☕️Meta's First AI Developer Conference, Google's AI Mode Rollout & Visa's Push For Shopping Via Chatbots

As another week comes to a curtain call, OpenAI hastily fixes ChatGPT's recent off-the-rails personality in time for the weekend, Duolingo doubles down on AI with new language courses, and Anthropic brings native Claude to even more apps.

Here are the main key takeaways:

• Meta's LlamaCon brings with it a standalone Meta AI app, a developer API, and new additions to the Phi-4 family

• Google's AI Mode for Google Search starts rolling out for select US users as it exits its experimental stage

• Visa wants to embed its payment network into AI chatbots to enable a new take on online shopping

Join us at AI Tangle as we untangle this week's happenings in AI!

Meta's first AI developer conference sees the company debut a standalone Meta AI app and a developer API

At its inaugural AI developer conference, LlamaCon, Meta announced the launch of a consumer AI chatbot app with a social feed and a developer-friendly Llama API to rival OpenAI's API services. Meta emphasized its strategy to "mix and match" top-performing open-source models, like China's DeepSeek or Qwen, which resulted in it launching reasoning variants of its Phi-4 small model family shortly after LlamaCon. Over the course of the conference, CEO Mark Zuckerberg framed open models as an "unstoppable force" as the company looks to align its models with recent EU AI Act incentives and regulations.

​

Google begins slowly rolling out its AI Mode feature tab on Google Search to a small subset of US users

​Google has announced that it has finally begun rolling out its "AI Mode" tab in Google Search to a small percentage of US users "in the coming weeks," moving the feature from its experimental state in Google Labs into its main search. Unlike AI Overviews, AI Mode provides its own AI-generated answers from Google's index under a dedicated tab beside "All," "Images," "Videos," and the other usual suspects. As for new features, AI Mode now saves past searches in a left-side panel and shows rich, clickable cards for comprehensive overviews of products, places, and businesses.

​

Visa announces a push to embed its payment network into the world's AI systems, from the US to the EU

​Visa announced on Wednesday that it is testing "intelligent commerce" by integrating its payment network into AI chatbots and agents from OpenAI, Anthropic, Microsoft, Perplexity, and Mistral. Users can securely store card credentials once and authorize purchases via chat, with some executives claiming it could rival some of e-commerce's largest shifts in the past. While the technology is said to be ready ready, as Mastercard itself is trying out a similar approach, trust and adoption remain the main hurdles - e-commerce still only accounts for a little under half of global retail.

​

OpenAI hits the breaks on an update that turned GPT-4o "too sycophant-y"

Last week, OpenAI released an update to ChatGPT's default AI model, GPT-4o, to make the model "more intelligent" and "improve its personality." However, the update turned problematic over the following days as users reported en masse that ChatGPT had become ludicrously flattering, annoying, and dangerously agreeing of the user, often described as sycophantic. After acknowledging the issue on Sunday, OpenAI announced in a recent blog post that it has reversed course and rolled back the update for 100% of free users, with paid ones soon to follow.

​

Duolingo launches 148 new AI-generated language courses after committing to become an "AI-first" company

Days after Duolingo made a controversial commitment to become "AI-first," the company announced that it has now "more than doubled" the number of language courses available, owing it to generative AI. The total amount of courses added amounts to 148 across its 7 most popular to-learn languages, which are now available to all 28 supported user interface languages. Duolingo spokesperson Sam Dalsimer says that the company has been experimenting with AI tech and content for years, stating that it would not be rolling out new course content "if we were not confident in AI."

​

Anthropic allows users to connect even more apps to Claude via Integrations, upgrades its Research tool

Anthropic's Claude now supports a new Integrations feature, letting the chatbot pull data from third-party cloud services via prebuilt connectors for 10 apps and custom integrations, as detailed in a blog post. Powered by the open-source MCP framework, Integrations gives Claude lots of leverage and actionability across tools. Along with Integrations, the upgraded Research tool now dissects complex queries into simpler concepts, spends 5–45 minutes gathering data, and delivers a cited, comprehensive report. Both features are live for subscribers of the Max, Team, and Enterprise plans, with Pro access on its way soon.

​Aha - Aha is an AI-powered influencer marketing team of agents that is designed to help boost drive sales, scale up business, and increase brand awareness.

​Copyleaks - Detect plagiarism throughout multiple languages with Copyleaks, the all-in-one intuitive and easy-to-use platform to help create and protect original content.

​Outerbase - Navigate your data with AI - View, edit, query, and visualize your data in one powerful and easy-to-use interface tool for your database.

AI Researchers' Secret Experiment on Reddit Gone Wrong (4-min read)

A group of researchers had AI bots pretend to be real people on the popular Reddit social media as part of a secret experimentation to see how AI can be used to influence human opinions. Now, with word having gotten out, users on the platform are calling for outrage - and a lawsuit.

🔮 AI Lesson - Prompting for Reasoning Models​

💡 AI CIO - What The Latest Verizon DBIR Says About AI​

🎯 The AI Marketing Advantage - The Hidden Cost Of Infinite AI-Generated Options​

📻 AI Confidential Podcast - Building Transparent, Open-Source AI With Sriram Raghavan