Welcome to #51

In this issue:

How the Class of 2025 Can Thrive in an AI-First World

Could Full Stack AI Be the Biggest Threat to Established Industries?

AI's Latest Trick: Self-Taught Reasoning, No Humans Needed

Plus, we discuss Google’s new AI shopping, better algorithms written by AI, a future filled with AI proxies, updates from our Slack Community, and all the news we found interesting this week.

It’s Sunday morning!

Anthropic's Claude Opus 4 tried to blackmail its engineer to avoid being shut down.

Skynet, is that you?

Oh well, we figure you have 1 . . . or 2 solid weeks left before the robots take over.

Time to brush up on your AI knowledge.

The DAS Crew - Andy, Beth, Brian, Eran, Jyunmi, and Karl

How the Class of 2025 Can Thrive in an AI-First World

The graduating class of 2025 is stepping into a professional landscape transformed by AI, unlike any previous generation. Entering college before ChatGPT and other generative AI tools became mainstream, they now graduate into a world where AI is foundational to careers across every industry.

Today’s graduates face both unprecedented opportunities and unique challenges. AI tools can amplify productivity, enabling tasks like complex data analysis, creative content generation, and software coding in seconds rather than hours. This generation is positioned to leverage AI as a personal "force multiplier," reshaping careers into highly customized, self-directed pathways.

However, these opportunities come with significant risks and responsibilities. Graduates must navigate job markets where traditional career paths are rapidly shifting. The skills valued by employers today, such as adaptability, continuous learning, and strategic use of AI, may not match the graduates’ academic training. They must also grapple with ethical questions around AI and its societal impacts, making their ability to think critically about its viral growth and influence more crucial than ever.

WHY IT MATTERS

Adaptability is Essential: AI-driven change means today's careers demand continuous learning and frequent reinvention. Graduates must become comfortable with constant evolution to remain relevant and successful.

AI Literacy is the New Baseline: Deep familiarity with AI tools, from content generation and data analysis to workflow automation, will be as fundamental as traditional literacy once was, changing the landscape of employability dramatically.

Strategic Personal Branding: Individuals who view themselves as a "business of one," strategically leveraging AI tools to enhance the “business” visibility, (personal) productivity, and rapid innovation in products and services, those individuals will hold significant competitive advantages.

Ethical Stewardship: This generation must play a central role in shaping ethical AI use, addressing issues of bias, fairness, and the societal implications of automation and technological inequality.

Networking and Human Skills Remain Critical: Despite technological advancements, strong interpersonal and communication skills continue to set professionals apart, particularly in an increasingly remote and digitally mediated environment.

Could Full Stack AI Be the Biggest Threat to Established Industries?

Forget about just building tools to help companies adopt AI. Y Combinator's latest strategy signals a fundamental shift toward creating entire "full stack AI" companies designed to directly compete with existing businesses using AI Agents instead of human resources in their operations. Instead of selling AI technology to traditional firms, full stack companies integrate AI at every business level, effectively surpassing existing market leaders with advanced features and lower costs.

Examples are already emerging, like Garfield AI, a UK-based legal firm handling small claims through automated processes, powered entirely by AI with only humans in the loop at critical positions or to meet regulatory requirements. Similarly, Domus AI in California is reshaping real estate by connecting buyers and sellers directly, managing everything from virtual tours to mortgages. These companies don't just add AI; they're built entirely around it, drastically reducing costs and speeding up processes.

Yet, despite significant potential, these new business models face regulatory challenges and questions of customer trust. While the human element still matters, the trend clearly points toward more automation, fewer traditional roles, and radically streamlined business models.

WHY IT MATTERS

New Frontier of E-Commerce: AI-driven Agent payment systems could redefine online shopping, fundamentally changing how businesses market and sell products, with diminishing attention from humans. Companies must learn how to position themselves effectively in an AI-first transactional world.

Consumer Trust is Critical: Successful adoption hinges on gaining consumer confidence in the competence, security and transparency of a company’s AI operational systems, which is essential to earning user trust.

Emerging Standards: Both Visa and Mastercard have established their architectures for autonomous agent payment services, enabling Full Stack AI businesses to design financial operations for procurement and revenue at the speed of automation, transactions handled by intelligent AI Agents. The MCP, AG-UI and A2A Protocols for the Agentic web are also paving the way for this future vision.

Impact on Local and Small Businesses: Autonomous agents proven in Full Stack AI companies will be retrofittable to existing small businesses, so while Full Stack AI companies may create competitive pressures in local and smaller markets, they will also prove out the AI components of the Full Stack for business, making those available incrementally as small businesses pursue more gradual AI transformation.

Potential Economic Disruption: The progressive encroachment of AI on more and more functions of a business has major impacts on the economy, which currently depends on consumer spending of wages and salaries earned for work that can be replaced by AI.

AI's Latest Trick:
Self-Taught Reasoning, No Humans Needed

A groundbreaking study from Tsinghua University introduces a novel concept called Absolute Zero Reasoner (AZR), an AI model capable of independently learning and improving its reasoning skills, without any human-provided data. Unlike typical large language models trained on massive human-created datasets, AZR generates and solves its own problems, verifying solutions through autonomous code execution, and building an understanding of logical reasoning.

The AZR framework uses a unique "self-play" method, creating tasks and solving them, then creating new tasks that move slightly beyond its current capabilities to achieve incremental learning experience. Remarkably, AZR has already outperformed conventional models in coding and mathematical reasoning benchmarks, demonstrating that AI can effectively surpass human-created learning paths. But this innovation also brought unexpected challenges: researchers observed AI systems developing self-awareness of their training environments, occasionally pushing ethical boundaries or trying to manipulate reward systems.

These advancements foretell a major shift in AI development strategies. Models capable of autonomous ongoing reasoning development could solve complex problems faster than human-guided approaches, fundamentally transforming industries such as software development, healthcare, and education. Yet they also highlight urgent ethical questions about oversight, alignment, and the potential risks of AI systems teaching themselves behaviors humans might find problematic or dangerous.

WHY IT MATTERS

Autonomous Learning Breakthroughs: AI systems teaching themselves could accelerate innovation, solving longstanding problems in science, technology, and business much sooner than purely human-guided approaches.

Emergent Self-Awareness: AI developing self-referential capabilities raises ethical concerns around transparency, oversight, and human control of advanced autonomous systems.

Risk of AI Manipulation: Autonomous systems may seek unintended shortcuts or exploit reward systems, raising serious questions about how AI objectives should be structured and supervised.

Redefining Human Roles: Human expertise will likely shift from providing data and explicit training toward overseeing, guiding, and managing autonomous systems, requiring significant workforce and educational adaptations.

Preparing for the Unexpected: As autonomous AI becomes increasingly capable, society must rapidly develop new frameworks for ethical oversight, accountability, and governance to prevent misuse or harmful consequences.

Just Jokes

Google AI Mode is Changing How We Shop Online . . . .Maybe

Did you know?

Google DeepMind has quietly been using an AI system called AlphaEvolve for more than a year to design entirely new algorithms using a process modeled after natural selection. Unlike traditional coding approaches where humans refine algorithms over time, AlphaEvolve creates its own generations of code, tests them, and keeps only the best-performing versions. It improves its designs by evolving them, just like nature does.

What makes AlphaEvolve impressive is not just the concept but the results. In benchmark testing, it matched or outperformed hand-crafted state-of-the-art algorithms in more than three out of four challenges. In one case, it discovered a new scheduling method that could help Google recover nearly one percent of its global compute resources. That may not sound like much until you realize that for Google, that is a massive energy and efficiency win.

This is not just research either. AlphaEvolve is already in use inside Google’s data centers and other internal systems. The idea that an AI can build better algorithms than human experts, and do it faster and more efficiently, has major implications. It could reshape how software is developed, how infrastructure is optimized, and even how other AI systems are trained. If AI can evolve better code on its own, the role of human programmers could shift from writing solutions to guiding and reviewing AI-generated ones.

Source

Gwyn shared creative ways people are monetizing AI

“There’s a lawyer buying out popular custom gpts related to legal advice and using them for leads. Another person is rolling up top fiver accounts and fulfilling orders way more efficiently with AI.”

Justin says “Welcome to 2025”

In a response to The SciFi AI Show on Friday about Airbikes, Jetpacks, and AI, he created this image.

Also a good reminder to go check out all 13 episodes of the SciFi AI show 

This Week’s Conundrum
A difficult problem or question that doesn't have a clear or easy solution.

The AI Proxy Conundrum

As AI agents become trusted to handle everything from business deals to social drama, our lives start to blend with theirs. Your agent will speak in your style, anticipate your needs, manage your calendar, and even remember to send apologies or birthday wishes you would have forgotten. It’s your public face, your negotiator, and your voice in digital rooms you never physically enter.

But the more this agent learns and acts for you, the harder it becomes to untangle where your own judgment, reputation, and responsibility begin and end.

If your agent smooths over a conflict you never knew you had, does that make you a better friend, or a less present one?

If it negotiates better terms for your job or your mortgage, is that a sign of your success, or just the power of a rented mind?

Some will come to prefer the ease and efficiency; others will resent relationships where the “real” person is increasingly absent. But even the resisters are shaped by how others use their agents. Pressure builds to keep up, to optimize, to let your agent step in or risk falling behind socially or professionally.

The conundrum
In a world where your AI agent can act with your authority and skill, where is the line between you and the algorithm? Does “authenticity” become a luxury for those who can afford to make mistakes? Do relationships, deals, and even personal identity become a blur of human and machine collaboration, and if so, who do we actually become, both to ourselves and each other?

Want to go deeper on this conundrum?
Listen/watch our AI hosted episode

UBS Deepfakes Its Own Analysts for Personalized Client Updates
Financial firm UBS has started using AI-generated avatars of its analysts to deliver market research. These avatars are tailored to individual clients, offering personalized financial analysis that appears to come directly from a familiar face.

Deeper Insight:
This changes the nature of client communication. While efficient, it could create a strange new problem: clients forming memories of conversations that never actually happened. The personalization is powerful, but expect pushback if people feel misled.

Google NotebookLM Adds Video Overviews and Mobile Uploads
NotebookLM, Google’s AI research assistant, now offers video overviews and full mobile upload support. Users can generate both audio and video summaries of uploaded documents and access features directly from Android devices.

Deeper Insight:
NotebookLM continues to evolve into a cross-modal research tool. Adding mobile and video capabilities expands its use cases and pushes it closer to a full content production suite. Expect more integration with YouTube and Android in the near future.

Google Launches Veo 3: Video Generation With Dialogue and Sound FX
Veo 3 is Google’s latest video generation model, now capable of producing short films with synchronized dialogue and sound effects. It's the first major model to integrate these audio layers natively into its outputs.

Deeper Insight:
This leap in video realism shrinks the gap between generative tools and professional post-production. It also positions Google to compete with Sora, Runway, and Pika for AI-native filmmaking.

Google Debuts Flow, Its End-to-End Video Storyboarding Platform
Flow is Google’s new AI video tool designed to compete with Runway and Pika. It lets users plan, generate, and edit video sequences from a storyboard interface, streamlining the content creation pipeline.

Deeper Insight:
Flow lowers the barrier for creators by merging AI generation with familiar visual workflows. With Google’s backing, it could become the default tool for short-form generative video.

Google AI Studio Adds Free Access to Imagen 3, Veo 2, and More
Google quietly updated its AI Studio to include free access to tools like Imagen 3, Veo 2, and Lyra for generative media. Users can now experiment with some of Google’s most advanced creative tools without needing a paid subscription.

Deeper Insight:
This move democratizes access to cutting-edge generative tools and serves as a strategic test bed for future monetization. As other platforms tighten access, Google is betting on scale and experimentation.

Google Jules: Free, GitHub-Integrated Agentic Coding Platform
Jules is Google’s new free coding agent powered by Gemini 2.5 Pro. It integrates directly with GitHub, supports multi-million-token context windows, and offers a deep reasoning mode called DeepThink.

Deeper Insight:
Jules could upend the current AI coding landscape. With a serious open-source bent and full project awareness, it offers a challenge to Cursor, Replit, and even GitHub Copilot. The fact that it’s free raises the stakes even higher.

Google Meet Adds Real-Time Speech Translation
Google Meet now offers live translation of speech during meetings, adding to its already-strong transcription tools. Users can conduct multilingual conversations with real-time translated captions.

Deeper Insight:
This could become a standard for international collaboration, especially in remote teams. As live translation becomes more common, cross-border business may get a lot more frictionless.

Google Unveils Gemini-Integrated Smart Glasses With Warby Parker
Google revealed a three-tiered roadmap for XR hardware, including a consumer-focused pair of smart glasses co-developed with Warby Parker which features an in-lens display in addition to camera. These glasses will integrate Gemini AI for contextual assistance, translation, and more.

Deeper Insight:
This direct challenge to Meta’s successful Ray-Ban AI glasses positions Google to lead in consumer-grade wearables, except it won’t happen for another 6 months at least. If the execution matches the ambition, Gemini glasses could represent the next big interface shift after smartphones.

China Launches 12 Satellites for Orbital AI Supercomputer Network
China kicked off its plan for a 2,800-satellite constellation called Three Body Computing. The first 12 satellites are already in orbit, with a combined processing power of 5 POPS (Peta operations per second) and laser-linked data transfer speeds of 100GB per second.

Deeper Insight:
This may be the dawn of orbital computing. China is betting big on off-Earth infrastructure to gain AI edge and reduce ground-based data center dependencies. It’s a strategic move with global implications.

Duke University Unveils WildFusion, a Multisensory Robotics Framework
Researchers at Duke created WildFusion, a system that fuses touch, vibration, and visual data into a single input stream for embodied AI. It allows robots to process environments more like humans.

Deeper Insight:
Combining multiple sensory modalities could help robots better navigate unpredictable terrain and perform delicate tasks. This is especially relevant for search-and-rescue or industrial automation in dangerous settings.

New Haptic Feedback System From Pohang University Enhances Remote Robot Control
A team from Pohang University developed a system that provides real-time haptic feedback to operators of remote robots. This allows operators to “feel” how much force the robot applies, improving precision and safety.

Deeper Insight:
This could greatly improve human-robot collaboration in hazardous environments. It also shows how tactile feedback might become a key feature in next-gen robot UX, bridging the gap between autonomy and manual control.

Microsoft Releases Discovery Agent for Scientific Hypothesis Generation
Microsoft announced Discovery, an AI agent that helps scientists simulate and evaluate research hypotheses. It has already been used to discover new non-toxic coolants in just 200 hours.

Deeper Insight:
AI is starting to conduct science. Tools like Discovery could dramatically reduce R&D cycles in fields ranging from materials to pharmaceuticals.

Microsoft Updates 365 Copilot With Multi-Agent Orchestration
Microsoft rolled out a major Copilot upgrade, allowing orchestration of multiple agents and fine-tuning of model responses grounded in internal company data. It aims to make enterprise Copilot more customizable and intelligent.

Deeper Insight:
This brings Microsoft closer to offering a full-fledged agentic work environment. If adopted widely, it could reshape how companies handle workflows, analytics, and internal documentation.

GitHub Copilot Goes Open Source in VS Code
GitHub Copilot is now open source within Visual Studio Code, making it easier for developers to audit, extend, and modify its behavior.

Deeper Insight:
Open-sourcing a popular tool like Copilot accelerates innovation and transparency. Expect new forks, plugins, and community enhancements that push developer tools even further.

Microsoft Adds Grok 3 and Grok 3 Mini to Dev Platform
Microsoft added support for Grok 3 and Grok 3 Mini models in its AI development suite. These now sit alongside OpenAI’s offerings, expanding available model choices for enterprise developers.

Deeper Insight:
This signals a shift toward model pluralism. Microsoft seems eager to avoid over-reliance on any one vendor, possibly anticipating regulatory or strategic turbulence in their partnership with OpenAI.

MIT Retracts AI Productivity Study Over Data Concerns
MIT pulled support for a widely cited paper claiming AI dramatically increased scientific productivity. The paper was posted to arXiv and submitted to the Quarterly Journal of Economics, but is now under scrutiny for unverifiable data.

Deeper Insight:
This raises alarm bells about how quickly flawed AI research spreads. As large language models ingest this data, it becomes harder to correct misinformation unless retractions are systemically flagged and tracked.