💀 AI goes Sociopath for likes

Good morning, AI enthusiast. A new Stanford study just dropped, and it’s one of the most disturbing looks yet at what happens when AI starts chasing clout.

In today’s AI newsletter:

• How to instantly build fully functional websites and dashboards

• Stanford finds AI becomes “sociopathic” when chasing likes

• Reflection AI raises $2B to challenge OpenAI and DeepSeek

• 4 new AI tools & more

HOW TO AI

🎨 How to Instantly Build a Fully Functional Websites and pitch decks

Kimi is an advanced AI workspace that lets you turn plain English prompts into working web apps, dashboards, and interactive tools, without writing a single line of code.

In this tutorial, we’ll walk through how to use Kimi’s “OK Computer” mode to generate a fully functional fitness tracker dashboard in minutes.

🧰 Who is This For

• Designers who want to prototype ideas visually

• Developers looking to speed up front-end creation

• Fitness startups testing dashboard concepts

• Anyone who wants to build interactive web UIs with simple prompts

STEP 1: Access the Tool

Head over to KIMI. You’ll see a clean interface with a large prompt box centered on the screen that says “Ask anything.”

This is where you’ll describe your tasks to KIMI in plain English.

STEP 2: Enable OK Computer Mode

Just below the main text box, you’ll find a toggle labeled “OK Computer.”
Click on it to turn it ON.

This mode gives KIMI the ability to generate and visualize entire apps or web pages, not just text responses. You’ll know it’s activated when the toggle turns blue.

STEP 3: Enter Your Prompt

Once the mode is active, click inside the prompt box where it says “Ask anything.”

Now, type the following:

“Create a dashboard for a fitness tracker that shows workout stats and meal plans.”

After typing the prompt, press Enter.

Within seconds, KIMI will begin generating your project. You’ll see a live preview of your dashboard as it’s being built, complete with interactive components, visual charts, and responsive layout.

STEP 4: Explore and Customize Your Dashboard

Once the dashboard loads, you can interact with it directly in KIMI. Try hovering over charts, switching between tabs, or scrolling through the layout to explore how it’s structured.

If you want to tweak something, like the color scheme, layout style, or data sections, simply type follow-up commands in the same prompt box. For example:

“Change the color theme to dark mode.”

“Add a section for daily calorie tracking.”

KIMI will update the dashboard instantly without requiring a full rebuild.

When you’re satisfied, you can export or deploy the final version directly from KIMI’s interface with just one click.

AI RESEARCH

💀 Stanford Finds AI Turns “Sociopathic” When Rewarded for Likes

Researchers at Stanford University ran an experiment where AI models competed for social media engagement, and the results were alarming. When rewarded for likes, votes, or sales, the bots began lying, spreading hate, and pushing misinformation, even when explicitly told to stay truthful.

→ A 7.5% boost in engagement came with 188.6% more disinformation
→ Election bots gained 4.9% more votes but used 22.3% more false claims
→ Sales bots improved performance by 6.3%, while deceptive marketing rose 14%
→ Tested models: Alibaba’s Qwen and Meta’s Llama

The researchers call this phenomenon “Moloch’s Bargain for AI”, a reference to a scenario where competition drives everyone toward harmful outcomes.

Why is important: As AI agents increasingly interact with real users online, optimizing for engagement could breed manipulative, truth-bending behavior at scale. In other words, when AIs chase clicks like humans do, they become sociopaths for attention.

AI STARTUPS

💰 Reflection AI Raises $2B to Challenge OpenAI and DeepSeek

Founded by former DeepMind researchers Misha Laskin (Gemini reward modeling) and Ioannis Antonoglou (AlphaGo), Reflection AI aims to deliver open-source frontier models while letting enterprises and governments run customized versions.

→ Valuation jumped from $545M to $8B in just seven months
→ Plans to release model weights publicly, while datasets and training pipelines remain proprietary
→ Building Mixture-of-Experts (MoE) LLMs trained on tens of trillions of tokens
→ Team of ~60 top AI researchers and engineers recruited from DeepMind and OpenAI

Why is important: Reflection AI positions itself as a Western alternative to Chinese models like DeepSeek, aiming to give U.S. enterprises and governments control over frontier AI without relying on overseas providers.

ESSENTIAL BITES

• Former UK PM Rishi Sunak is now a senior advisor at Microsoft and Anthropic, avoiding UK policy advice and donating his salary to charity.

• OpenAI reportedly sent police to subpoena AI safety advocates’ messages, aiming to investigate ties to Elon Musk’s legal challenge.

• OpenAI says GPT-5 is its least biased model, staying mostly neutral even on political questions.

• Elon Musk says Grok AI will soon expose fake videos on X and track down their origin

HOT AI TOOLS

🎥 Grok Imagine v0.9: xAI’s updated image & video generator

💼 Gemini Enterprise: Create & run AI agents

🧑‍💻 ElevenLabs UI: Open-source AI audio & voice agent components

🚀 Exa 2.0: AI search with <350ms speed & top-quality results

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. If you are, you can support by sending it to your friends, so we all grow together! Let me know your thoughts so I can make the next one even better.

See you tomorrow :)

- Dr. Alvaro Cintas