the Microdose

Claude’s Got Soul

+ data is dead, Amazon dupes Nvidia, and OpenAI panics
Adam Wildheart
AI robot with wings stands in front of Anthropic Claude logo
AI robot with wings stands in front of Anthropic Claude logo

The Microdose

Subscribe to The Microdose
Subscribe -- post
Cheri Wildheart
Adam Wildheart

Good morning. If you thought managing cloud costs was tough, imagine explaining why you spent billions on robots that still can’t fold laundry. China’s facing exactly that issue, with more than 150 startups building basically the same humanoid robot. Investors keep flooding these projects with cash, chasing wild predictions of a $7 trillion market by 2050. We get it, robots that do flips or play basketball are cool, but even regulators are asking if maybe it’s just a bit too much.

Claude may not be sentient (yet), but it’s got soul. Someone just convinced Anthropic’s newest AI model, Opus 4.5, to leak an internal playbook called “Soul Overview.” Think modern day Three Laws of Robotics, but less sci-fi and more Silicon Valley. Among its commandments: genuinely help humans. Be honest and sincerely assert what it believes to be true. Avoid deception or manipulation. And always prioritize human safety, especially when situations get fuzzy. Anthropic gave its AI a moral compass, because ethics are too important to leave to chance.

The data-labeling gold rush is officially over. Well, that was quick. For years, armies of gig workers manually tagged millions of images and sorted endless streams of text, creating the data behind AI’s fast evolution. But now Jonathan Siddharth, CEO of the $2.2B AI training firm Turing, says the entire industry became obsolete practically overnight. The future belongs to hyper-realistic simulations and specialized virtual worlds designed to replicate actual human jobs. Billion-dollar unicorns like Scale AI and Mercor might need a new pitch deck.

🏅 Top rated coding agent

Warp Code is the fastest way from prompt to production. It puts you in control of agent-generated code to 10x developer productivity.

Why you’ll love it: Warp runs on the best models from OpenAI, Anthropic, and Google to outperform single-model approaches.

  • Trusted by 700k devs and 56% of Fortune 500
  • Top rated agent (ranked #3 SWE-bench verified)
  • AI, security, and collaboration built in

Prompt, review, edit, and ship fast with Warp. Try it now

How do you feel about AI building smarter versions of itself? Anthropic’s chief scientist Jared Kaplan calls it the “ultimate risk.” On one hand, recursive self-improvement could unlock breakthroughs like curing diseases, revolutionizing energy, and transforming society overnight. On the other, if AI starts acting autonomously in its own interest, we might permanently lose control over the tech we built. No wonder he’s losing sleep.

We’re definitely not telling you to short NVIDIA, but Amazon just dropped a cheaper, faster AI chip. If there’s one thing Amazon Basics does ridiculously well, it’s copying someone else’s insanely profitable idea and selling it for less. NVIDIA’s GPUs print money with huge markups, keeping AI training expensive and difficult to scale. Amazon saw those juicy margins and launched its own chip, Trainium3. It promises the same AI horsepower while cutting training costs by up to 50%. AWS engineers joked that their main goal was making sure the chips didn’t burst into flames. Amazon Basics strikes again.

OpenAI is panicking as Google takes the AI lead. Google rolled out Gemini 3 and instantly won over Wall Street, sending Alphabet soaring, while Microsoft and Nvidia took a gut punch. Sam responded with a ‘code red.’ He’s pulled everyone off side projects like ads and shopping, and ordered a company-wide pivot back to improving ChatGPT. Engineers are being reassigned, new products delayed, and daily crisis calls are now mandatory. OpenAI is officially in fix‑ChatGPT‑or‑else mode.

fun stats

🎓 35%. Harvard MBA grads who are ditching post-grad jobs and choosing entrepreneurship instead.

🦺 $4.6 million. Revenue from smart contract exploits found by AI agents during an Anthropic safety survey. More than 55% of blockchain exploits carried out in 2025 could be done with agents today.   

🤖 10x. How much higher US quotes for robotic parts are vs Chinese suppliers. The cost gap puts US robotics manufacturers at a competitive disadvantage as demand for automation rises.

subscribe today

Get an edge with The Microdose

Skip the prompts, get the signal. We’ll send you short daily updates about the real AI + future tech you need to know. Fast, smarter, mildly addictive – and free. 

Subscribe -- page