The Future of Shopping? AI + Actual Humans.

AI has changed how consumers shop by speeding up research. But one thing hasn’t changed: shoppers still trust people more than AI.

Levanta’s new Affiliate 3.0 Consumer Report reveals a major shift in how shoppers blend AI tools with human influence. Consumers use AI to explore options, but when it comes time to buy, they still turn to creators, communities, and real experiences to validate their decisions.

The data shows:

• Only 10% of shoppers buy through AI-recommended links

• 87% discover products through creators, blogs, or communities they trust

• Human sources like reviews and creators rank higher in trust than AI recommendations

The most effective brands are combining AI discovery with authentic human influence to drive measurable conversions.

Affiliate marketing isn’t being replaced by AI, it’s being amplified by it.

Download the full report to see what this means for your brand.

Hey there!

Three stories jumped out this week, and they all rhyme in a funny way: the industry is racing to make AI understand the world, feel more human, and move in the real world.

Today:

• AI Agents Run Businesses

• LeCun's New $5B Startup 

• Customizing ChatGPT's Tone

• Robot Backup Dancers

• AI Generates Image Layers 

• OpenAI Costs Dropping

AI earns while you sleep

Researchers at Andon Labs are testing whether self-running (autonomous) AI agents can operate real businesses. Their “vending bench” test gives each model $500 and a simulated snack stand. Early versions lost money, buying odd items like tungsten cubes, but updated agents—led by Google’s Gemini 3 Pro—now 10-fold profits by following strict checklists and using extra helper code. Adding a boss-like AI, “Seymour Cash,” further cut giveaways. 

The gap between playful mistakes and reliable performance is shrinking fast, yet procedures, oversight and specialized training remain vital. A new test will watch AI DJs build an online radio brand and earn sponsorships soon.

WATCH THE VIDEO ON YOUTUBE

Yann LeCun confirms his “world model” startup — and the valuation talk is… loud

Meta’s chief AI scientist (and deep learning legend) Yann LeCun confirmed he’s launching a new startup called AMI / AMI Labs, with Alex LeBrun (Nabla co-founder/CEO) transitioning to CEO while LeCun serves as executive chairman.

Here’s the spicy part: reporting pegs the fundraising target at €500M and a valuation around €3B (~$3.5B), while TechCrunch frames it as “$5B+” chatter depending on how people are rounding/positioning it. 

Why I think this matters: “World models” are essentially a bet that today’s LLMs aren’t enough — that the next big leap comes from systems that can simulate cause-and-effect and do better “what-if” prediction, not just text completion. That’s explicitly how TechCrunch describes the pitch (and the hallucination angle). 

If this works, it’s a straight line to better robotics, planning, and “AI that doesn’t confidently invent reality.”

OpenAI adds direct controls for ChatGPT “warmth” + “enthusiasm”

OpenAI is now letting users dial ChatGPT’s warmth and enthusiasm up or down, plus tweak emoji usage (and even formatting behaviors like headers/lists). You’ll find it under Personalization → Add Characteristics (rolling out gradually).

This is partly a response to a real tension: people want friendly and natural… but not so friendly it becomes “sycophant-y.” TechCrunch notes OpenAI previously rolled back an update for that exact vibe problem, and also points out critics who argue overly affirming chatbots can be a dark pattern.

My take: this is a quiet “product maturity” milestone. Personality is no longer a side effect — it’s becoming a user-controlled setting. That’s going to matter a lot for work use (professional tone) and mental ergonomics (less hype, less emotional mirroring).

Robots as backup dancers (yes, really)

A fun one — and also weirdly telling. Videos from a Wang Leehom concert in Chengdu show robotic dancers performing in sync on stage during “Open Fire.” The artist’s site even frames it as a deliberate “tech + live music” moment.

Futurism reports the robots were Unitree G1 humanoids, and highlights how convincing the choreography looked (including flips in some clips).

Why it’s not just a gimmick: entertainment is basically a stress-test for coordination, timing, and reliability in chaotic real-world settings. When robots can hit marks on a live stage, you’re staring at a capability curve that eventually spills into everything else.

🧠RESEARCH

DEER: Draft with Diffusion, Verify with Autoregressive Models

This paper introduces DEER, a system that makes AI write faster by using a "diffusion" model—which refines random noise into text—to draft sentences, while a standard model checks them. This avoids the confusion that slows down other fast-writing methods, speeding up text generation significantly.

Can LLMs Guide Their Own Exploration? Gradient-Guided Reinforcement Learning for LLM Reasoning

Researchers found a way to help AI learn to reason better by using "gradients"—mathematical signals that show the direction of improvement. Instead of guessing randomly during training, the AI uses these signals to explore smarter problem-solving paths, essentially teaching itself to think more effectively.

End-to-End Training for Autoregressive Video Diffusion via Self-Resampling

This study presents a method to train video-making AI from scratch without needing a separate "teacher" model. By simulating errors during training, the system learns to fix its own mistakes. This results in longer, smoother videos where events follow a logical timeline.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

AutoApplyAI – Let AI Apply to Jobs For You - Wonsulting’s automated job‑application tool.

AutoDraw – AI Drawing Assistant - free, web‑based drawing tool from Google Creative Lab that uses machine learning to guess what you’re doodling and instantly swap your scribbles for clean, artist‑made icons.

AutoGPT: AI Framework for Autonomous Task Execution - pen-source platform for creating, deploying, and managing continuous AI agents—software workers that can run on their own and automate multi‑step workflows.

📲SOCIAL MEDIA

🗞️MORE NEWS

Qwen-Image-Layered This new AI model can break down any image into separate, transparent layers for easy editing. Users can move, resize, or recolor individual objects without messing up the rest of the picture. It bridges the gap between static images and editable design files.

OpenAI Internal Financials Internal documents reveal that OpenAI is becoming much more efficient at turning computing costs into profit. While their total server bill remains huge, the company is squeezing more revenue out of every dollar spent on running ChatGPT. This improved "compute margin" is a positive sign for their long-term business model.

China’s Light-Based AI Chips Chinese researchers have reportedly developed a new photonic chip that uses light instead of electricity to process data. This "Taichi" chip is claimed to be 100 times faster and more energy-efficient than Nvidia’s top GPUs for certain tasks. If scalable, this technology could bypass traditional restrictions on semiconductor power.

OpenAI’s ChatGPT Problem Internal struggles and organizational messiness are reportedly slowing down ChatGPT’s development. Delays in releasing new features like AI agents and video generation have allowed competitors to catch up. The company is fighting to regain its momentum amidst these growing internal hurdles.