I'm not covering it here today but
figure one just recently released a
video showcasing how their robot can be
hooked up into chat GPT so maybe we
really aren't far off from a real life
Westworld hopefully we have learned the
lesson not to shoot the robots because
you know eventually they will shoot back
on that note I thank you for watching my
name is
Tim
Ver Más Videos Relacionados
GPT-4o|ChatGPT最新モデルは何が凄いのか?
Microsoft's new "Embodied AI" SHOCKS the Entire Industry! | Microsoft's Robots, Gaussian Splat & EMO
WWDC Drama I'm Watching For at Apple's AI Event
Day 29/75 Build Text-to-Video AI with LLM [Explained] OpenAI SORA Stable Diffusion VideoPoet Runway
これさえ押さえておけばOK! 生成AI時流を解説 〜3月後半の生成AIトレンドをご紹介〜(2024/04/03)
松田語録:Soraへの反発〜今どきのラッダイト運動
5.0 / 5 (0 votes)