Google Puts Ads in AI Chats, Duolingo Faces AI Backlash, & Visa & Tinder Adopt Worldcoin

Summary

AI Deep Dive Podcast Summary
Episode: Google Puts Ads in AI Chats, Duolingo Faces AI Backlash, & Visa & Tinder Adopt Worldcoin
Release Date: May 2, 2025
Host: Daily Deep Dives

---

Introduction

In this episode of the AI Deep Dive Podcast, hosts A and B explore four major developments in the artificial intelligence landscape: Duolingo's transition to AI-generated language courses, Worldcoin's expansion into biometric ID partnerships with giants like Visa and Tinder, Google's initiative to embed advertisements within AI chatbots, and the controversy surrounding the LM Arena AI benchmark's alleged favoritism towards major AI labs. Each topic is dissected to reveal the implications for technology, businesses, and users alike.

---

1. Duolingo's AI-Driven Language Courses

Duolingo has embarked on a transformative journey by launching 148 new AI-generated language courses, marking their most significant content expansion to date. This shift to an AI-first model underscores Duolingo's commitment to leveraging artificial intelligence for rapid and high-quality content creation.

• Notable Quote:
  “He [CEO Louis Von Ahn] sees it as vital for their mission, teaching, reaching more learners fast, unprecedented speed and quality.”
  (00:43)

The integration of AI allows Duolingo to produce language courses at a pace that would have taken twelve years to achieve manually. However, this move isn't without its drawbacks. The transition entails a reduced reliance on human contractors, sparking concerns among both employees and users.

• Notable Quote:
  “The message was gradually stop using contractors to do work that AI can handle and new hires only if you absolutely can't automate it.”
  (02:05)

The backlash was immediate, with users expressing fears that the AI-driven approach might compromise the quality and effectiveness of language learning. Some even deleted the app, highlighting a significant apprehension about the diminishing human element in educational tools.

• Notable Quote:
  “There’s this real concern about losing the human element in language learning.”
  (02:30)

Duolingo maintains that the initial focus is on beginner-level courses, such as Stories and Duo Radio features, with plans to expand into more advanced topics in the future. This strategy aims to balance rapid content generation with maintaining educational standards.

---

2. Worldcoin's Biometric ID Partnerships with Visa and Tinder

Worldcoin, known for its Orb biometric scanning technology, is strategically expanding its footprint through partnerships with major companies like Tinder and Visa. These collaborations aim to mainstream biometric ID verification, enhancing security and user verification processes across platforms.

• Notable Quote:
  “The Orb might have been a bit cumbersome. Maybe a barrier for some. The card, the Stripe integration. That feels like a push for much broader adoption.”
  (04:11)

With Tinder in Japan, Worldcoin seeks to verify users as genuine individuals, reducing the prevalence of fake profiles. Additionally, partnerships with Kalshi and Morpho enable users to log into services using their World ID, streamlining access and enhancing security.

The most prominent partnership with Visa introduces the World Card, slated for a U.S. release later in 2025. This card facilitates cryptocurrency spending anywhere Visa is accepted by automatically converting crypto to fiat currency. Future incentives may include rewards linked to AI subscriptions, further integrating Worldcoin's technology into everyday financial transactions.

• Notable Quote:
  “It connects to their World app and lets you spend crypto anywhere. Visas accepted.”
  (03:44)

Moreover, an upcoming collaboration with Stripe aims to enable World ID payments on websites, although a release date has yet to be announced. These initiatives signify Worldcoin's efforts to transition from niche biometric scanning to a pervasive tool embedded in daily financial and online interactions.

• Notable Quote:
  “It's about making the ID useful beyond just getting scanned by the Orb.”
  (03:34)

This expansion raises pivotal questions about privacy and the societal acceptance of biometric IDs for routine activities, highlighting the delicate balance between technological advancement and user comfort.

---

3. Google's Integration of Ads into AI Chatbots

Google is venturing into the realm of advertising within AI chat interfaces. According to Bloomberg, Google's AdSense network now supports the placement of ads inside chats facilitated by third-party AI chatbots such as Iask and Liner.

• Notable Quote:
  “They see the shift people using chatbots for searches and they want their ad model to work there too.”
  (05:11)

This initiative appears to be a strategic move to maintain Google's dominance in the advertising sector as user behavior shifts towards conversational AI for information retrieval. By embedding ads directly into chat responses, Google aims to ensure that its AdSense remains relevant in an evolving digital landscape.

• Notable Quote:
  “Will it feel intrusive? Will people just accept it? Could it, you know, subtly influence the answers the AI gives.”
  (05:34)

The integration of ads into chatbots poses significant questions about user experience and the potential manipulation of information. Unlike traditional search page ads, these in-chat ads could blend seamlessly into conversations, making them less conspicuous and potentially more persuasive.

• Notable Quote:
  “You definitely could blur the lines. It’s different from seeing an ad clearly marked on a search page.”
  (05:48)

Google's own AI projects, like Gemini, alongside these advertising strategies, suggest a comprehensive approach to embedding AI deeper into both user interactions and monetization models.

---

4. LM Arena Benchmark Controversy

The LM Arena, a prominent AI benchmark platform, is embroiled in controversy following a study by researchers from Cohere, Stanford, and MIT AI2. The study alleges that LM Arena has been favoring major AI labs—such as Meta, OpenAI, Google, and Amazon—thereby skewing performance rankings and undermining the benchmark's credibility.

• Notable Quote:
  “The core accusation is that LM Arena let companies like Meta, OpenAI, Google, Amazon privately test lots of different versions of their models.”
  (06:14)

Specifically, it is claimed that LM Arena permitted these companies to conduct extensive private testing of their AI models, subsequently releasing only the top-performing scores. For instance, Sara Hooker from Cohere noted that Meta tested 27 versions of Llama 4 but initially disclosed only one superior score.

• Notable Quote:
  “If true, it absolutely could. El Amarina is a really popular benchmark. If people think it’s not a level playing field, it’s. It damages trust in how we compare these models.”
  (06:50)

LM Arena's co-founder, Ion Stoica, has responded by branding the study as inaccurate and questioning its analysis. He reaffirms LM Arena's commitment to fair evaluations and encourages more providers to submit diverse models to restore trust.

• Notable Quote:
  “They’ve said they’ll create a new sampling algorithm, but they rejected the calls for like transparent limits on private testing or making pre-release scores public.”
  (07:41)

Additionally, past incidents—such as Meta optimizing a version of Llama 4 specifically for the Chatbot Arena without releasing it—have further fueled skepticism about the fairness of LM Arena's benchmarking processes.

• Notable Quote:
  “Tuning a model specifically to ace the benchmark's conversationality metric. So there's a pattern of questions being raised.”
  (07:58)

As LM Arena plans to transform into a standalone company and seek investment, the emphasis on transparency and equitable assessment becomes even more critical to maintaining industry trust and integrity.

---

Conclusion

This episode of AI Deep Dive traverses the expansive and often contentious terrain of artificial intelligence advancements and their societal implications. From Duolingo's ambitious AI integration in education and Worldcoin's biometric ID expansions to Google's advertising strategies within AI chats and the ethical dilemmas posed by LM Arena's benchmarking practices, each discussion underscores AI's profound and multifaceted impact.

• Recap Quote:
  “Each one gives you a really glimpse into how AI is just rapidly changing things. Tech advances clashing with human impact, privacy concerns, new business models, and fundamental questions.”
  (08:46)

As AI technologies continue to evolve at a breakneck pace, the podcast invites listeners to ponder critical questions about responsible development, ethical safeguards, and the envisioned future shaped by these innovations.

• Closing Quote:
  “As AI keeps racing ahead, what principles, what safeguards do you think are most important to make sure it develops responsibly, beneficially.”
  (09:13)

The hosts hint at delving deeper into these ethical considerations in future episodes, encouraging ongoing reflection and dialogue within the AI community.

---

Tune in next time as Daily Deep Dives continues to explore the dynamic world of artificial intelligence, uncovering the trends and debates that will shape our future.

Transcript

[A]

Okay, let's unpack this. Today's deep dive, we're looking at some pretty significant AI developments. Stuff you flagged from AI deep dives News roundup.

[B]

Right, you wanted the quick take the key shifts without getting bogged down in all the technical jargon.

[A]

Exactly. So our mission basically is to pull out those crucial bits and, you know, connect the dots for you.

[B]

Makes sense. And we've got four main things on the docket today. Okay, first up, Duolingo. They've gone massive with AI generated language courses. Shifting to be AI first, which has, well, implications for contractors.

[A]

Right. Then world. The eyeball scanning ID company.

[B]

Yeah, they're partnering up with Tinder, Visa, some others, trying to get that biometric tech more mainstream.

[A]

Okay, interesting. What's third?

[B]

Third is Google. Reports say they're starting to put ads inside chats with some of those third party AI chatbots.

[A]

Ads in the chat. Okay.

[B]

And finally, and finally a bit of controversy. A study's accusing element, you know, the chatbot arena benchmark people of maybe tilting the scales for the big AI labs.

[A]

Hmm. Okay, so quite a range. Education, ID advertising, even how we measure AI itself.

[B]

Exactly. Different facets, but all showing AI's growing footprint.

[A]

Alright, let's, let's kick things off with Duolingo then. 148 new courses, AI generated. That sounds huge.

[B]

It is huge. I mean think about it, it took them something like 12 years to build their first hundred courses.

[A]

Wow. So this is their biggest content expansion.

[B]

Ever by far, by a long shot. And the CEO, Louis Von Ahn, he's quite clear this is down to their investment in AI and automation.

[A]

He said it lets them create content at, what was it? Unprecedented speed and quality.

[B]

That's the quote. He sees it as vital for their mission, teaching, reaching more learners fast, unprecedented speed.

[A]

Okay, but speed isn't everything, is it? Because this came alongside another announcement.

[B]

Right, the internal stuff moving to an AI first model.

[A]

And that means less reliance on human contractors.

[B]

Pretty much. The message was gradually stop using contractors to do work that AI can handle and new hires only if you absolutely can't automate it.

[A]

Okay. And that, well, that didn't go down too well, did it?

[B]

No. Definitely sparked some backlash. We saw it on social media. Users worried the AI focus means, you know, mistakes, lower quality content.

[A]

Some people actually deleted the app.

[B]

Yeah, some deleted it, encouraged others to do the same. There's this real concern about losing the human element in language learning.

[A]

I guess it's a tricky balance. They do say these new courses are mostly for beginners, Right? Like the Stories and Duo Radio features.

[B]

Correct. They're starting there and aiming for beginner levels. And they do plan more advanced stuff later.

[A]

Still, it makes you wonder about the future for language apps and, well, human teachers too.

[B]

Absolutely. How far can AI go effectively, especially with the nuances of language? That's the big question.

[A]

Okay, let's switch gears. World, the Orb company, Eyeball Scanning. What are they up to?

[B]

They're making some pretty strategic moves to get their biometric ID tech used more widely. Partnerships are key here.

[A]

Like with Tinder in Japan. Yep.

[B]

With Match Group for Tinder in Japan, the goal is verifying users are, well, human.

[A]

Verifying humanness. Okay. And others.

[B]

They've also teamed up with Kalshi, that's a prediction market startup, and Morpho, a decentralized lending platform.

[A]

So you could log into those using your World id.

[B]

Exactly. Using your existing World id. It's about making the ID useful beyond just getting scanned by the Orb.

[A]

But the really big one seems to be this Visa partnership. The World Card.

[B]

Yeah, that's definitely grabbing headlines. Planned for the US later this year. It connects to their World app and.

[A]

Lets you spend crypto anywhere. Visas accepted.

[B]

That's the idea. It converts crypto to fiat automatically. Plus they're hinting at rewards may be tied to like AI subscriptions.

[A]

Huh. And Stripe too. Paying with World ID on websites.

[B]

Also in the works, partnering with Stripe for online payments. No firm date on that one yet though.

[A]

So they raised a lot of money, signed up millions with the Orbit. This feels like the next phase, right? Trying to overcome the whole scan your eyeballs thing.

[B]

Precisely. The Orb might have been a bit cumbersome. Maybe a barrier for some. The card, the stripe integration. That feels like a push for much broader adoption. Embedding it into everyday transactions.

[A]

It really does push the question, doesn't it? How comfortable are we getting with biometric IDs for just everyday stuff?

[B]

It's a leap for sure. From unlocking your phone to potentially linking your financial life to an iris scan.

[A]

Let's pivot again. Google and ads in AI chatbots.

[B]

Yeah. Bloomberg reported this Google's AdSense network is apparently now supporting ads within chats with some third party AI bots.

[A]

They tested this first, right? With a couple of AI search startups.

[B]

Uh huh. Iask and Liner were mentioned as test partners and Google confirmed it. AdSense for search is available for sites wanting ads in their conversational AI.

[A]

So this feels like Google trying to stay relevant as people use chatbots like ChatGPT or Claude More for finding info Maybe protecting their search ad business.

[B]

Seems very likely. They see the shift people using chatbots for searches and they want their ad model to work there too. Remember, they're already putting ads in their own AI overviews in search, right?

[A]

Right.

[B]

So extending it to third party bots using their network. It fits the pattern. They're investing heavily in their own AI, like Gemini, but also hedging their bets, perhaps.

[A]

What do you make of ads inside a chat though? Does it change the experience?

[B]

That's the million dollar question, isn't it? Will it feel intrusive? Will people just accept it? Could it, you know, subtly influence the answers the AI gives?

[A]

Yeah, that's the worry. Does the AI start pushing sponsored links disguised as helpful info?

[B]

You definitely could blur the lines. It's different from seeing an ad clearly marked on a search page. Feels more integrated, potentially more persuasive, maybe.

[A]

Okay, one more big topic. This controversy around LM arena, the AI benchmark.

[B]

Right. This is pretty significant. A study came out. Researchers From Cohere, Stanford, MIT AI2 basically accusing LM arena of giving some big AI labs a leg up.

[A]

A leg up how?

[B]

The core accusation is that LM arena let companies like Meta, OpenAI, Google, Amazon privately test lots of different versions of their models.

[A]

Okay.

[B]

And then they only had to publicly release the scores for the best ones. But the study claims this private testing fast track wasn't open to everyone. They call it gamification.

[A]

So they could iterate privately until they got a great score, then just show that one while others couldn't.

[B]

That's the allegation. Sara Hooker from Cohere gave an example. Meta supposedly tested 27 versions of llama 4 privately, but only the score for one top performer was revealed initially.

[A]

That does sound like it could skew the rankings.

[B]

If true, it absolutely could. El Amarina is a really popular benchmark. If people think it's not a level playing field, it's. It damages trust in how we compare these models.

[A]

Has Ella Marina responded?

[B]

Oh, yeah. Ion Stoica, one of the co founders, called the study inaccurate and the analysis questionable. He insists they're committed to fair community evaluations and wants all providers to submit more models.

[A]

But the study had other points too, right? Something about sampling rates.

[B]

Yes. They also claimed Ella Marina gave certain companies models more battles, more exposure to user comparisons. More data, basically.

[A]

And more data could help them perform better on other benchmarks LM Marina runs.

[B]

The researchers estimate it could give a significant boost on, say, the arena hard benchmark. Ela Marina disputes a direct link, but acknowledges the sampling point somewhat.

[A]

So what's Ela Marina doing about it?

[B]

They've said they'll create a new sampling algorithm, but they rejected the calls for like transparent limits on private testing or making pre release scores public.

[A]

And this comes after that other issue with meta Optimizing a llama 4 version just for Chatbot arena but not releasing it.

[B]

Exactly. That raised eyebrows too. Tuning a model specifically to ace the benchmark's conversationality metric. So there's a pattern of questions being.

[A]

Raised, especially since Ala Marino wants to become a company now, raise money.

[B]

Right. That makes transparency and perceived fairness even more crucial. How can we trust the benchmark if there are these kinds of questions swirling around?

[A]

Okay, so let's just quickly recap then. We've covered a lot of ground.

[B]

We have Duolingo's big AI push in language learning and the human cost questions.

[A]

That raises then World trying to weave its biometric ID into payments and online logins via partners like Visa and Tinder.

[B]

Google experimenting with putting ads directly into AI chat conversations.

[A]

And finally this controversy over whether AI benchmarks like LM arena are truly fair or if they're being gamed by the big players.

[B]

Each one gives you a really glimpse into how AI is just rapidly changing things. Tech advances clashing with human impact, privacy concerns, new business models, and fundamental questions.

[A]

About how we even measure progress fairly.

[B]

Exactly. It all leads to a pretty big question, really, for you to mull over.

[A]

Yeah.

[B]

As AI keeps racing ahead, what principles, what safeguards do you think are most important to make sure it develops responsibly, beneficially?

[A]

What kind of future are these trends pointing towards?

[B]

That's the one. Something to think about. Maybe we can dive into some of those bigger ethical questions next time.