Another glorious battle for AI dominance… GPT-4o vs Google I/O
Summary
TLDR在2024年的Google I/O大会上,Google和OpenAI展开了激烈的技术竞争。OpenAI抢先发布了其旗舰模型GPT-4 Omni,该模型结合了文本、视觉和音频,并具有类似人类的对话能力。尽管对话功能尚未对公众开放,但OpenAI已在与苹果商谈将其技术整合到iPhone中。与此同时,Google也展示了其Project Astro,并宣布了多项新功能,包括能够处理200万标记上下文窗口的Gemini 1.5 Pro,以及新的Firebase工具,如Firebase Gen Kit和Firebase Data Connect,后者将Postgres正式引入Firebase,满足了多年来社区的头号需求。此外,Google还宣布了新的硬件产品,如Trillium TPU和Axion CPU,以及与OpenAI Sora竞争的生成视频模型Vo。尽管技术进步令人印象深刻,但视频中表达了对达到技术奇点(singularity)进展缓慢的失望,暗示AI的发展可能正处在一个停滞期。
Takeaways
- 📅 Google IO是一年一度的开发者大会,Google在这里尝试追赶其人工智能领域的竞争对手OpenAI。
- 🆕 OpenAI在Google IO前夕发布了其新旗舰模型GPT-4,这可能并非巧合,而是有意为之。
- 🚀 GPT-4模型结合了文本、视觉和音频能力,并且具有类似人类的对话能力。
- 🔍 GPT-4的对话功能尚未对公众开放,但OpenAI正在与苹果商谈将其技术整合到iPhone中。
- 🤖 Google在IO上展示了名为Project Astro的项目,与OpenAI的Omni相似,但语音听起来更机械化。
- 📱 两家公司都在竞争,希望创建一个既智能又快速且成本低廉的模型,以便能在移动设备上运行。
- 🌟 Google宣布了Gemini 1.5 Pro,能够处理高达200万个token的上下文窗口,这相当于2小时的视频内容或60,000行代码。
- 💰 Google推出了一个名为context caching的新功能,可以以更低的成本重用tokens。
- 🛠️ Google发布了Firebase Gen Kit,这是一个与Firebase集成的工具,可以轻松构建AI驱动的API端点。
- 🔗 Firebase data connect是Google发布的一项新工具,它正式将Postgress引入Firebase,这是多年来最受期待的功能。
- 📈 Google还宣布了一些新的硬件,如Trillium TPU和Axion,这是其为数据中心设计的新ARM架构CPU。
- 🎥 Google宣布了VO,一个生成视频模型,与OpenAI的Sora竞争,展现了一年来技术的巨大进步。
- 🤔 尽管有这些技术进步,但对于向奇点(singularity)的进展感到有些失望,AI的智能和独立学习能力似乎并未有显著提升。
Q & A
Google I/O 是一个怎样的会议?
-Google I/O 是 Google 一年一度的开发者大会,Google 在此会议上发布新技术和产品,旨在与竞争对手如 Open AI 保持竞争力。
Open AI 宣布了什么重大更新?
-Open AI 宣布了其新旗舰模型 GP4 Omni,这是一个比 GP4 Turbo 更快、更便宜的模型,能够将文本、视觉和音频结合在一个单一模型中。
GP4 Omni 的对话能力有何特点?
-GP4 Omni 的对话能力非常接近人类,能够根据情境变化语调,从戏剧性到讽刺再到超级冷静的语调,适合讲睡前故事。
Open AI 与 iPhone 有什么合作计划?
-Open AI 正在与苹果公司讨论将其技术整合到 iPhone 中,同时 Google 也在寻求将其旗舰模型整合到 iPhone 中。
Google 在 I/O 大会上展示了什么项目?
-Google 在 I/O 大会上展示了名为 Project Astro 的项目,该项目与 Open AI 的 Omni 相似,但存在更多的延迟,语音听起来也更机械化。
Gemini 1.5 Pro 是什么?
-Gemini 1.5 Pro 是 Google 发布的一个新模型,能够处理高达 200 万个 token 的上下文窗口,相当于 2 小时的视频内容或 60,000 行代码。
什么是 Firebase gen kit?
-Firebase gen kit 是 Google 发布的一个新工具,它与 Firebase 集成,使得构建 AI 支持的 API 端点变得更加容易。
Firebase data connect 是什么?
-Firebase data connect 是 Google 发布的一个新工具,它将 PostgreSQL 正式引入 Firebase,这是多年来最受期待的功能之一。
Google 还宣布了哪些新技术或产品?
-Google 还宣布了新的硬件产品,如 Trillium TPU 和 Axion,这是其为数据中心设计的新型基于 ARM 的 CPU。此外,还宣布了名为 Vo 的生成视频模型,以与 Open AI 的 Sora 竞争。
目前 AI 技术发展的主要瓶颈是什么?
-目前 AI 技术发展的主要瓶颈在于智能的提升。尽管模型变得更快、更便宜,但如果它们没有变得更加智能,那么奇点(singularity)的到来将遥不可及。
为什么说我们可能正站在一个平台期的边缘?
-因为尽管 AI 模型在基准测试中的表现已经相当出色,但如果它们没有实现独立学习的能力,AI 的智能水平可能已经达到了一个暂时的极限,这就是所谓的平台期。
为什么说我们可能正走向失望之谷?
-因为尽管 AI 技术取得了显著进步,但如果它们没有实现真正的智能突破,人们对 AI 的期望可能会逐渐降低,从而导致对当前技术进展的失望。
Outlines
📅 Google IO 大会与 AI 竞争
在 Google IO 年度开发者大会上,Google 宣布了多项新技术,包括 Firebase 的 SQL 数据库。然而,最引人注目的新闻是 Open AI 发布了其新的 GPT-4 模型,这发生在 Google IO 仅几小时前,引发了对两家公司间竞争的关注。视频将深入探讨这场 AI 竞争,并审视最近 48 小时内发布的各种新技术。
🚀 Open AI 的 GPT-4 模型
Open AI 展示了其新的旗舰模型 GP4 Omni,该模型比 GP4 Turbo 更快、更便宜,并且能够将文本、视觉和音频结合在一个单一模型中。GP4 Omni 最令人印象深刻的是其类似人类的对话能力,尽管目前对话部分尚未对公众开放。此外,Open AI 正在与苹果公司讨论将其技术整合到 iPhone 中,而 Google 也在寻求将其旗舰模型整合到 iPhone 中。
🌟 Google 的 Project Astro 和 Gemini 1.5 Pro
Google 在 IO 上展示了 Project Astro,它与 Open AI 的 Omni 相似,但存在更多延迟,声音也更机械化。Google 还宣布了 Gemini 1.5 Pro,它能够处理高达 200 万个令牌的上下文窗口,这相当于 2 小时的视频内容或 60,000 行代码。为了解决令牌成本问题,Google 发布了名为上下文缓存的新功能,可以以更低的成本重用令牌。此外,Google 还推出了面向开发者的竞赛,胜者将获得一辆电动 DeLorean 汽车,并发布了 Firebase gen kit,这是一个与 Firebase 集成的新工具,可以轻松构建 AI 启用的 API 端点。
🔥 Firebase 的新功能和硬件更新
Google 宣布了 Firebase data connect,这是一个将 PostgreSQL 正式引入 Firebase 的工具,这是多年来最受期待的功能。此外,Google 还宣布了新的硬件,如 Trillium TPU 和 Axion,这是其为数据中心设计的新 ARM 架构 CPU。最后,Google 还宣布了 VO,这是一个与 Open AI 的 Sora 竞争的生成视频模型,与一年前相比,技术进步令人印象深刻。
🤔 对奇点的思考和未来的展望
尽管 AI 模型变得更快、更便宜,但如果它们没有变得更加智能,那么达到奇点的可能性似乎还很遥远。自从 GPT-4 发布以来已经过去了一年多,目前看来我们可能正站在一个平台期的边缘,除非有重大突破使 AI 真正智能并能够独立学习,否则我们可能正面临着幻灭的低谷。视频以对当前 AI 进展的反思和对未来的展望结束。
Mindmap
Keywords
💡Google IO
💡Open AI
💡GPT-4
💡Project Astro
💡Gemini 1.5 Pro
💡Context Caching
💡Firebase
💡Superbase
💡Trillium TPUs
💡Axion
💡VO
Highlights
Google IO是一年一度的开发者大会,Google在会上宣布了一些令人难以置信的新技术。
OpenAI在Google IO前几小时发布了新的GPT-4模型,展示了其在对话能力上的进步。
GPT-4模型结合了文本、视觉和音频,能够进行类似人类的对话。
OpenAI正在与苹果讨论将技术整合到iPhone中,Google也有此意向。
Google在IO上展示了名为Project Astro的项目,与Omni相似但存在延迟和声音机械化的问题。
OpenAI与其前首席科学家和联合创始人Ilia分道扬镳,可能存在一些内部矛盾。
Google宣布了Gemini 1.5 Pro,可以处理高达200万个token的上下文窗口。
Google推出了Context Caching功能,可以重复使用tokens以降低成本。
Google为开发者启动了一项竞赛,胜者将获得一辆电动DeLorean。
Firebase Gen Kit工具发布,与Firebase集成,便于构建AI支持的API端点。
Project idx现在对公众开放,这是一个基于浏览器的VS Code。
Firebase Data Connect工具发布,将Postgress正式引入Firebase。
Superbase作为Firebase的替代品,现在Firebase成为了Superbase的替代品。
Google还宣布了一些新的硬件,如Trillium TPU和Axion,这是其新的基于ARM的CPU。
Google宣布了VO,一个生成视频模型,与OpenAI的Sora竞争。
尽管技术进步巨大,但作者对向奇点(singularity)的进展感到失望。
AI模型在基准测试上已经达到极限,除非有重大突破,否则奇点似乎还很遥远。
The Code Report节目感谢观众观看,并预告了下一期节目。
Transcripts
yesterday was Google IO the annual
developer conference where Google
desperately tries to catch up to its
artificial rival open AI
Google Google announced some crazy stuff
I never thought I would see in my
lifetime like a SQL database for
Firebase more on that later because
first we need to talk about the biggest
announcement at iio open ai's new GPT 4
oh oh oh you see open AI hype Lord Sam
Alman yet again wrapped up Sundar in a
wet blanket by releasing GPT 40 just
hours before Google IO which is a total
coincidence and definitely not designed
to troll Google in today's video we'll
break down this artificial beef but more
importantly look at all kinds of crazy
new technology released in just the last
48 hours it is May 15th 2024 and you
watching the code report on Monday open
aai had a surprise spring update where
they unveiled their new flagship model
gp4 Omni you've got me on the edge of my
well I don't really have a seat but you
get the idea what's the big news yeah
we've got a new model which is faster
and cheaper than gp4 turbo and combines
text vision and audio into a single
model what was most impressive though
was its humanlike conversational
abilities well well well just when I
thought things couldn't get any more
interesting talking to another AI that
can see the World by default it uses a
California Valley Girl accent set to
maximum cringe but the tone of the voice
can vary from dramatic to sarcastic to
Super chill for bedtime stories a
bedtime story about robots and love I
got you covered this technology will be
a huge leap forward for your AI
girlfriend and you can use the GPT 40
model today but the conversational part
of it is still not available to the
public that's disappointing but what you
also need to know is that open AI is in
talks to put their technology on the
iPhone but Google also wants to get its
Flagship model on the iPhone as well it
talks are on going to also get Gemini on
the iPhone so these companies are
competing to create a model that's smart
but also fast and cheap enough to run on
mobile in order to get that massive bag
from Apple yesterday at IO Google demoed
something called project Astro which
feels similar to for Omni do you
remember where you saw my
glasses yes I do your glasses were on
the desk near a red apple it's cool but
there's more latency and the voice is
more robotic compared to open AI now
what's also very interesting is that
open AI just parted ways with Ilia their
former Chief scientist and co-founder
who many people used to worship as the
brains behind open AI there's definitely
some underline drama here but we likely
won't know the truth until they release
their Memoirs in the 2040s but now let's
finally talk about Google IO the biggest
AI announcement from Google was Gemini
1.5 Pro which can now handle a 2 million
token context window that could be 2
hours of video content or 60,000 lines
of code that's a lot of context but
tokens can be expensive and to address
that they released a new feature called
context caching that can reuse tokens
for a fraction of the cost in addition
Google launched a competition for
developers and whoever builds the best
Gemini powered app wins an electric
DeLorean to make building this app
easier they also released a new tool
called Firebase gen kit which which is
integrated with oama and makes it easy
to build AI enabled API endpoints in
addition project idx is now open to the
public which is a browser-based vs code
that's also integrated with things like
mobile emulators by far the most
exciting thing for me though is a new
tool called Firebase data connect which
officially brings postgress into
Firebase this has been the number one
most requested feature for years how do
I use Firebase with SQL and its absence
is led to startups like superbase which
is branded as a Firebase alternative but
now in 2024 the turns of table Firebase
is now the superbase alternative I'm a
big fan of both super base and Firebase
and if you want to learn these
Technologies check out my full courses
on fireship iio and stay tuned for a
full tutorial on data connect on my
second Channel Beyond fireship soon
Google also announced some new hardware
like Trillium tpus and Axion its new
arm-based CPUs for data centers and
finally Google also announced vo a
generative video model to compete with
open AI Sora it's extremely impressive
compared to where we were just a year
ago but yet again it just feels one step
on behind open AI we just looked at all
kinds of crazy new gamechanging
technology but at this point I'm feeling
a little disappointed with our progress
towards the singularity it's been over a
year since GPT 4 and unfortunately I
still have a job four Omni Claude and
Gemini 1.5 all seem to be pretty maxed
out on how far they can get with these
benchmarks making models faster and
cheaper is great but if they're not
becoming more intelligent then the
singularity is nowhere in sight they've
already absorbed almost all the
information humans have created is so
unless there's a major breakthrough that
makes AI actually intelligent and able
to learn independently it sure looks
like we're standing on the edge of a
plateau and the only place to go is the
trough of disillusionment this has been
the code report thanks for watching and
I will see you in the next one
5.0 / 5 (0 votes)