Dear Readers,

The smartest AI systems on Earth just failed a test that children ace without breaking a sweat: ARC-AGI-3 dropped yesterday, and frontier models scored below 1% where humans hit a perfect 100%, exposing a canyon between pattern recall and genuine reasoning that no amount of scaling has closed.

But while one benchmark humbles the machines, the rest of today's issue shows an industry charging forward at full speed: Google's Lyria 3 turns text prompts into production-ready music tracks, Apple is quietly distilling Gemini into bite-sized models that run directly on your iPhone, and Perplexity's CEO argues that the wave of AI layoffs might actually birth a golden age of solo founders building million-dollar companies with a laptop and a prompt.

We've also got NVIDIA's VP of AI Models explaining why the company gives away its best work for free (yes, that's my interview with Kari Briski :)), engineering job openings hitting a 3-year high despite all the doom talk, and a McMaster University study linking ultra-processed food to a 60% drop in fertility — a reminder that not every disruption worth worrying about comes from silicon.

Grab your coffee, this one's packed.

Google’s new Lyria 3 and Lyria 3 Pro models bring high-quality, structured music generation to developers, enabling full songs (up to ~3 minutes!) or fast 30-second clips with realistic vocals and multi-genre support. With precise controls like tempo setting, time-aligned lyrics, and even image-to-music inputs, creators can build highly customized audio experiences, from video soundtracks to dynamic alarm songs.

This pushes AI music beyond simple loops into production-ready, expressive compositions while maintaining transparency via SynthID watermarking—opening serious opportunities for apps, content creation, and personalized media. And it pushes even more pressure on SunoAI.

Engineering job openings have surged to a 3-year high, with over 67,000 global roles, raising the question of whether AI is driving or limiting demand.

{Text Goes Here if Any}

The smartest AI Models on the planet just got schooled, by mini-games a child could beat. The ARC Prize Foundation launched ARC-AGI-3 yesterday, and it could rewrite the rules of how we measure machine intelligence. Unlike previous versions that tested static grid puzzles, this benchmark drops AI agents into interactive, video-game-like environments with no instructions, no stated goals, and no prior training data to lean on.

Think of it like handing someone a game controller and saying: "Figure it out." Humans do this effortlessly, we explore, form theories, test them, adapt. The best AI agent during the preview phase scored just 12.58%, while frontier LLMs landed below 1%. Humans? A perfect 100%.

The ARC Prize 2026 competition offers over $2 million across three tracks, and every winning solution must be fully open-sourced. François Chollet, who created the original ARC test in 2019, has long argued that memorization is not intelligence and ARC-AGI-3 is his sharpest proof yet. Over 1,200 human players tested the benchmark during its preview, establishing the baseline every AI agent is now scored against.

Why it matters: ARC-AGI-3 exposes a fundamental gap between pattern recall and genuine reasoning that today's most powerful AI models still can't close. It sets a new, measurable standard for what real progress toward AGI looks like, and makes that progress open-source by design.

Sources:
🔗 https://arcprize.org/competitions/2026/arc-agi-3

Attio is the AI CRM with conversational AI built directly into your workspace. Every morning, Ask Attio handles my prep:

All in seconds. No searching, no switching tabs, no manual updates.

Ready to scale faster?

Start now →

Nemotron 3 Super is now a leading reasoning foundation for OpenClaw and complex agentic workflows, with 1.5M+ downloads in its first two weeks.

That bag of chips might cost more than you think, and the price could be measured in fertility. A first-of-its-kind study from McMaster University has found that women who eat more ultra-processed foods have significantly lower odds of conceiving, even after researchers accounted for age, weight, and lifestyle factors.

The team analyzed data from over 2,500 women using the U.S. National Health and Nutrition Examination Survey, and the numbers are striking: in fully adjusted models, higher ultra-processed food intake was linked to roughly 60 percent lower odds of fertility. That's not a rounding error, that's a red flag. The culprit likely isn't just calories. Ultra-processed foods often carry chemicals like phthalates, BPA, and acrylamides that can leach from packaging and processing machinery, compounds known to disrupt hormones.

Interestingly, the Mediterranean diet showed a positive link to fertility, but that benefit vanished once obesity was factored in, suggesting it works mainly through weight management. A separate study published in Human Reproduction the same week found that ultra-processed food consumption in men was also linked to reduced fertility and slower early embryonic development, making this a problem for both partners. The researchers stress this isn't about perfection. It's about awareness. Even small shifts toward whole, recognizable ingredients could reduce exposure to chemicals science is only beginning to understand. So here's a question worth asking at the dinner table: How much of what's on your plate was made in a factory?

This study reframes the ultra-processed food debate from a weight issue to a reproductive health crisis, revealing that hidden chemicals in everyday foods may disrupt hormones in ways that go far beyond calories. With infertility rates climbing globally, dietary choices could become one of the most accessible levers for change.

AI will eliminate 300 million jobs in the next 5 years.

Yours doesn't have to be one of them.

Here's how to future-proof your career:

Start learning AI now