AI / AGI = 3 / 4 ?

 Demis Hassabis: We're Three Quarters of the Way to AGI - YouTube

In this conversation, Demis Hassabis, co-founder of Google DeepMind, discusses his vision for the future of AI and science. Here are the key takeaways:

• The Path to AGI: Hassabis maintains that AGI remains a realistic goal for 2030, a target he has consistently projected for years.
• AI for Science: The primary mission of DeepMind has always been using AI as a tool to accelerate scientific discovery. He highlights AlphaFold as a major breakthrough that successfully tackled a long-standing grand challenge in biology.
• Revolutionizing Drug Discovery: By leveraging AI in biochemistry and chemistry—specifically through Isomorphic Labs—the goal is to move the majority of the drug discovery process into a simulated environment, potentially reducing development timelines from years to days.
• The Role of Simulations: He argues that AI-driven simulations offer a way to study emergent systems—like biology, economics, and climate—in ways that traditional mathematics or isolated experiments cannot.
• Information as Fundamental: Hassabis proposes a philosophical perspective that information is the most fundamental substance in the universe, suggesting that AI's ability to process and organize information is profoundly significant.
• Philosophy and Science: He draws inspiration from philosophers like Kant and Spinoza, viewing his work in AI and science as a means to understand the nature of reality and the human mind.

Simpler code with early returns: "Guard Clauses"

This is how "ideomatic Go" (or any other code) should like, simpler.

Guard Clauses: Simplifying Code with Early Returns | by Vaibhav Mojidra | Medium

Guard clauses (or early returns) are conditional statements placed at the beginning of a function that exit early if specific conditions are not met, reducing nested if-else structures. They improve readability by separating error handling and validations from the main logic

JS

function getMemberDiscount(member) {
  if (member) {
    if (member.isPremium) {
      return "20% off";
    } else {
      return "10% off";
    }
  } else {
    return "No discount";
  }
}

=>
function getMemberDiscount(member) {
  // Guard clauses
  if (!member) return "No discount";
  if (!member.isPremium) return "10% off";
  // Main logic (no nesting)
  return "20% off";
}

In modern programming languages like C#, Java, and Dart, guard clauses can be used within switch expressions to add additional conditions to a pattern match. These are often called case guards or when clauses. 

C#
public static string GetShippingCategory(int weight) => weight switch
{
    <= 0                  => "Invalid weight",          // Base pattern
    var w when w < 5      => "Lightweight",             // Guard clause (when w < 5)
    var w when w < 20     => "Standard",                // Guard clause (when w < 20)
    _                     => "Heavy / Freight"          // Discard (default)
};

Java

static String checkString(Object obj) {
    return switch (obj) {
        case String s when s.length() == 1 -> "Short string"; // Guard clause
        case String s                      -> "Long string";  // General pattern
        default                            -> "Not a string";
    };
}

Guard Clauses: Simplifying Code with Early Returns | by Vaibhav Mojidra | Medium

Living in simulation? 1M people => Mars by SpaceX, Musk

Nobel Prize Just Given for Proving the Universe Isn't Real! - YouTube @ImpactTheory

• The Simulation Hypothesis: The universe functions similarly to a video game, where objects are only rendered when they are observed or interacted with, rather than maintaining a constant, permanent state.
• Quantum Realities: Experimental evidence, such as the double-slit experiment, reveals that quantum particles exist as probabilities rather than definite objects until measured, challenging classical notions of physical realism.
• Retroactive Causality: The delayed-choice experiment demonstrates that current observations can retroactively influence the past state of a particle, suggesting that the universe constructs history as needed to satisfy current conditions.
• Nobel Prize-Winning Evidence: The 2022 Nobel Prize in Physics recognized work proving that the universe is not locally real. This means objects are not independent of one another and distance is an illusion, similar to how distant objects in a game exist in the same computational memory.
• Computational Universe: The base layer of reality appears to be information processing, math, and code rather than physical matter. This makes the distinction between our world and a highly advanced simulation functionally non-existent.

The Nobel Prize mentioned in the video is the 2022 Nobel Prize in Physics, which was awarded to Alain Aspect, John F. Clauser, and Anton Zeilinger for their groundbreaking experimental work with entangled photons (19:36 - 23:56).

Why it matters:

• Violation of Bell Inequalities: Their experiments proved that the universe does not adhere to "local realism," meaning particles remain linked regardless of the distance between them, and their states are not predetermined by "hidden variables" (21:36 - 23:47).
• Pioneering Quantum Information Science: This work established the experimental foundation for modern quantum technologies, including quantum computing and secure quantum communication (23:56).
• Computational Implications: The video argues that these findings mirror how video game engines work, where distance is an illusion and objects are processed by a central system only when they need to be "rendered" via observation (24:26 - 26:41).

Pioneering Quantum Physicists Win Nobel Prize in Physics | Quanta Magazine

Core Discovery: Quantum Entanglement

The laureates conducted experiments that proved quantum entanglement—a phenomenon where two particles remain connected and influence each other instantly, regardless of the distance between them.

Key Research Milestones

• John Clauser: Built the first practical experiment to test Bell’s Theorem, proving that quantum mechanics cannot be explained by "hidden variables" (Einstein’s preferred view).

• Alain Aspect: Developed a more sophisticated setup that closed a major loophole by switching measurement settings while particles were in flight, leaving no room for classical communication.

• Anton Zeilinger: Demonstrated quantum teleportation and entanglement swapping, showing that quantum states can be transferred across distances without moving the physical object itself.

Why It Matters

• Settled a Century-Old Debate: Their work confirmed that the universe is fundamentally "spooky" at a quantum level, vindicating the views of Niels Bohr and Erwin Schrödinger over Albert Einstein.

• Technology Foundation: These experiments transitioned entanglement from a philosophical debate to a quantum resource.

• Future Applications: Their work is the backbone for emerging technologies like unbreakable cryptography, high-speed quantum computing, and secure global quantum networks.

SpaceX ties Musk compensation to Mars colonization goal - YouTube

• Valuation Target: Musk is eligible for $200 million in super-voting restricted shares if SpaceX achieves a market valuation of $7.5 trillion.
• Mars Colonization & Compute Goals: He can receive an additional 60.4 million in restricted shares if the company successfully establishes a permanent human colony on Mars (housing at least 1 million people) and operates space-based data centers with at least 100 terowatts of compute capacity.
• Performance Structure: The awards utilize Class B super-voting stock (10 votes per share) and vest in tranches as the company value grows. Notably, Musk receives zero shares if the company fails to meet these board-set valuation targets.

SpaceX Ties Musk Pay To Mars Colony

A CEO compensation plan hinges not on revenue or profit, but on putting 1 million people on Mars and building massive space-based data centers. SpaceX approved a pay package for Elon Musk awarding up to 200 million super-voting shares if the company reaches a $7.5 trillion valuation and establishes a permanent Mars colony. Additional awards include 60.4 million shares tied to operating space data centers delivering 100 terawatts of compute capacity.

 

TypeScript faster than Rust=>WASM ?!

Rewriting our Rust WASM Parser in TypeScript | OpenUI.

...openui-lang parser in Rust and compiled it to WASM. The logic was sound: Rust is fast, WASM gives you near-native speed in the browser, and our parser is a reasonably complex multi-stage pipeline. Why wouldn't you want that in Rust?

Turns out we were optimising the wrong thing.

End result: 2.2-4.6x faster per call and 2.6-3.3x lower total streaming cost.

AI news: many.... GitHub issues

 China Takes Over - YouTube by Mat.B.

DeepSeek 4: open frontier model for fraction of price?

OpenAI's Co-Founders Finally Tell Their Side Of Everything - YouTube

Sam Altman and Greg Brockman story...

Told you so - YouTube by Max S.

GitHub Subscription: no "free tokens", just pre-paid tokens

AI subscriptions are becoming less attractive - YouTube

Github are you joking? - YouTube by PrimeTime

GitHub ruining repos?

GitHub is facing HUGE problems! - YouTube Max S.

GitHub is currently facing a combination of severe technical and strategic issues that have impacted its reliability as a development platform:

• Critical Security Vulnerabilities: The platform recently addressed a major remote code execution flaw, which highlighted risks in how push requests are handled.
• Infrastructure and Reliability Issues: Persistent downtime and bugs in core features, such as merge queues, have disrupted user workflows and decreased platform trust.
• AI-Driven Traffic Surge: A massive, rapid increase in code and project activity—attributed to the rise of AI—has placed immense strain on GitHub's infrastructure while it attempts to migrate to a modern, cloud-based microservices architecture.
• Strategic Misalignment: Since being integrated into Microsoft's Core AI division, GitHub has focused heavily on AI tools like GitHub Copilot. Critics argue this shift has caused the company to neglect foundational features, leaving open-source maintainers overwhelmed by low-quality, AI-generated contributions.
• Leadership Concerns: The lack of a dedicated CEO has fueled perceptions that the platform’s original mission of supporting developers is being subordinated to broader corporate AI goals.

GitHub RCE Vulnerability: CVE-2026-3854 Breakdown | Wiz Blog

An update on GitHub availability - The GitHub Blog

Securing the git push pipeline: Responding to a critical remote code execution vulnerability - The GitHub Blog

TypeScript validation: Zod performance: interpreter vs compiler

Intro | Zod  (//zod.dev/)

TypeScript-first schema validation with static type inference

Defining schemas | Zod

• Zero external dependencies
• Works in Node.js and all modern browsers
• Tiny: 2kb core bundle (gzipped)
• Immutable API: methods return a new instance
• Concise interface
• Works with TypeScript and plain JS
• Built-in JSON Schema conversion
• Extensive ecosystem

Why is Zod so slow? - LogRocket Blog

Zod is the undisputed king of TypeScript validation. In a remarkably short time, it has become the gold standard, celebrated across the ecosystem for its phenomenal developer experience, elegant chainable API, and robust type inference. For countless developers, Zod is not just a library — it’s the default, instinctive choice for ensuring data integrity in their applications.

While the high-performance library Typia clocks in at an astonishing ~76 million operations per second, the latest version of Zod (v4) performs around 6.7 million ops/sec.

To understand why the performance gap exists, we need to look under the hood. The difference between Zod and its faster competitors isn’t about small optimizations; it’s a fundamental split in their architectural philosophy. It’s a classic tale of the interpreter versus the compiler.

TypeScript Runtype Benchmarks

AI summary

Zod is widely considered the de facto industry standard for runtime data validation in modern TypeScript projects. Its ability to provide full static type inference from schema definitions makes it the preferred "TypeScript-first" choice, frequently used in frontend forms, API request validation, and frontend-backend communication. 

Key Reasons for Its "Standard" Status:

• TypeScript-First Design: Unlike older libraries like Joi, Zod was built specifically for TypeScript, allowing developers to create a single source of truth for both runtime validation and type definitions.
• Widespread Ecosystem Adoption: It is highly popular in frameworks like Next.js, Nuxt, and in tools like tRPC, TanStack Form, and Hono.
• Developer Experience: Known for an elegant, chainable API and robust developer experience.
• Zero Dependencies: Zod is lightweight and brings no external dependencies into a project. 

Industry Standing vs. Alternatives:

• Zod vs. Yup: Zod has largely taken over from older libraries like Yup, which was previously standard but lacks native, seamless TypeScript type inference.
• Zod vs. Others: While libraries like Joi and AJV exist (often used for specific JSON schema needs), Zod has become the default, instinctive choice for new TypeScript applications.
• Future Trends: While heavily used, newer libraries like Valibot or Effect Schema are sometimes preferred in performance-critical scenarios (due to better tree-shaking or speed), but Zod remains the most popular choice. 

Zod is a TypeScript-first schema declaration and validation library. In the context of AI and APIs, it is primarily used to ensure structured output from Large Language Models (LLMs) and to provide runtime type safety for API integrations. 

Key Uses in AI and APIs

• Structured AI Outputs: Zod schemas are used to define the exact shape of data an LLM should return (e.g., JSON). Tools like the Vercel AI SDK use Zod to validate and parse these AI-generated objects.
• OpenAI Function Calling: Developers use Zod to define the parameters for OpenAI Tool/Function calls, ensuring the model sends correctly formatted arguments to your code.
• API Validation: Libraries like Zodios or zod-to-openapi use Zod to automatically generate OpenAPI/Swagger documentation and validate incoming request bodies or outgoing responses.
• Error Handling: Zod acts as a "bridge" for external APIs with unpredictable data, allowing you to gracefully handle schema mismatches before they break your application. 

Popular AI Libraries Supporting Zod 

1. Vercel AI SDK: Deeply integrates Zod via the zodSchema() function for generating structured objects.
2. LangChain: Uses Zod for output parsers to convert raw text from LLMs into typed TypeScript objects.
3. Zod-GPT: A specialized utility that uses Zod to get fully typed and validated responses from GPT models.
4. OpenAI-Zod-Functions: A helper library for defining OpenAI functions with Zod schemas instead of raw JSON Schema.