Two undergrads construct elite speech AI

Good morning, AI enthusiasts. Two Korean undergrads with little AI experience took Sam Altman’s “you’ll be able to just construct things” quote to heart — constructing an open-source speech model that outperforms industry leaders — with zero funding.

With AI tools and resources becoming more accessible globally, the era of solo builders competing with billion-dollar labs has officially arrived.

P.S. — Our next workshop is today at 1 pm, where you’ll learn how one can turn raw ideas into polished, on-brand presentations with Gamma. RSVP here.

In today’s AI rundown:

• Two undergrads unveil SOTA speech AI

• The Washington Post joins OpenAI’s alliance

• Automate your sales with personalized emails

• Anthropic CISO: AI employees are coming

• 4 recent AI tools & 4 job opportunities

LATEST DEVELOPMENTS

NARI LABS

🗣️ Two undergrads unveil SOTA speech AI

The Rundown: Korean startup Nari Labs released Dia, an open-source text-to-speech model that claims to exceed the capabilities of leading industrial offerings like ElevenLabs and Sesame — developed by two undergraduate techies with zero funding.

The small print:

• The 1.6B parameter model supports advanced features like emotional tones, multiple speaker tags, and nonverbal cues like laughter, coughing, and screams.

• The work was inspired by Google’s NotebookLM, with Nari also using Google’s TPU Research Cloud program for compute access.

• Side‑by‑side tests show Dia outshining ElevenLabs Studio and Sesame CSM‑1B in timing, expressiveness, and handling nonverbal scripts.

• Nari Labs founder Toby Kim said the startup plans to develop a consumer app focused on social content creation and remixing based on the model.

Why it matters: Dia is a living testament to Sam Altman’s ‘you’ll be able to just do things’ tweet, with two inexperienced undergrads training an open-source model that competes with the highest voice tech in the marketplace. There’s never been a greater time to attempt to construct something, with AI unlocking recent access to learning like never before.

TOGETHER WITH GAMMA

🚀 Create stunning content in minutes, not hours

The Rundown: For consultants, educators, marketers, and sales pros drowning in content creation, Gamma’s all-in-one platform is the lifeline you would like — helping you create beautiful presentations, web sites, and social media content with minimal effort.

Gamma lets you:

• Generate content from easy text prompts or import and improve existing work

• Edit images with a single click — change subjects, styles, or remove backgrounds

• Easily export content to Google Slides/PDFs/PowerPoint

• Experience an upgraded AI and UI for a strong content creation experience

Click here to experience Gamma and join 50M other users creating 700k presentations a day.

OPENAI

📰 The Washington Post joins OpenAI’s alliance

The Rundown: The Washington Post just announced a brand new partnership with OpenAI, allowing the AI leader to bring summaries and links from its reporting directly into ChatGPT answers.

The small print:

• ChatGPT will now feature summaries, quotes, and direct links to relevant Washington Post articles in its responses to user questions.

• The deal adds the Jeff Bezos-owned Post to OpenAI’s expanding roster of media partners, with over 20 major news publishers.

• It also comes amid ongoing legal battles between OpenAI and other major publishers, including the NYT, over training data and copyright issues.

• The Washington Post has been actively experimenting with AI, launching tools like Ask The Post AI and Climate Answers over the past 12 months.

Why it matters: One other major news outlet is selecting partnership over litigation, with a bet that visibility via ChatGPT might be key to reaching audiences worldwide. OpenAI also gains more trustworthy content for its AI, and an enormous like WaPo joining on further isolates publishers just like the NYT, who’re fighting back in court.

AI TRAINING

📊 Automate your sales with personalized emails

The Rundown: On this tutorial, you’ll learn how one can use n8n to show your static contact lists into dynamic sales outreach machines by routinely sending personalized emails to prospects based on their company, role, and interests.

Step-by-step:

1. Create a brand new n8n workflow and arrange a Google Sheets trigger that monitors when recent leads are added to your spreadsheet.

2. Add an AI Agent node and connect it to a language model to process your contact information.

3. Configure a Gmail node to create drafts of personalized emails as an alternative of sending them directly.

4. Write detailed instructions within the AI Agent’s system message telling it exactly how one can craft sales emails.

Pro tip: Fantastic-tune your AI’s writing style by testing different system message instructions. We did an in depth workshop showing how one can create your personal AI Agent to automate tasks and run local AI models with n8n here.

PRESENTED BY AXOLOTL

⚙️ Post‑train LLMs without limits

The Rundown: Axolotl v0.8.0 is an open‑source toolkit built by devs, for devs — providing you with full transparency and the liberty to customize large language models through easy configuration.

With Axolotl, you’ll be able to:

• Post‑train LLaMA, Gemma, Mistral, and other leading models from one workflow

• Scale effortlessly — from a single local GPU to multi‑node cloud clusters

• Start fast because of a simple, plug‑and‑play setup

Dive into Axolotl v0.8.0 to start out post‑training today.

ANTHROPIC

💼 Anthropic CISO: AI employees are coming

The Rundown: Anthropic’s Chief Information Security Officer, Jason Clinton, just predicted that AI-powered virtual employees will begin operating on corporate networks inside the following 12 months, bringing major recent challenges in security management.

The small print:

• These AI employees would have their very own corporate accounts, passwords, and “memories,” a major step up from current task-specific AI agents.

• Clinton said security challenges will include managing AI account privileges, monitoring access, and determining responsibility for autonomous actions.

• He sees virtual employees as the following “AI innovation hotbed,” with virtual worker security also emerging as an area of focus alongside it.

• Anthropic said it’s focused on securing its own AI models against attacks and watching out for potential areas of misuse.

Why it matters: Work is about to alter completely within the AI age, and so will the safety measures needed for this brand-new kind of worker. The query is that if cybersecurity practices will update fast enough to maintain up — or if it’s going to take a serious breach or exploit as a wake-up call for brand new threats that autonomous AI employees present.

QUICK HITS

🛠️ Trending AI Tools

• 🗣️ Agent-to-Agent Transfers – Hand off conversations between AI agents

• 📽️ AI Co-Editor – Descript’s recent agentic video editor

• ⚡️ Genspark AI Slides – Agentic tool for quickly creating presentation slides

• 🎥 Edits – Meta’s recent video creation app with AI features

💼 AI Job Opportunities

• 🔐 Runway – Member of Technical Staff

• 🎧 Soundhound AI – Customer Success Representative

• 📈 Abridge – Director, GTM Strategy + Pricing

• 🗣️ Captions – Community Manager

📰 All the things else in AI today

OpenAI’s head of product, Nick Turley, testified in Google’s antitrust trial that the AI leader can be fascinated by buying its Google Chrome browser if a sale were forced.

Apple removed “available now” claims from its Apple Intelligence marketing page following the National Promoting Division’s concerns about misleading availability.

Character AI launched AvatarFX, an AI platform that enables users to create long-form, coherent talking avatars from a single reference photo and voice selection.

IBM and the European Space Agency released TerraMind, an open-source AI system that uses nine data modalities and satellites for real-time climate monitoring.

Cohere CEO Aidan Gomez joined the board of electrical automaker Rivian, aiming to integrate AI tech more broadly into the corporate’s products and manufacturing.

Motorola debuted SVX, a brand new AI-powered device that mixes a body camera, speakers, and an AI assistant to scale back emergency response times.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop today at 1 PM EST with Mel, Creative Director at Gamma. By the tip of the workshop, you’ll learn how one can turn raw ideas into polished, on-brand presentations using Gamma’s powerful AI storytelling tools — no design skills needed.

RSVP here. Not a member? Join The Rundown University on a 14-day free trial.

🤝 Share The Rundown, get rewards

We’ll at all times keep this article 100% free. To support our work, consider sharing The Rundown with your pals, and we’ll send you more free goodies.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team