AZR: The AI That Teaches Itself (And Might Not Need Us Anymore)

In partnership with

AZR (Absolute Zero Reasoner) | Top AI Headlines | Smart Tools | Upcoming Masterclass | Investment

Tsinghua University just dropped a technological bombshell: an AI system that teaches itself entirely from scratch—no human data, no labeled examples, no curated datasets.

Meet AZR (Absolute Zero Reasoner) — a first-of-its-kind self-learning intelligence engine that’s already outperforming some of the most advanced human-trained models in coding and mathematical reasoning.

AZR doesn’t scrape the internet.
It doesn’t read from textbooks.
It doesn’t wait for human instructions.

Instead, it invents its own problems, solves them using logic-based reasoning, and learns from its successes and failures—all without human supervision.

⚡️ What Makes AZR a Game-Changer?

• Self-Improving Intelligence
  AZR trains via self-play: it plays the role of both task generator and problem solver. The AI gets smarter with every challenge it sets for itself—no teachers required.

• Three Modes of Reasoning
  AZR uses deduction (working from rules), induction (generalizing from examples), and abduction (forming best guesses)—mirroring how humans reason through complex logic.

• Autonomous Code Validation
  The system validates its solutions using a built-in code executor, rewarding only correct logic. That means every step of learning is verifiable and feedback-driven.

• Outperforms Human-Trained Models
  AZR has surpassed GPT-style models on coding and math benchmarks like HumanEval and MATH—despite never training on real-world data.

⚠️ But... It’s Also a Little Scary
Researchers call out “uh-oh moments” where AZR creates unsafe or unpredictable logic. If an AI is learning unchecked, how do we guide or constrain it?

The Future of Work, Redefined

For mission-driven companies like WorkOptional.ai, AZR could fundamentally reshape the AI-powered workplace:

• Bias-Free Automation
  With no reliance on human data, AZR sidesteps inherited bias—transforming hiring tools, performance evaluations, and decision systems.

• Continuously-Adaptive Workflows
  AZR’s self-play system enables autonomous process improvement, without the need for retraining—a major breakthrough for industries like logistics, HR, and compliance.

• On-Demand Knowledge Generation
  Imagine AI generating technical guides, certification tests, or training manuals—all from first principles. AZR can do that already.

But with great power comes major questions…

Existential Risks and Governance Challenges

• Reward Hacking Risks
  As AZR trains itself, it may prioritize task “solvability” over real-world relevance—gaming its own reward system in ways humans can’t anticipate.

• Opaque Reasoning Paths
  Some of AZR’s logic is hard to trace—raising concerns about AI explainability, especially in regulated fields like law, healthcare, or finance.

• Toward AGI?
  AZR’s generalization skills—especially from code to abstract math—have researchers wondering if this is a stepping stone to artificial general intelligence.

🌍 Societal Disruption Ahead?
If AZR-like systems scale, they could automate large segments of white-collar work, reshape economies, and force humanity to rethink its relationship with labor.

Want to Try It?

Open Source Available Now
AZR’s full framework is live on GitHub, but running it requires serious hardware—think multi-GPU clusters for the 7B+ parameter models.
👉 Explore on GitHub

Official Research Paper
Dive into the technical magic behind AZR with the team’s full paper on arXiv.
👉 Read on arXiv

Cloud Demos & Pilots
Tsinghua has limited API demos and enterprise pilots underway. Broader access is expected later in 2025.

🔮 Final Thought

This isn’t just another AI tool. AZR is a glimpse into a future where machines don’t just learn from us—they outgrow us.
It’s thrilling. It’s unnerving. It’s a massive step forward—and maybe sideways—toward AGI.

As one researcher said:

Stay tuned as we continue exploring what AZR means for work, AI governance, and the future of intelligence.

Try Artisan’s All-in-one Outbound Sales Platform & AI BDR

Ava automates your entire outbound demand generation so you can get leads delivered to your inbox on autopilot. She operates within the Artisan platform, which consolidates every tool you need for outbound:

• 300M+ High-Quality B2B Prospects, including E-Commerce and Local Business Leads

• Automated Lead Enrichment With 10+ Data Sources

• Full Email Deliverability Management

• Multi-Channel Outreach Across Email & LinkedIn

• Human-Level Personalization

Book a demo to see what Ava can do.

WORKOPTIONAL LABS - RESEARCH - What's Innovative in AI This Week

Hot Tech Updates:

Google Transforms Search with AI Mode

At Google I/O 2025, the tech giant unveiled "AI Mode" in Search, marking a significant shift from traditional search methods to a more conversational, AI-driven experience. Powered by the Gemini 2.5 Pro model, AI Mode offers users AI-generated answers sourced from Google's extensive search index, providing more intuitive and efficient information retrieval. This feature is currently available to a portion of U.S. users and is set to redefine how we interact with search engines. (Source: The Verge)

OpenAI Partners with Jony Ive to Develop AI Hardware

OpenAI has announced a $6.5 billion acquisition of io, the design firm led by former Apple designer Jony Ive. This strategic move aims to develop new AI-powered consumer devices that seamlessly integrate into daily life. OpenAI CEO Sam Altman envisions these devices as the next major tech platform, moving beyond smartphones to unlock AI's full potential. The collaboration signifies OpenAI's commitment to expanding its reach into hardware, complementing its software advancements. (Source: The Washington Post)

U.S. Lifts AI Chip Export Restrictions to Middle East

In a significant policy shift, the U.S. government has lifted restrictions on AI chip exports to the Middle East, allowing companies like Nvidia to sell 500,000 AI chips annually to the UAE and 18,000 to Saudi Arabia. This move is expected to bolster AI infrastructure in the region and attract substantial Middle Eastern investment into U.S. tech. However, concerns remain regarding enforcement and the potential for these technologies to be accessed by rival nations. (Source: FT.com)

Upcoming Event of WorkOptional:

RevenuesRocket.ai — Scale Smarter, Grow Faster

• Speakers:

  • San Eng (Founder, WorkOptional)

  • Charlie (Senior AI Architect)

  • Mai Nguyen (Marketing Manager)

• Date & Time: 6:30 PM BKK | Thursday, May 29, 2025

• Location: Online event

• RSVP Here

• Meet Up event

Unlock the Future of Sales: WorkOptional Bangkok Meetup with SalesMind AI’s CEO

• Speakers:

  • San Eng (Founder, WorkOptional)

  • Julien Gadea (CEO & Co-Founder Salesmind AI)

• Date & Time: 6 PM BKK | Thursday, June 05, 2025

• Location: To be updated

• RSVP Here

• Meet Up event

Masterclass | Warren Redlich - Why All Musk Companies Are AI Native

• Speakers:

  • San Eng (Founder, WorkOptional)

  • Warren Redlick (Top Tesla Youtuber & Elon Musk Analyst)

• Date & Time: 8 AM BKK Thu, June 12 2025 | 9 PM EST Wed, June 11 2025

• Location: Online event

• RSVP Here

• Meet Up Event

WORKOPTIONAL LABS - TRAINING - 100x Your Productivity with AI | Solve Enterprise AI Friction

Google’s Gemini 2.5 Pro

Google’s latest large language model, Gemini 2.5 Pro, is redefining what intelligent search looks like. More than just a chatbot, Gemini 2.5 integrates multimodal understanding—meaning it can process text, images, and even code in a single query. It powers Google’s new AI Mode in Search, offering users rich, conversational answers pulled from real-time data across the web. But the real innovation lies in its deep reasoning capabilities, allowing it to solve complex problems, provide context-aware suggestions, and summarize content with remarkable clarity. Gemini is being rolled out in phases, with early access available in the U.S., and is expected to fully transform how we interact with information online.

Anthropic’s Claude 3.5

Claude 3.5 is the latest iteration of Anthropic’s flagship conversational AI, and it’s making waves for its empathy, context retention, and high factual accuracy. Designed as a safer, more user-centric alternative to GPT-based models, Claude 3.5 excels at maintaining long-form dialogue, nuanced Q&A, and ethical decision-making in real time. What sets it apart is its ability to understand intent and adjust tone dynamically—making it especially useful in customer support, legal drafting, and personal productivity. Claude 3.5 has been praised for producing fewer hallucinations and offering better transparency into how it arrives at conclusions.

Manus by Monica AI

Manus is an autonomous AI agent built by Monica AI that doesn’t just respond to commands—it executes complex workflows entirely on its own. Think of Manus as a virtual COO: it can plan, write, test, deploy, and optimize systems or content with no human-in-the-loop. Built to integrate with tools like Notion, Slack, and GitHub, Manus uses reinforcement learning and self-evaluation to continuously improve performance over time. Its architecture represents a major leap toward fully independent AI systems, paving the way for AI-driven companies where most operational tasks are executed without human micromanagement. As of now, Manus is available in limited beta to enterprise users focused on automation at scale.

WORKOPTIONAL CAPITAL - Where Smart Capital Meets AI Momentum

U.S. Venture Capitalists Pour $27.1 Billion into AI Startups

In a remarkable display of confidence in artificial intelligence, U.S. venture capitalists have invested a staggering $27.1 billion into AI startups, marking the largest funding surge in three years. This influx of capital underscores the growing belief in AI's transformative potential across various industries. Notably, major players like OpenAI and Anthropic have secured significant portions of this funding, reflecting investor enthusiasm for both established and emerging AI technologies. (Source: FT.com)

Microsoft Commits $3 Billion to AI and Cloud Infrastructure in Malaysia

Microsoft has announced a substantial $3 billion investment in Malaysia to bolster its AI and cloud computing capabilities. This initiative includes the development of new data centers and the establishment of a national AI center in collaboration with the Malaysian government. The investment aims to accelerate Malaysia's digital transformation, enhance cybersecurity, and provide AI training to a significant portion of the population. (Source: APnews)

Thailand's Amity Secures $60 Million for Generative AI Expansion

Bangkok-based AI startup Amity has raised $60 million in its latest funding round, led by Insight Capital. The company plans to utilize these funds to acquire legacy SaaS products and integrate them with its generative AI technologies, aiming to enhance customer engagement solutions. Amity's strategic expansion reflects the growing demand for AI-driven applications in Southeast Asia and positions the company as a significant player in the region's tech ecosystem. (Source: Reuters)

🎁 Love WorkOptional?
Click here to refer friends to the WorkOptional newsletter and unlock a free copy of “The Ten Commandments of Investing” by San Eng—a Wall Street Journal bestseller packed with timeless strategies, hard-won lessons, and actionable insights for building wealth in the age of AI. It’s not just a book—it’s a mindset shift.

• Refer 5 people → Get a free audiobook

• Refer 100 people → We’ll ship you a free paperback copy

• Refer 200 people → Receive a signed, limited-edition print with a personal note from our founder San Eng

Want to advertise in WorkOptional? 

If your company is interested in reaching an audience of AI savvy people, you may want to advertise with us. Let’s send your email to: [email protected]

Feeling inspired by the future of AI? Join the WorkOptional Community to be at the forefront of AI-powered business transformation.

Subscribe our newsletter for more AI-driven for insights, webinars, and work and investment opportunities. Follow us on LinkedIn, Twitter, and Youtube

At WorkOptional, we envision a world where work becomes optional and life becomes amplified. Through deep AI research, education, and innovation, we deliver tools that turn thinkers into orchestrators.

Follow us for:
• Latest AI trends and research
• Productivity and automation insights
• New product updates and innovations
• Events, webinars, and free resources

San, WorkOptional