AI PULSE - AI by Mark Zuckerberg

Microsoft Unveils Enhanced Phi-4 Models for Improved Efficiency and Reasoning in AI Tasks Anthropic lets users connect more apps to Claude Google's AI Mode Revolutionizes Search with Real-Time Data and Personalized Responses Apple Intelligence To Add Google Gemini & Other AI Soon Apple and Anthropic Collaborate on AI-Powered "Vibe-Coding" Platform for Developers Meta's AI Strategy: Revolutionizing Advertising, Social Experiences, and Emerging Technologies #AI, #MarkZuckerberg, #Microsoft, #Google, #Apple, #Anthropic, #Meta

Episode Transcript

(00:00):
Welcome to Innovation Pulse, your quick no-nonsense update on the latest in AI.

(00:09):
First, we will cover the latest news.
Microsoft launches the five-four reasoning models to boost productivity.
Anthropic enhances Claude with new features, and Google's AI mode transforms search capabilities.
After this, we'll dive deep into how Apple and Anthropic are collaborating on a groundbreaking

(00:29):
Vibe Coding software platform using generative AI.
Microsoft has introduced a new era for small language models with five-four reasoning,
including five-four reasoning plus and five-four mini-reasoning.
These models excel in complex reasoning tasks, such as mathematical problem solving,

(00:50):
through advanced techniques like distillation and reinforcement learning.
Despite their smaller size, they rival larger models in performance,
making them suitable for various applications, including educational tools and mobile systems.
Available on Azure AI Foundry and Huggingface, these models are optimized for efficient use

(01:12):
on devices with limited resources. Microsoft emphasizes responsible AI development,
ensuring the models adhere to principles of accountability, transparency, and safety.
The five models integrate seamlessly into the PC ecosystem, enhancing productivity applications
like Outlook. They are designed to operate efficiently on upcoming co-pilot plus PC

(01:38):
NPUs, offering fast and reliable performance.
Anthropic has introduced new features for its AI chatbot, Claude, including app connections called
integrations and an expanded research tool, advanced research. These are initially available in beta

(02:00):
for certain subscription plans. Integrations allow developers to enhance Claude's capabilities by
connecting it to various app servers, drawing data from business tools, content repositories, and more.
For example, Claude can now summarize and create pages in Atlassian's confluence,
or connect to Zapier's workflows. Advanced research enables Claude to deliver comprehensive

(02:26):
reports by searching multiple sources with capabilities to cite original materials.
This enhancement aims to compete with similar features in chatbots like
Google's Gemini and OpenAI's ChatGPT. Anthropic is working towards significant revenue growth,
with a target of $34,500 million by 2027, though current annualized revenue is about $1,400 million.

(03:01):
Join us as we discuss the enhanced AI search capabilities. Google's new AI mode enhances
search capabilities, particularly for planning trips, shopping, and learning. It merges real-time
data with generative AI for longer queries and personalized responses, making it ideal for
detailed questions. The update introduces visual place and product cards, offering current information

(03:27):
like ratings and local inventory. Users can access their past searches via a new panel,
aiding in ongoing research. Initially accessible through Google Labs, AI mode now offers immediate
access for US users. This feature is seen as an extension of AI overviews, allowing users to request

(03:48):
AI-generated responses on demand. While AI overviews provide summaries, AI mode serves as a research
tool. Unlike Gemini, which acts as a productivity assistant, AI mode focuses on easy information
access. User feedback is being used to refine the model.

(04:12):
Google has confirmed that Apple will integrate its Gemini AI service into Apple Intelligence by
the end of 2020. Reports indicate Apple plans to add other AI services as well.
This was hinted at by Apple Senior Vice President Craig Federighi, who suggested users might choose
models like Google Gemini in the future. This update came from a Department of Justice inquiry,

(04:37):
where Sundar Pichai confirmed discussions with Apple CEO Tim Cook about evolving AI technologies.
Cook expressed interest in Google's plans, discussing Gemini's distribution and the addition
of more third-party AI models on Apple Intelligence. These developments align with predictions that

(04:58):
Apple Intelligence would select third-party AI to provide optimal answers to user queries,
indicating a strategic enhancement of Apple's AI capabilities in the near future.
Apple and Anthropic are collaborating on a Vibe Coding software platform using
Generative AI to write, edit and test code, as reported by Bloomberg. This new system is an

(05:23):
updated version of Apple's Xcode, incorporating Anthropic's Claude Sonet model. Currently,
Apple plans to use the software internally, with no decision on a public launch yet.
To enhance its AI capabilities, Apple is engaging with various partners. Open AI's chat

(05:44):
GPT is already integrated into Apple's Intelligence features, and Google's Gemini might be included
in the future. Anthropic's latest Claude models are gaining popularity among developers,
particularly for coding tasks on platforms like Cursor and Windsurf. This collaboration

(06:04):
reflects Apple's strategic move to expand its AI infrastructure by leveraging external expertise.
And now, pivot our discussion towards the main AI topic.
Welcome to Innovation Pulse. I'm Jakob Lasker. Today, we're diving into a fascinating

(06:27):
interview with Mark Zuckerberg from Strathechery, where he outlines Metta's ambitious AI strategy
in a way that I found both revealing and significant for the tech landscape.
The interview took place at Metta HQ just before LamaCon, their developer conference focused on
their open-source AI model. What struck me immediately was how clearly Zuckerberg has

(06:51):
positioned AI as central to Metta's future, not as a side project, but as something that touches
every aspect of their business. Let me break down what I see as the most important revelations.
First, Zuckerberg outlined four major AI opportunities that Metta is pursuing. The
clarity of this vision shows they're not just throwing things at the wall to see what sticks,

(07:16):
they have a comprehensive strategy. The first opportunity is revolutionizing their advertising
business. Zuckerberg describes what he calls the ultimate business agent, where a business can
simply specify desired outcomes, connect their bank account, and Metta's AI handles everything else.

(07:37):
No need for the business to create content, understand targeting, or even know who their
customers are. The AI will generate creative content, find the right audience, and deliver
measurable results. This would redefine advertising entirely. Zuckerberg believes
advertising as a percentage of GDP could grow substantially as this removes traditional constraints.

(08:02):
The second opportunity is enhancing engagement across their platforms through AI-generated
content. Zuckerberg described social media's evolution through three distinct epics. First,
content from your friends. Second, creator content that's algorithmically recommended.
And now, a third epic of AI-generated content. What's fascinating here is his prediction that,

(08:27):
as AI increases productivity, people will spend less time working and more time on entertainment
and culture, making these AI-enhanced feed services even more central to daily life.
The third opportunity is business messaging, particularly on WhatsApp,
which has nearly three billion users but generates far less revenue than Facebook or Instagram.

(08:50):
Zuckerberg pointed to Thailand and Vietnam, where business messaging is already substantial.
These countries rank in Metta's top 10 for revenue, despite being around 30th in GDP rankings.
He estimates 2% of Thailand's GDP flows through transactions on Metta's messaging apps.
The key to unlocking this globally is AI replacing human labor for customer service and sales.

(09:17):
The fourth opportunity is AI-native experiences, like Metta AI, which Zuckerberg says already has
a billion monthly users across Metta's apps. They've also launched a standalone Metta AI app
for those who want a dedicated experience. Now let's talk about Lama and Metta's open-source
AI approach. This is particularly intriguing to me because it diverges from the closed models of

(09:41):
competitors like open AI and Anthropic. Zuckerberg announced a Lama API at the conference but
emphasized it's not meant to be a major business. Instead, it's positioned as a reference implementation,
offering the vanilla version of their model at close to their cost of capital.
Why open-source? Zuckerberg explained that open-source AI allows developers more control and

(10:08):
customization than closed APIs, where providers can change things overnight or censor certain
queries. He predicts 2025 will be the year open-source becomes the largest type of model
that developers work with. But here's the strategy behind it. Metta is building Lama primarily for

(10:28):
their own use cases. The training costs are enormous, but Zuckerberg is planning financially
as if Metta is bearing the full cost alone. Any industry contribution would be upside.
While he's approached other companies about sharing training costs,
most have chosen to develop their own models instead. What's particularly telling is how

(10:53):
Zuckerberg described the economics. He said the marginal return has always been higher to
allocate GPUs to their recommendation systems and ads than to an API business. There's a constant
internal calculation about where to direct computing resources for maximum return. This
offers rare insight into how these decisions are made at Metta's scale. The Metta AI vision is

(11:18):
also fascinating. Zuckerberg believes Metta has a unique advantage in building personal AI
because of their long experience with social technology. He said something I found quite
insightful. That a good personal AI isn't just about knowing your basic preferences,
but about developing a theory of mind about you, understanding your world in a deep way.

(11:42):
Zuckerberg envisions Metta AI as a personal companion that can help people process issues,
prepare for difficult conversations, or simply be there to talk. He cited research suggesting the
average American has fewer than three friends but would like more, positioning AI as potentially
filling some of this gap. He also suggested everyone should have a therapist and for those who don't,

(12:09):
everyone will have an AI. This takes us full circle from Facebook's original mission of
connecting people. Metta started with human connections, pivoted to entertainment under
competitive pressure and now seems to be returning to connection, but with AI as the connection point.
When pressed on this, Zuckerberg acknowledged Metta's particular attunement to people's desire

(12:34):
for self-expression, understanding, and avoiding loneliness. Another significant revelation was
how Metta AI integrates with their vision for AR and VR. Zuckerberg sees glasses as the ultimate
form factor for AI assistance because they allow the AI to see what you see and hear what you hear

(12:55):
while being able to whisper in your ear throughout the day. He firmly believes that within 10 years,
every pair of regular glasses will become AI glasses at minimum and AI glasses with holograms
at maximum. What surprised me was his clear distinction between AR and VR futures. He described

(13:15):
glasses as the phones of the future and VR as the TV of the future. He expects VR to replace
tablets and many TVs as the quality improves while acknowledging the physical limitations of AR,
which can only add photons to the world, versus VR, which gives you a blank canvas.

(13:36):
The Metta name itself seems perfectly positioned for this future. Zuckerberg noted that when they
renamed the company, he expected the fully holographic world to arrive before advanced AI.
Instead, AI is arriving sooner, but both remain parts of the same vision of blending digital and

(13:57):
physical worlds. Perhaps what impressed me most was Zuckerberg's long-term thinking.
He credited Metta's controlled company structure for allowing them to see past short-term pain to
make necessary investments. When downturns hit, they've typically gained market share despite
revenue impacts. As he put it, he doesn't need to worry about making the quarter to keep his job.

(14:23):
This gives Metta a strategic advantage in making decade-long investments in AR,
VR, and now AI. I think there are several key takeaways here.
First, Metta's four-pronged AI strategy touches every aspect of their business, from ads to social
media to messaging to new experiences. Second, their commitment to open-source AI with Lama

(14:47):
sets them apart in an industry increasingly dominated by closed models. Third, their vision
of personal AI assistance leverages their deep understanding of human connection needs.
And finally, their long-term investment horizon allows them to pursue these massive technological
shifts patiently. What's clear is that Metta is positioning itself not just as an AI competitor,

(15:14):
but as a company fundamentally transforming itself around AI while maintaining its social DNA.
As Zuckerberg said, they don't need all four of their AI strategies to succeed,
just two or three to hit out of the park to make their massive AI investment worthwhile.

(15:34):
This is Jakob Lasker for Innovation Pulse. Thanks for listening, and we'll see you in the next
time as we continue tracking the bleeding edge of tech innovation.
And that's a wrap for today's discussion. Microsoft, Anthropik, and Apple are pushing

(15:55):
the boundaries of AI with innovative models and collaborations, while Google and Metta focus on
enhancing user experiences and open-source strategies. Don't forget to like, subscribe,
and share this episode with your friends and colleagues, so they can also stay updated on
the latest news and gain powerful insights. Stay tuned for more updates.