- The Rundown AI
- Posts
- AI avatars outsell humans in 6-hour livestream
AI avatars outsell humans in 6-hour livestream
PLUS: OpenAI lands $200M Pentagon defense contract
Zach Mink & Rowan Cheung
June 18, 2025
Good morning, AI enthusiasts. The future of e-commerce just arrived in China, with an entrepreneur’s six-hour AI avatar livestream raking in millions and outperforming real human creators.
With hyper-realistic AI twins now able to effectively sell products 24/7, is this streaming success a peek at a new blueprint for retail's automated future?
In today’s AI rundown:
China’s AI avatars outsell humans in livestream
OpenAI lands $200M Pentagon contract
How to design perfect interfaces with v0
Gemini 2.5 family goes GA with new flash-lite
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
AI AVATARS
Image source: Baidu
The Rundown: Chinese entrepreneur Luo Yonghao just utilized an AI digital twin on a six-hour livestream on Baidu’s e-commerce platform, outperforming his human-led streams with over $7M in sales generated during the broadcast.
The details:
Two AI-generated hosts promoted 133 products in the session, showcasing items while utilizing human gestures and handling real-time viewer interactions.
The stream reached 13M viewers and beat Luo’s “real” stream in May in just 26 minutes, with Baidu’s ERNIE crafting 97K+ characters of product descriptions.
Baidu said the stream was the first to feature “dual digital avatars”, with Luo and his digital co-host interacting in natural conversation and movements.
Over 100k digital humans reportedly work in China's $946B live commerce sector, slashing costs by 80% and increasing transactions by 62% on average.
Why it matters: It’s hard to compete with an AI salesperson able to perfectly (and endlessly) sell products, especially when the difference in realism is imperceptible for the average viewer. The question turns to whether authentic human interactions retain any commercial advantage in markets where efficiency often trumps everything else.
TOGETHER WITH SANA
The Rundown: Sana Agents provides a no-code platform for deploying AI across your entire organization, enabling teams to automate complex workflows and integrate AI directly into existing tools in minutes, not months.
With Sana Agents, you can:
Seamlessly connect AI to 100+ platforms, including Slack, Teams, and your CRM
Build specialized agents for different departments in one intuitive interface
Ensure enterprise-grade security with zero data leakage
Slash operational overhead by 10-20X
OPENAI
Image source: o3 / The Rundown
The Rundown: OpenAI announced “OpenAI for Government,” an initiative that brings the company’s federal efforts under a single umbrella — while also securing a $200M DoD contract to address national security issues in “warfighting and enterprise” areas.
The details:
The one-year deal marks OpenAI's debut as an official Pentagon contractor, with work centered in the Washington, D.C. region.
ChatGPT Enterprise will aid service members in admin tasks like navigating benefits, with custom models tackling areas like proactive cyber defense.
“OpenAI for Government" moves existing partnerships with NASA, NIH, Air Force Research Lab, and Treasury under a single initiative.
The DoD’s contract listed the role as developing “prototype frontier AI” for “warfighting and enterprise,” though OAI said it would follow usage policies.
Why it matters: Between AI giants revising policies, partnering with governments and defense firms, and the modernization of warfare with AI and drone tech, it’s clear where things are trending. Government usage and control of top models could (unfortunately) be the key to the global military powers that define future conflicts.
AI TRAINING
The Rundown: In this tutorial, you will learn how to use v0's new Design Mode to create stunning user interfaces with AI and then manually fine-tune every element without additional prompts or credits.
Step-by-step:
Visit v0 and prompt your desired user interface, e.g., "Create a landing page for a content creation AI tool"
Click the "Design" tab to enter manual editing mode
Select any element (text, buttons, sections) to customize copy, typography, layout, colors, and styling
Preview changes in real-time and click "Save" when ready
Pro tip: Use detailed AI prompts for the foundation (80%), then Design Mode for perfection (20%).
PRESENTED BY IBM
The Rundown: Come meet the next evolution of IBM® watsonx.data®, an open, hybrid lakehouse built to help improve AI accuracy.
Watsonx.data is designed to:
Automate the ingestion, enrichment and governance of data at scale
Optimize for cost and performance through purpose-built query engines
Work across on-prem, cloud and multi-cloud environments
Help improve the accuracy of gen AI apps
What else can watsonx.data do for you? See for yourself.
Image source: Google
The Rundown: Google just graduated its Gemini 2.5 Pro and Flash models to stable production status after months of preview version updates, while also rolling out a new, hyper-efficient Flash-Lite variant.
The details:
The Gemini 2.5 Pro and Flash models exit preview and are now generally available, with Pro topping the leaderboards alongside OpenAI’s o3-pro.
2.5 Flash-Lite launches in preview, beating previous Lite models across benchmarks while maintaining the massive 1M token context window.
All three models feature adjustable "thinking" capabilities that let users control reasoning and cost, with Lite defaulting to thinking off for maximum speed.
Why it matters: After making a series of impressive “preview” updates to Gemini’s 2.5 family, Google is finally cementing the models in with a GA launch. The move now turns focus away from incremental updates and towards the teased 2.5 DeepThink and eventual larger leap forward with the Gemini 3 family.
QUICK HITS
🔁 The Rundown - Strategic Partnerships (AI University)
📦 Figure AI - Global Supply Manager, Electronics
🔬 Meta - Research Scientist
🖥️ xAI - Datacenter Technician Lead
MiniMax debuted Hailuo 02, a new AI video model (tested under the “Kangaroo” codename) that moves to No. 2 on the Artificial Analysis leaderboard, passing Veo 3.
Amazon CEO Andy Jassy said in a letter to employees that the company’s AI push will trim its corporate headcount in the coming years with agents and automation advances.
Krea AI launched its debut Krea 1 image model as a free public beta, showcasing advanced style control and image quality.
Intelligent Internet introduced an updated version of its open II-Medical model, surpassing Google’s MedGemma across benchmarks despite its smaller size.
Adobe released new mobile apps for its Firefly platform, allowing users to access its AI image, video, and other creative tools via iOS and Android.
xAI is reportedly aiming to raise $4.3B in new funding for its AI operations, with the company valued at $80B as of the end of Q1.
COMMUNITY
Join our next workshop this Friday, June 20th, at 4 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. By the end of the workshop, you’ll be able to confidently build and run your automation agents using n8n.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
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