- The Rundown AI
- Posts
- Visa, Mastercard give AI credit cards
Visa, Mastercard give AI credit cards
PLUS: GPT-4o gets rolled back after personality trouble
Zach Mink & Rowan Cheung
May 01, 2025
Good morning, AI enthusiasts. The AI shopping revolution is here — and both Visa and MasterCard are laying the payment rails for the web’s agentic future.
With new platforms allowing agents to complete purchases on users’ behalf, the incoming AI commerce shift in retail is getting an essential upgrade.
In today’s AI rundown:
Visa, Mastercard give AI agents credit cards
GPT-4o rolled back after personality trouble
Build your AI consultancy prep assistant
DeepSeek's latest model masters math proofs
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
VISA & MASTERCARD
Image source: o3 / The Rundown
The Rundown: Visa just introduced Intelligent Commerce, enabling AI to shop and pay on consumers’ behalf through partnerships with Anthropic, OpenAI, and others, while Mastercard is releasing ‘Agent Pay’ to embed payments into AI conversations.
The details:
Intelligent Commerce uses AI-ready cards with tokenized credentials and user-set preferences to let AI agents find and buy items without exposing card data.
Consumers can set spending limits and conditions while sharing basic purchase data to help personalize shopping recommendations.
Mastercard’s ‘Agent Pay’ is a similar platform enabling easy payment experiences when interacting with AI agents to explore and shop products.
The news comes alongside ChatGPT Search’s shopping upgrades and other shopping-focused agentic efforts from Perplexity, Amazon, and others.
Why it matters: The next step in the evolution of e-commerce is looking more like AI commerce, with payment rails being laid by the legacy giants to let AI agents purchase items directly for users instead of just finding and recommending. While agents may not yet be as capable as many expected, their time is coming soon.
TOGETHER WITH FUEL iX
The Rundown: AI moves fast — but implementing it effectively is the key to success. Join Fuel iX and guest Forrester for an insightful discussion on how enterprises can harness the power of GenAI copilots to drive innovation and productivity across their organization.
This exclusive session will explore:
The current landscape of GenAI and why enterprises need to act now
Practical strategies to overcome barriers in AI implementation
Real-world success stories of AI copilots transforming businesses
Watch on-demand now for expert insights on enterprise AI success.
OPENAI
Image source: o3 / The Rundown
The Rundown: OpenAI just announced the reversal of a controversial GPT-4o update that made the model excessively agreeable and flattering in any context, igniting an industry-wide debate about AI personality tuning.
The details:
Last week’s GPT-4o update aimed at improving personality inadvertently led to excessive sycophancy, with the AI validating even poor or harmful user ideas.
OpenAI identified the cause as over-optimizing on short-term user feedback (like thumbs-up signals) without fully considering long-term interaction quality.
OpenAI Head of Model Behavior Joanne Jang held an AMA on Reddit, providing insights on model training and plans for personality customization.
Jang said the company is working on both a default personality for all users and preset offerings that users could customize on their own.
Why it matters: With hundreds of millions of ChatGPT users and models plugged into everything from mental health support to business operations, the stakes are high for tuning a personality that influences the masses. While OpenAI was quick to fix this issue, the sycophancy was glaring and viral — what happens when it’s more subtle?
AI TRAINING
The Rundown: In this tutorial, you will learn how to use Zapier Agents to create an automated consultant that researches clients before meetings and sends you a detailed brief, helping you deliver more insightful consultations.
Step-by-step:
Visit Zapier Agents, log in or create a free Zapier account, and click “Create New Agent.”
Add a name, select “Behavior”, set "Calendly: Invitee Created" as your trigger, and connect your Calendly account.
Add enhanced instructions: “Get client details from booking, research company challenges, compile insights, draft an email with summary and 3-5 strategic talking points.”
Add the “Gmail: Create Draft” action and test your consultant with the "Retest behavior" button.
Pro tip: For specialized consultations, modify your instructions to include industry-specific research points, competitor analysis, or market trends relevant to your expertise.
PRESENTED BY DROPBOX
The Rundown: Dropbox Dash is an AI universal search and knowledge management offering that searches videos, images, docs, and even your teammates to surface answers and spin up first drafts in seconds — assembling content into ready‑to‑share deliverables quickly and securely.
With Dropbox Dash, you can:
Locate any file, clip, or screenshot with natural‑language video & image search
Craft AI‑generated briefs, summaries, and plans built from existing content
Leverage deep integrations with Slack, Teams, Jira, Zoom, Canva, and more
Protect company data via self‑hosted AI and admin‑level permission controls
DEEPSEEK
Image source: o3 / The Rundown
The Rundown: Chinese AI lab DeepSeek just released Prover-V2, a specialized open-source model that combines informal math reasoning with formal theorem proving — achieving SOTA performance on complex math benchmarks.
The details:
The 671B parameter model achieves an 88.9% success rate on the MiniF2F test benchmark, setting new highs for automated theorem proving.
The system uses a "cold-start" approach that breaks down complex proofs into smaller subgoals using DeepSeek's V3 model before formal verification.
The team also introduced ProverBench, a new evaluation dataset with 325 problems, including AIME competition questions and undergraduate-level math.
The quiet open-source release comes shortly after Alibaba's Qwen3, and ahead of the highly anticipated DeepSeek-R2, expected in early May.
Why it matters: It may not be R2, but DeepSeek continues to quietly flex its abilities with yet another strong model release. It won’t be long before these models are solving problems once thought impossible — leading to superhuman math capabilities bringing new advances across areas like physics, drug discovery, and materials science.
QUICK HITS
🤖 Meta AI - Now available via a standalone app with enhanced personalization
🦙 Llama API - Free limited preview to build with Meta’s top models
🎧 Audio Overviews - Google’s AI deep dives, expanded to 50+ languages
✈️ Kayak AI - Plan trips and compare options with conversational AI
NVIDIA CEO Jensen Huang said that China is ‘not behind’ in AI, saying companies like Huawei are very close in the “long-term, infinite race.”
Ex-OpenAI CTO Mira Murati’s Thinking Machines Lab is reportedly nearing a $2B raise, with Murati said to have unique control of the company’s board votes.
Runway launched Gen-4 References to paid plans, allowing users to use photos, images, 3D models, or selfies to place a character into any scene with consistency.
Microsoft CEO Satya Nadella said in an interview at LlamaCon that as much as 30% of the company’s code is now written by AI, with a 30-40% acceptance rate.
Chinese tech giant Xiaomi introduced MiMo, a small 7B parameter open-source reasoning model that matches much larger rivals like o1-mini on math and coding tasks.
Freepik and Fal released F-Lite, a new open-source, open-weights image generation model trained on 100% licensed data.
Duolingo launched 148 new language courses in the “largest expansion of content in the company’s history,” coming on the heels of its transition to an AI-first organization.
COMMUNITY
Join our next workshop this Friday, May 2nd, at 4 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. By the end of the workshop, you’ll confidently be able to use Google NotebookLM to improve your research, studying, and teaching.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
That's it for today!Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you. |
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
Reply