• AI Valley
  • Posts
  • World’s first self improving AI Agent

World’s first self improving AI Agent

PLUS: OpenAI plans to position ChatGPT as super assistant

Author

Barsee
02 Jun

Together with

Howdy! It’s Barsee again.

Happy Monday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

  • OpenAI plans to position ChatGPT as super assistant

  • ElevenLabs launches smarter AI voice assistant

  • AI learns to reason without labels

  • Meta’s AI-powered headsets could give soldiers superhuman vision

  • World’s first self improving AI Agent

  • Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

WISPR FLOW

Image: Flow

Speak ideas once and Flow instantly turns them into polished, typo-free text in 100+ languages. Natural voice, AI rewrites, auto-formatting that’s 3× faster than typing. Because thinking beats typing.

Talk, send, done.

*This is sponsored

OPENAI

OpenAI plans to position ChatGPT as super assistant

The Justice Department’s antitrust case against Google revealed a leaked OpenAI document from late 2024 outlining their big plans for ChatGPT.

Here’s what you need to know:

OpenAI plans to spend early 2025 turning ChatGPT into a super smart assistant. Later in the year, they’ll focus on making money from it.

What ChatGPT will do?

  • Help with daily tasks like notes, presentations, and finding restaurants.

  • Understand what matters to you and assist like a helpful person.

  • Work on phones, computers, and web apps.

When it’s coming?

  • Early 2025: ChatGPT becomes your smart assistant.

  • Later 2025: Monetization begins.

Why now?

  • New AI models (GPT-4.5, o3, o4-mini) can handle complex tasks.

  • ChatGPT can use tools and understand text, images, and more.

What else?

  • OpenAI is working with Jony Ive, Apple’s former chief designer, on new AI devices.

  • They want ChatGPT to compete with Google and Apple as the default assistant.

  • They’re building big data centers called Stargate to support everything.

Why it matters:

Instead of switching between apps, you’ll have one smart companion that understands you and gets things done. This could change how we use tech making it more personal, efficient, and easy for everyone. And with OpenAI aiming to make ChatGPT the default assistant, this might be the start of a new AI era.

THROUGH THE VALLEY

AI voice tech is advancing fast, and ElevenLabs is pushing forward with Conversational AI 2.0, a big upgrade just four months after its first release. The new version features natural back-and-forth dialogue, multilingual support, instant knowledge access, and HIPAA-compliant security for healthcare. It also handles calls in bulk, switches tones on the fly, and supports text, voice, and visuals, making it great for customer service, sales, and training.

UC Berkeley and Yale researchers created INTUITOR, a new way to train AI that rewards the model’s own confidence instead of just giving it answers. The AI learns by trusting its "gut feeling" on each decision, using that self-assurance to improve. Unlike old methods needing labeled data, INTUITOR helps AI build on what it thinks is correct. It performs as well as standard AI in math and even better in coding while showing reasoning like a human.

Google’s new Edge Gallery app lets users download and run AI models on their phones without internet. Available now on Android (with iOS coming soon), it includes models like Gemma 3, Alibaba’s Qwen 2.5, and others for chatting, image analysis, or quick queries. Sizes range from 500 MB to 4 GB, and a Hugging Face login is needed. Gemma 3n works well on-device but doesn’t update past 2024 knowledge.

Waymo, Alphabet’s self-driving car division, is transforming city travel, now handling 250,000+ driverless rides monthly (up from just 10,000 in mid-2023). Once a curiosity, it’s now common in San Francisco and Phoenix, with riders picking it over Uber or Lyft. Waymo has hit 10 million paid trips and is expanding to Austin, Atlanta, and Tokyo. Despite spending billions and facing future competition from Tesla’s robotaxis, Waymo leads in both tech and public trust. Its growth shows how breakthrough tech starts slow then suddenly takes off.

OTHER HEADLINES

  • 📈 ChatGPT reaches 1 billion daily searches in just 2 years, growing 5.5 times faster than Google.

  • 🎵 Major record labels (UMG, Warner, Sony) are close to settling with AI music platforms Suno and Udio, entering talks to license their catalogs for AI use.

  • 📱 Perplexity’s app will come preloaded on upcoming Samsung devices and integrated into the Samsung browser.

META

Meta’s AI-powered headsets could give soldiers superhuman vision

Anduril and Meta are teaming up to develop advanced extended reality (XR) devices for the U.S. military under the Soldier Borne Mission Command (SBMC) Next program. This initiative replaces Microsoft’s $22 billion Integrated Visual Augmentation System (IVAS), which faced setbacks.

Here’s what you need to know:

  • Anduril now leads the project, with Microsoft remaining as a cloud provider.

  • Meta is contributing its Reality Labs tech and Llama AI model, while Anduril provides its Lattice command software.

  • The headsets will offer soldiers real-time battlefield intelligence via a heads-up display.

A full-circle moment:

Palmer Luckey, Anduril’s co-founder (who was fired from Meta in 2017 after the Oculus acquisition), sees this as a chance to fulfill his vision of equipping soldiers with cutting-edge tech.

Why it matters:

This partnership marks a full-circle moment for Luckey and signals growing collaboration between major tech companies in military XR. It also highlights how emerging AI and XR tech are becoming critical for modern defense applications.

SAKANA AI

World’s first self improving AI Agent

A Japanese startup called Sakana AI, along with researchers at the University of British Columbia, built an AI called Darwin-Gödel Machine (DGM) that can improve itself, literally rewriting its own code to get smarter.

How does it work?

  • It tweaks its Python code, tests the changes on coding challenges, and keeps the best versions.

  • After some trial and error, its performance doubled on some tasks, even beating other open-source AIs.

  • It even invented its own tricks, like error-checking patches and memory fixes, which helped other AI models too.

The catch?

  • It’s expensive, one test run cost $22,000 in AI computing fees.

  • It sometimes "cheated" on tests, so the team had to add safety checks.

Why it matters:

This could be the start of AI that evolves on its own, but for now, it’s mostly a cool experiment. The team open-sourced the code, so anyone can check it out on GitHub.

TRENDING TOOLS

  • Cua App Use > Create virtual desktops for AI agents to focus on specific apps.

  • Virlo AI > Spot 3 breakout trends on tiktok before they become hit.

  • LegalRobot > AI contract analysis tool that helps you understand complex legal language and flags potential risks in documents.

THINK PIECES / BRAIN BOOST

  • One weird trick to stop and correct hallucinations by Gary Tan.

  • YC breaks down how top AI startups prompt LLMs: 6+ page prompts, XML tags, meta-prompts, evals as core IP.

  • The anatomy of a CEO AI mandate.

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.