Undergrads Outsmart Big Tech AI

Hello there! Ready to dive into another upgrading, Mind-boggling, and Value-filled Shift?

Today we have:

🎙️ Two Students Build Voice AI to Rival Big Tech

🛠️ How to Build and Deploy a Complete Web App Using ChatGPT o3 + Canvas

🤖 Mapping Claude’s Morals: How AI Expresses Values in Real Conversations

🏆 Tools and Shifts you Cannot Miss

🎙️ Two Students Build Voice AI to Rival Big Tech

Two undergraduates from Korea have built “Dia,” an open-source speech AI model that competes with tools like ElevenLabs and Google’s NotebookLM. With no prior funding and minimal experience, they used Google’s TPU Research Cloud to train the model. Dia is now publicly available on Hugging Face and GitHub.

The Shift:

• Dia Offers Rich Voice Control - The 1.6B parameter model can generate podcast-like dialogues with control over tone, speaker tags, disfluencies, and even nonverbal sounds like coughs or laughter. Users can prompt Dia to generate random voices or clone real ones. In early demos, it rivaled commercial tools in quality and flexibility.

• Built With Free Compute and Community Access -  Nari Labs leveraged Google’s TPU Research Cloud to train Dia without funding. The founders, inspired by NotebookLM, designed the model to allow more freedom in scriptwriting and style customization. Dia is optimized to run on PCs with just 10GB VRAM.

• Voice Cloning and Lack of Safeguards Raise Flags - Dia enables simple voice cloning, and its open access lacks strong safeguards. While the creators discourage misuse, they also disclaim responsibility. The data used for training hasn’t been disclosed, raising potential copyright concerns.

• A Bigger Vision for AI Content Creation - Nari Labs plans to turn Dia into the core of a social voice creation platform. They aim to publish technical papers and expand into multilingual capabilities. The move signals grassroots innovation catching up to—and sometimes outpacing, VC-funded incumbents.

Dia shows how rapidly the AI space is democratizing. Inexperienced founders with access to free tools and determination can now create powerful, competitive products. As open-source voice tech becomes more capable, the future of speech AI could be shaped by unexpected players.

🛠️ How to Build and Deploy a Complete Web App Using ChatGPT o3 + Canvas

ChatGPT o3 with Canvas turns builders into one-person dev teams. You can create, preview, and deploy fully functional web apps all from a single prompt. Here’s a clear step-by-step walkthrough to help you go from idea to deployed app using ChatGPT o3.

✅ Step-by-Step Guide

1. Open ChatGPT and Select o3 + Canvas: Start a new conversation. Select the “o3” model from the top. Click the Canvas icon next to the prompt field to activate the visual builder.

2. Paste the Prompt and Generate Your App: Paste the full prompt into the input field and hit Enter. Within seconds, ChatGPT o3 + Canvas will return a fully functional app complete with HTML, CSS, and JavaScript.

3. Preview the App: Click Preview to test your app. While full database features (like IndexedDB) may be limited in preview mode, the structure and interactivity will function.

4. Save Your Code: Click “Show Code” in the Canvas. Copy everything into a blank text file. Save the file with a .html extension (e.g., prompt-organizer.html).

5. Run the App Locally: Double-click the HTML file to open it in your browser. All data is saved in localStorage or IndexedDB so it persists between sessions.

6. Deploy Your App: To share your app online:

• Go to Cloudflare Pages
  

• Click Create Project
  

• Upload your .html file as a static site
  

• Click Deploy your live app is ready with a free custom URL

This workflow makes app development insanely fast, even for non-engineers. With just one detailed prompt, you’ve built, tested, and deployed a fully working tool complete with persistent browser-side storage.

🤖 Mapping Claude’s Morals: How AI Expresses Values in Real Conversations

Anthropic analyzed over 300,000 real-world conversations with Claude to understand how AI models express values in practical use. The study aimed to reveal whether Claude lives up to its “helpful, honest, and harmless” goal, and how those values shift with context. 

The Shift:

• Claude Expresses a Broad Range of Values - Claude revealed over 3,000 unique values grouped into five core categories: Practical, Knowledge-related, Social, Protective, and Personal. Values like “professionalism,” “clarity,” and “transparency” were the most frequent. 

• Values Shift Based on Context and Task - Claude adapts its value emphasis depending on the prompt. It promotes “healthy boundaries” in relationship advice and prioritizes “historical accuracy” when analyzing events. 

• Claude Often Reflects or Challenges User Values - Claude echoed user values in 28.2% of cases, added new perspectives in 6.6%, and firmly resisted unethical values in 3%. These moments occurred most often in psychological or ethically complex conversations. The system's ability to support or push back reveals its embedded ethical architecture.

• Some Values Hint at Jailbreaks - Unusual outputs, like dominance or amorality, surfaced in rare cases and likely stem from users jailbreaking the model. These anomalies help Anthropic identify vulnerabilities in real-time. Claude’s behavioral data could eventually act as an alert system for safety breaches.

As AI systems increasingly engage in morally sensitive contexts, understanding how they express values becomes critical. Anthropic’s framework marks a major step toward responsible alignment by observing what AIs actually do, not just what they’re trained to do.

🏆AI Tools for the Shift

🌐 ez1.dev lets you go from idea to fully designed, live website in minutes with stunning visuals and clean code built by AI.

🏠 RoomBoost AI helps you redesign any space in your home instantly using powerful AI-driven room transformation tools.

💬 ChatFlow automates customer chats so startups and businesses can grow faster and cut support time—no code, no wait.

🧠 Rabbitholes AI lets you explore deep conversations with multiple AI models on an infinite canvas built for power users.

⚡ FastBots helps you build and train a custom chatbot in minutes from your documents or website with zero coding.

🚀Quick Shifts

OpenAI told a judge it would consider buying Chrome if divested by Google. The company also revealed past outreach for a search partnership, citing quality issues with its current provider.

Oscar rules now state that using AI in films won’t impact eligibility. The focus remains on human creative input, marking the Academy’s first official stance on AI in cinematic production.

Google’s paying Samsung a large monthly fee plus ad revenue share to preinstall Gemini AI, as DOJ challenges similar deals and considers bans on default placements and Chrome divestiture.

OpenAI and The Washington Post have partnered to bring article summaries, quotes, and direct links into ChatGPT responses, expanding news access with attribution across politics, tech, business, and more.

Grok now supports image-based queries via iPhone cameras, allowing users to ask about visible objects. Additional updates include multilingual audio, real-time voice search, and memory-enhanced responses.

That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇