AI's $500M Reality Check: When Hype Meets Hard Cash

Periodic Labs just raised $300 million to do what everyone talks about but nobody actually does: real science with AI. Founded by OpenAI's Liam Fedus and Google Brain's Ekin Dogus Cubuk, this isn't another chatbot company promising to revolutionize everything.

Here's what makes this different: they're building actual robotic labs that mix chemicals, run experiments, and analyze results. The AI suggests compounds, robots synthesize them, and machine learning models learn from both successes and failures. It's the first time someone's putting serious money behind AI that gets its hands dirty in the physical world.

The timing isn't coincidental. Robotic synthesis finally works reliably, ML simulations can model complex physical systems, and LLMs have reasoning capabilities that can actually interpret experimental results. Cubuk already proved this works at Google, creating 41 novel compounds using AI-suggested recipes.

But here's the real insight: failed experiments are just as valuable as successful ones because they generate training data. This flips the traditional scientific incentive system on its head, where failure means no publication and no funding. In AI science, every experiment teaches the model something new.

OpenAI is using classic Big Tech intimidation tactics, and it's blowing up in their faces. The company has been carpet-bombing nonprofits with subpoenas, demanding everything from funding sources to private communications about OpenAI's for-profit conversion.

The targets? Small nonprofits like The Midas Project (annual budget: under $75,000) and policy groups that dared to question OpenAI's transformation from nonprofit to money-making machine. These aren't Elon Musk's shadow organizations—they're legitimate advocacy groups doing exactly what they should be doing: holding powerful companies accountable.

The subpoenas are so broad they're essentially fishing expeditions. OpenAI wants to see every document, every communication, every funding source these organizations have ever had. One nonprofit couldn't even get legal insurance afterward because insurers were spooked by the OpenAI litigation.

Even OpenAI's own employees are calling this out. Mission alignment team lead Joshua Achiam publicly criticized the tactics, saying "We can't be doing things that make us into a frightening power instead of a virtuous one." When your own people are breaking ranks, you've crossed a line.

NGOs are flooding social media with AI-generated images of extreme poverty, and it's exactly as problematic as you'd expect. Stock photo sites now sell "photorealistic" images of suffering children, refugee camps, and sexual violence survivors—all completely artificial but designed to tug at heartstrings and open wallets.

The images are racially stereotyped and exaggerated beyond reality. Adobe sells licenses for AI-generated photos captioned "Asian children swim in a river full of waste" and "Caucasian white volunteer provides medical consultation to young black children in African village" for about £60 each.

Here's the twisted logic: it's cheaper and you don't need consent. As one researcher put it, "It is quite clear that various organisations are starting to consider synthetic images instead of real photography, because it's cheap and you don't need to bother with consent and everything."

The UN even posted AI-generated "re-enactments" of sexual violence, including fake testimony from a Burundian woman describing rape. They pulled the video only after media inquiries. This isn't just about ethics—these biased images get fed back into AI training data, amplifying prejudice in the next generation of models.

A 2-by-2-millimeter device is giving blind people their sight back, and it actually works. Researchers implanted tiny photovoltaic panels under patients' retinas, paired with camera-equipped smart glasses that beam images directly to the optic nerve.

The results are remarkable: 26 out of 32 patients could see well enough to read books and fill out crossword puzzles after one year. That's an 80% success rate for people who had lost central vision to age-related macular degeneration. The vision isn't perfect—it's blurry and black-and-white—but patients who couldn't see anything can now read again.

The technology comes from Science Corporation, founded by Max Hodak, who co-founded Neuralink with Elon Musk. The company acquired the retinal implant tech from French startup Pixium Vision after it ran out of money, proving that sometimes the best innovations come from rescuing abandoned projects.

This is brain-computer interface technology that's actually helping people today, not promising to help them someday. While other BCI companies chase headlines with flashy demos, Science Corporation is quietly restoring sight to people who thought they'd never see again.

The Trump administration is systematically deleting government guidance on AI risks, and it's happening faster than anyone expected. The FTC has removed blog posts about AI dangers, open-source models, and consumer protection—essentially erasing years of policy work overnight.

Three key posts have vanished: guidance on "open-weight" AI models, warnings about AI consumer harms, and analysis of AI fraud risks. These weren't partisan screeds—they were technical guidance that companies actually used to build safer AI systems. One deleted post even won an award from the Aspen Institute for making AI accessible to the public.

The removals violate federal record-keeping laws, according to FTC sources. The Federal Records Act requires agencies to preserve documents with administrative, legal, or historical value. But when you're trying to erase the previous administration's work, legal compliance becomes secondary to political messaging.

Former FTC public affairs director Douglas Farrar called it "shocking" that the new FTC leadership would contradict Trump's own AI Action Plan, which actually supports open-source AI development. It's policy incoherence driven by pure partisan reflex.