Amazon Brings AI to Shopping, Qwen’s New Multimodal AI, & OpenAI Adopts Anthropic’s Standard

Summary

AI Deep Dive Podcast Summary
Episode: Amazon Brings AI to Shopping, Qwen’s New Multimodal AI, & OpenAI Adopts Anthropic’s Standard
Host: Daily Deep Dives
Release Date: March 27, 2025

---

Introduction

In this episode of the AI Deep Dive Podcast, hosts Alex and Bailey explore the latest advancements in artificial intelligence, focusing on three major topics: Amazon's new AI-driven shopping experience, Alibaba’s cutting-edge multimodal AI model Qwen 2.5 Omni, and OpenAI’s adoption of Anthropic’s Model Context Protocol (MCP). Additionally, they delve into the rising threat of deepfakes and the innovative solutions being developed to combat them.

---

Amazon’s Personalized AI Shopping Experience

Timestamp 01:32 – 03:52

Alex and Bailey kick off the discussion with Amazon's latest AI initiative aimed at revolutionizing the online shopping experience. Amazon has introduced a feature called Interests, which leverages large language models to provide highly personalized shopping recommendations.

• Personalized Search: Instead of generic search terms like "coffee maker," users can input more nuanced queries such as "brewing tools and gadgets for coffee lovers," allowing the AI to understand and cater to specific hobbies and budgets. As Bailey explains, "The real breakthrough here is that they're using large language models to understand those natural language queries" (02:09).

• Automated Notifications: The Interests feature doesn't just enhance search capabilities; it also operates in the background, sending notifications about new products, restocks, and special offers aligned with the user's interests (02:42).

• Integration with Existing AI Tools: This feature builds on Amazon's existing AI tools like Rufus, their AI shopping assistant, and AI-generated customer review summaries, marking the next step in making online shopping feel like an ongoing conversation rather than a mere transaction (03:28).

• Broader Industry Trends: The discussion highlights that Amazon is not alone in this venture. Other companies, such as Google with its Vision Match tool, are also integrating similar AI-driven personalization features into their shopping platforms (04:03).

Notable Quote:
Bailey emphasizes the importance of this feature by stating, "The Interests feature really feels like the next step in this strategy of AI integration" (03:46).

---

Alibaba’s Qwen 2.5 Omni: A New Multimodal AI

Timestamp 04:57 – 09:12

The conversation shifts to Qwen 2.5 Omni, Alibaba’s latest multimodal AI model, renowned for its ability to process various types of inputs simultaneously, including text, images, audio, and video.

• Multimodal Capabilities: Unlike traditional models that handle one type of input at a time, Qwen 2.5 Omni integrates multiple modalities within a single framework, allowing for real-time understanding and responses. Bailey describes it as "an all seeing, all hearing AI" (05:09).

• Real-Time Interaction: The model can generate text and synthesize natural speech instantaneously, making it highly suitable for interactive applications. Alex marvels, “So it can understand what you're saying or showing it and respond instantly with either text or spoken words?” (05:46).

• Innovative Architecture: The Thinker-Talker architecture is highlighted as a key innovation. The "Thinker" processes and understands the inputs, while the "Talker" converts this understanding into natural-sounding speech, all within a cohesive, end-to-end trained model (06:21).

• Performance and Accessibility: Qwen 2.5 Omni outperforms single-modality models and is available to researchers and developers through platforms like Hugging Face and GitHub. Its novel time-aligned multimodal rope position embedding ensures nuanced understanding of audiovisual data (07:16).

• Future Applications: The hosts discuss potential applications in accessibility, personalized education, customer service, and interactive entertainment, envisioning a future where such AI models revolutionize various sectors (08:43).

Notable Quote:
Bailey underscores the model’s impact by stating, "They also have a longer term vision of integrating even more modalities" (08:32).

---

OpenAI Adopts Anthropic’s Model Context Protocol (MCP)

Timestamp 09:25 – 12:04

Next, Alex and Bailey delve into OpenAI's strategic move to adopt Anthropic’s Model Context Protocol (MCP), an open-source standard designed to enhance interoperability among AI assistants.

• Interoperability and Connectivity: MCP allows AI models to connect with various data sources, enabling them to pull information from business tools, software applications, content repositories, and development environments. Bailey explains, "So it allows them to draw data from a variety of sources" (10:06).

• Bidirectional Data Flow: The protocol supports two-way communication between AI applications and data sources, allowing not only data retrieval but also interactions that can modify or update the connected systems (10:22).

• Industry Support and Integration: OpenAI's CEO, Sam Altman, has endorsed MCP, planning to integrate it across products like ChatGPT and their responses API. Other companies such as Block, Apollo, Replit, and Codeum are also supporting and integrating MCP, fostering a more interconnected AI ecosystem (11:13).

• Practical Implications: For users, this means more intelligent and context-aware applications that can seamlessly integrate into their workflows. For instance, a chatbot integrated with MCP could access a company's knowledge base or project management software to provide more accurate and relevant responses (10:30).

Notable Quote:
Sam Altman highlights the importance of MCP by stating, "The true potential of AI is unlocked when it can interact with the real world" (11:31).

---

The Rise of Deepfakes and Get Real’s Response

Timestamp 12:07 – 15:01

The final segment addresses the growing menace of deepfakes and introduces Get Real, a startup dedicated to combating this threat through advanced detection technologies.

• Threat of Deepfakes: Deepfakes pose significant risks, including scams, misinformation, and threats to national security. Get Real is focused on developing technologies to detect deepfakes across audio, video, and image mediums (12:11).

• Funding and Support: Recently securing $17.5 million in funding, Get Real is launching its forensic platform as a service, which includes a web interface, API, threat exposure dashboard, and specialized tools like Inspector for protecting executives from digital spoofing (12:32).

• Expertise and Leadership: Co-founded by Hani Farid, a pioneer in deepfake detection, Get Real benefits from strong leadership and credibility. The involvement of ForgePoint Capital and Ballistic Ventures further underscores the seriousness of their mission (13:00).

• Industry Adoption: Get Real has attracted clients from heavily regulated industries such as financial institutions and government agencies, which are particularly vulnerable to deepfake threats. Major customers include John Deere and Visa (14:29).

• Future Plans: While currently focused on audio, video, and image deepfakes, Get Real plans to tackle text-based impersonations in the future, aiming to provide comprehensive protection against all forms of digital deception (14:34).

Notable Quote:
Get Real CEO Matt Moynihan describes deepfakes as "a ubiquitous threat" likening them to computer viruses, highlighting the pervasive vulnerability in our increasingly digital business landscape (13:40).

---

Conclusion and Ethical Considerations

Timestamp 15:01 – End

As the episode wraps up, Alex and Bailey reflect on the interconnectedness of these AI advancements and their implications for the future.

• Rapid AI Integration: The discussed technologies – personalized shopping, multimodal AI, open standards, and deepfake detection – all signal a future where AI plays a pivotal role in everyday life and various industries.

• Ethical Implications: The hosts prompt listeners to consider the ethical dimensions of these advancements, such as privacy concerns, the potential for filter bubbles in personalized shopping, and the importance of maintaining trust and security in the digital realm (15:19).

• Future Exploration: They encourage ongoing exploration and discussion about how to navigate the ethical landscape shaped by AI’s rapid advancement, emphasizing the need for proactive measures to address emerging challenges.

Notable Quote:
Bailey concludes, "I think they're going to be absolutely essential for maintaining trust and security in the digital world" when discussing the role of tools like Get Real in combating deepfakes (14:54).

---

Key Takeaways

1. Amazon’s Interests Feature: Represents a significant leap in personalized shopping through advanced AI, enhancing user experience by understanding natural language queries and providing tailored recommendations.

2. Alibaba’s Qwen 2.5 Omni: Showcases the future of multimodal AI with real-time processing and response capabilities across multiple data types, promising wide-ranging applications from accessibility to interactive entertainment.

3. OpenAI’s Adoption of MCP: Marks a move towards greater AI interoperability, enabling AI assistants to seamlessly integrate with diverse data sources and applications, thereby unlocking their full potential.

4. Get Real’s Deepfake Solutions: Highlights the urgent need for robust deepfake detection technologies to safeguard against the growing threats of digital deception in various sectors.

5. Ethical Considerations: As AI technologies become more pervasive, it is crucial to address the ethical implications to ensure trust, security, and equitable access in an AI-driven future.

---

This episode of the AI Deep Dive Podcast offers a comprehensive overview of significant AI developments that are shaping industries and daily life. From enhancing online shopping experiences to pioneering multimodal AI and establishing open standards for AI interoperability, the discussed advancements underscore the transformative power of artificial intelligence. Simultaneously, the pressing issue of deepfake threats and the innovative measures to counter them highlight the dual-edged nature of AI progress. As AI continues to evolve, the conversation around ethical considerations remains paramount, inviting listeners to engage thoughtfully with the technologies that are redefining our world.

Transcript

[Alex]

Ever get that feeling like you're just completely overwhelmed with information?

[Bailey]

Oh, yeah.

[Alex]

Every day. It's just this flood of news and studies and updates. And honestly, it can be really hard to figure out what actually matters, you know, what's actually going to change things.

[Bailey]

Yeah. Distilling actual insight from all that noise. It really takes focus and understanding the bigger picture. Having a targeted approach to your information is really crucial.

[Alex]

That's the whole idea behind the Deep Dive. Right. You send us what you're reading and what you're interested in and we become your expert guides. We pull out the key takeaways, the aha moments.

[Bailey]

Exactly. So you can stay ahead of the curve without feeling overwhelmed, like you're drowning.

[Alex]

Yeah. And today we are diving deep into the world of AI.

[Bailey]

We are.

[Alex]

You put together a really fascinating set of news items for us.

[Bailey]

I did.

[Alex]

And our mission today is to dig into some of these latest advancements and explore how they might impact our daily lives in different industries.

[Bailey]

Absolutely.

[Alex]

So what do we have on the docket for today?

[Bailey]

Well, we've got a really interesting mix. We're going to be looking at how AI is starting to personalize our shopping experience. Some really cutting edge developments in AI that can understand all sorts of information.

[Alex]

Wow.

[Bailey]

The growing movement behind making AI data more openly accessible. And then finally, the really vital work that's being done to get a handle on deepfakes.

[Alex]

Okay, a lot to unpack there. The first thing that jumped out at me was this news about personalized AI shopping. Amazon seems to be taking the idea of like tailored recommendations to a whole new level with this thing called Interests. So tell us more about that.

[Bailey]

Yeah, this Interests feature, it's really interesting. Amazon is leveraging the power of AI to create a really intuitive and personalized way to shop online. The feature essentially allows you to use everyday language when you're searching. You can reflect your specific hobbies, passions, and even, you know, what kind of budget you have in mind.

[Alex]

So instead of just typing coffee maker, you could say something like brewing tools and gadgets for coffee lovers.

[Bailey]

Exactly. The reporting actually mentioned examples like model building kits and accessories for hobbyist engineers too. The real breakthrough here is that they're using large language models to understand those natural language queries.

[Alex]

Okay.

[Bailey]

The same technology that powers a lot of the generative AI we're seeing. This should lead to much more relevant product suggestions.

[Alex]

Yeah, it's moving beyond just surface level matching to actually understanding what the user is trying to find. And it sounds like this isn't just a one off search thing. Either.

[Bailey]

Right. It's designed to work in the background for you.

[Alex]

Okay.

[Bailey]

It'll actually send you notifications about new products that align with the interest you've specified. Interesting updates on items that you might be keeping an eye on. You know, like if they come back in stock or if there's any special.

[Alex]

Offers that could actually be super useful for spaying on top of things that you're genuinely passionate about. Is this something that everybody can use right now?

[Bailey]

So right now it's only being rolled out to a limited number of users in the US and this is through the Amazon Shopping app on on iOS and Android. It's also accessible on their mobile website. You can find it under the Me tab. They are planning to expand access to more customers across the US in the coming months.

[Alex]

Okay. And this Interests feature, it feels like it's not just happening in a vacuum. The articles mention a lot of other AI powered features that Amazon has already introduced.

[Bailey]

Absolutely. You can see a clear trend of Amazon's steadily weaving AI into the shopping experience. They already have Rufus, which is their AI shopping assistant.

[Alex]

Okay.

[Bailey]

AI powered shopping guides. AI generated summaries of customer reviews. They even have AI that helps generate product information.

[Alex]

Wow.

[Bailey]

The Interests feature really feels like the next step in this strategy of AI integration.

[Alex]

It's almost like they're trying to make the entire online shopping experience feel less like a transaction and more like an ongoing conversation.

[Bailey]

Exactly.

[Alex]

And it seems like Amazon isn't the only one that's exploring this approach.

[Bailey]

Right. Other companies are also starting to experiment with similar AI driven features. Google, for example, recently upgraded its Shopping tab with something called a Vision Match tool.

[Alex]

Okay.

[Bailey]

And this lets you describe a piece of clothing you have in mind, and their AI will suggest similar items. They also have an AI tool that summarizes product information.

[Alex]

I feel like these personalized AI powered shopping experiences are going to become really common.

[Bailey]

I think so too, across all sorts of online platforms.

[Alex]

So, thinking about you, the listener, how might these personalized features change the way you shop online? Could it make it easier to find what you're looking for? Or do you see potential downsides like getting stuck in a filter bubble?

[Bailey]

That's a great question and a really important one. While the convenience is appealing, it's important to be aware of how these systems might narrow our focus and limit our exposure to a broader range of products and ideas.

[Alex]

Okay, let's shift gears now to something that sounds incredibly cutting edge. The next piece of news is about Quinn 2.5 Omni from Alibaba. They're calling it an all seeing, all hearing AI.

[Bailey]

It's pretty extraordinary.

[Alex]

Yeah, that sounds pretty wild.

[Bailey]

It is. Quinn 2.5 Omni is Alibaba's latest end to end multimodal AI model.

[Alex]

Okay.

[Bailey]

And the key word here is multimodal. It's designed to process a really wide variety of inputs.

[Alex]

Like what?

[Bailey]

Text, images, audio and video.

[Alex]

All within a single model.

[Bailey]

All within a single model.

[Alex]

So it's not just understanding text in one part and images in another. It can take all of that in at the same time.

[Bailey]

Exactly.

[Alex]

That's pretty amazing.

[Bailey]

And what's particularly impressive is its ability to deliver responses in real time, both through generating text and synthesizing natural speech.

[Alex]

Wow. So it can understand what you're saying or showing it and respond instantly with either text or spoken words?

[Bailey]

That's right.

[Alex]

It's mind boggling that real time capability sounds super important for any interact applications. And is this something that researchers or developers can actually get their hands on?

[Bailey]

Yes. Alibaba has made it openly available on several platforms, Hugging Face, GitHub, as well as their own platforms, Model Scope and Dash Scope. They've also released really detailed documentation that explains its architecture and capabilities.

[Alex]

Okay. And the architecture itself sounds really innovative. This Thinker talker concept.

[Bailey]

So yeah, the Thinker talker architecture, think of the thinker component as the brain.

[Alex]

Okay.

[Bailey]

It's responsible for processing and understanding all those different kinds of inputs.

[Alex]

Right.

[Bailey]

And generating a high level understanding, then translating that understanding into text. Then you have the talker.

[Alex]

Okay.

[Bailey]

And this acts as the mouth.

[Alex]

Gotcha.

[Bailey]

It takes that high level understanding and the text and turns it into natural sounding speech.

[Alex]

So a two stage process, but it all happens seamlessly in real time.

[Bailey]

Precisely.

[Alex]

That's incredible.

[Bailey]

The reporting also emphasizes that the entire architecture is designed as one cohesive model which allows for end to end training. So this system learns and operates as one unit. They've also developed a novel time aligned multimodal rope or TM rope position embedding. And this is crucial for ensuring that when the AI processes a video with audio, it understands which words correspond to which visual elements.

[Alex]

That sounds like a small detail, but it makes a huge difference.

[Bailey]

It does. It makes its understanding much more nuanced, especially in real world scenarios.

[Alex]

Okay. And how does it actually perform? How does it stack up against other models?

[Bailey]

Well, the evaluations show really strong performance across all modalities. It outperforms similarly sized AI models that are only specialized in one area, like just audio or images. It even achieves comparable performance to Quin 2.5 VL7B, which is specifically designed for vision language tasks.

[Alex]

That's really impressive. So it's not just decent at everything, it's actually excelling. Yeah, it's very good. The article also mentions it's good at end to end speech instruction following. What does that mean?

[Bailey]

So this means it can understand and follow instructions given through spoken language just as well as if they were written down.

[Alex]

Wow.

[Bailey]

They run benchmarks like MMLU and GSMAK and it seems to support that.

[Alex]

So that opens up a lot of possibilities for voice based interactions with AI. What are Alibaba's future plans?

[Bailey]

So their immediate goals are to further refine its ability to understand voice commands and enhance its comprehension of audiovisual collaboration.

[Alex]

So understanding actions and events through sight and sound.

[Bailey]

Exactly.

[Alex]

Wow.

[Bailey]

They also have a longer term vision of integrating even more modalities.

[Alex]

So moving towards a true Omni model.

[Bailey]

Right. One that can handle a really broad range of sensory inputs.

[Alex]

What kind of applications do you see benefiting from this? The speed and versatility?

[Bailey]

I think a lot of them.

[Alex]

Yeah.

[Bailey]

You know, accessibility, personalized education. Even in crafting more immersive and interactive entertainment experiences, it's pretty remarkable.

[Alex]

Thinking about you, the listener, can you start to imagine all the potential applications for an AI like Quen 2.5? Omni could be a real game changer.

[Bailey]

Absolutely. Revolutionize customer service interactions, open up all new avenues in creative fields. It could really change the way we interact with technology.

[Alex]

Okay, let's move on to another significant development in AI. This news about OpenAI supporting Anthropic's model Context protocol. This sounds like a step towards more open standards in AI.

[Bailey]

It is. OpenAI's decision to adopt MCP is a move towards interoperability. So MCP is an open source standard that allows AI assistants to connect with different data sources.

[Alex]

Okay. So instead of each AI model operating in a silo, it can now tap into other systems.

[Bailey]

Exactly. Sam Altman actually voiced his support for MCP.

[Alex]

Who's that?

[Bailey]

Oh, he's the CEO of OpenAI.

[Alex]

Okay.

[Bailey]

And he said that they are planning to integrate it across all their products. That includes ChatGPT and their responses API.

[Alex]

That's a huge endorsement coming from a major player like that. What does McP allow these AI models to do?

[Bailey]

So it allows them to draw data from a variety of sources.

[Alex]

Okay.

[Bailey]

Business tools, software applications, content repositories, even app development environments.

[Alex]

Wow.

[Bailey]

So when you ask an AI assistant a question, it can actually access and incorporate information from these other sources.

[Alex]

So if I'm using a chatbot, that's integrated with mcp. It could pull information from my company's knowledge base. Yes, or my project management software.

[Bailey]

Precisely.

[Alex]

To answer my question, that's right.

[Bailey]

And it's not just information flowing into the AI. MCP also allows for connections that go both ways. Yeah. Between data sources and AI applications.

[Alex]

That sounds like it could be a game changer. The article mentions MCP servers and MCP clients. What are those?

[Bailey]

So MCP servers are the systems that are making their data accessible.

[Alex]

Okay.

[Bailey]

And MCP clients are the applications and workflows that connect to these servers.

[Alex]

So like chatbots.

[Bailey]

Exactly.

[Alex]

Okay, so it's a way for AI tools and data sources to communicate with each other.

[Bailey]

Yes, a standardized way.

[Alex]

And it sounds like Anthropic isn't the only one involved in this.

[Bailey]

No. Several other companies have announced support and are integrating mcp.

[Alex]

Like who?

[Bailey]

Block, Apollo, Replit, Codeum, sourcegraph. Wow. Mike Krieger, the Chief Product Officer of anthropic, welcomed OpenAI's decision.

[Alex]

Okay.

[Bailey]

He made a really important point. He said that large language models are most useful when they can connect to data and software that we're already using.

[Alex]

That makes sense. The true potential of AI is unlocked when it can interact with the real world. And OpenAI said they'll be sharing more details about their MCP integration plan soon.

[Bailey]

Yes.

[Alex]

This is definitely something to keep an eye on.

[Bailey]

Definitely. The adoption of open standards like this could lead to a much more interconnected AI ecosystem.

[Alex]

What kind of possibilities could this unlock? I'm thinking about the listener here. What could this mean for them?

[Bailey]

More powerful and integrated AI tools in their work and personal lives. More intelligent and context aware applications.

[Alex]

Okay. Finally, let's talk about the rise of deepfakes.

[Bailey]

Okay.

[Alex]

It seems like the startup Get Real is trying to tackle this head on.

[Bailey]

They are. The ease with which deepfakes can be created is a huge concern. The potential for scams, misinformation, even threats to national security. Get Real is focused on developing technologies to detect and stop these deep fakes across all mediums, audio, video and images.

[Alex]

And they just got a lot of funding to do this?

[Bailey]

Yeah. $17.5 million.

[Alex]

Wow.

[Bailey]

And they're also launching their forensics platform as a service.

[Alex]

Okay, so what's included in this platform?

[Bailey]

Well, it includes a web interface, an API, a threat exposure dashboard.

[Alex]

Okay.

[Bailey]

They also have a specialized tool called Inspector, designed to protect executives from being digitally spoofed.

[Alex]

Interesting.

[Bailey]

A protect tool for screening media and a respond capability which involves deeper analysis by their human experts.

[Alex]

And Hani Farid is one of the co founders.

[Bailey]

He is.

[Alex]

He's a pioneer in deepfake detection.

[Bailey]

Yeah.

[Alex]

That lends a lot of credibility to what they're doing.

[Bailey]

Absolutely. And ForgePoint Capital, a firm specializing in cybersecurity and AI, was the lead investor.

[Alex]

Okay.

[Bailey]

That shows how serious this threat is. Ballistic Ventures also participated.

[Alex]

And it sounds like Ballistic Ventures was involved early on.

[Bailey]

Yes. Getreal was actually incubated within Ballistic Ventures. They also led their seed funding round.

[Alex]

Wow. Okay.

[Bailey]

Ted Schlein, the founder of Ballistic, is also chairman and co founder of Getreal.

[Alex]

Wow. They're really intertwined. What does the GetReal CEO Matt Moynihan have to say about all of this?

[Bailey]

He calls deepfakes a ubiquitous threat. He says they're like computer viruses and that because business is increasingly digital, we're all vulnerable.

[Alex]

It sounds like they're trying to scale Hani Farid's expertise.

[Bailey]

They are. Hani's service.

[Alex]

That's funny.

[Bailey]

Yeah. They want to turn his understanding into code and services that are widely accessible.

[Alex]

And who else is backing them?

[Bailey]

Well, Cisco Investments, Capital One Ventures, and.

[Alex]

In qtel, in Q Tel. They have ties to the CIA?

[Bailey]

They do.

[Alex]

Okay, so this is getting some serious attention.

[Bailey]

It is. Heavily regulated industries are particularly interested in.

[Alex]

Solutions like this, like financial institutions and government agencies.

[Bailey]

Right. They're worried about CEO impersonations and being tricked by fake information.

[Alex]

That makes sense.

[Bailey]

They already have big customers like John Deere and Visa.

[Alex]

Wow.

[Bailey]

But right now they're focused on audio, video and image deepfakes, not text.

[Alex]

Oh, okay. Interesting.

[Bailey]

Reid says that text based impersonations are a different challenge. They will address them in the future, but not right now.

[Alex]

So, listener, how do you feel about all of this? Deepfakes are getting more sophisticated. They're getting harder to detect. What do you think about the role of tools like Get Real in safeguarding us?

[Bailey]

I think they're going to be absolutely essential for maintaining trust and security in the digital world. We have to be able to detect these things early.

[Alex]

Yeah. Well, this has been a fascinating and at times unsettling look at recent developments in AI.

[Bailey]

It has.

[Alex]

We talked about personalized online shopping, the capabilities of multimodal AI models, open standards for data access, and the threat of deepfakes.

[Bailey]

A lot to think about.

[Alex]

It is. And while these may seem like separate areas, they all point towards a future that is heavily influenced by AI rapid advancement. So, listener, what new ethical considerations do you think we need to be prepared for in this AI driven world?

[Bailey]

Lot to consider.

[Alex]

It is. It's definitely something to explore further.