Good morning. You know how people joke about being replaced by robots? Well, Mark Zuckerberg took that personally and decided to replace himself first. He’s teaching his photorealistic AI clone the fine art of micromanaging employees. The 3D Zuck bot is programmed to think and sound exactly like him, but early reports suggest it’s slightly less robotic and capable of blinking. Is this part of Meta’s push toward personal superintelligence, or just a clever way to keep the real Zuck out of the office?
AI agents are mostly useless without human babysitters. New research from Scale AI explains why agents dazzle in demos but flop in production. Given clear instructions, agents nail up to 89% accuracy on coding and SQL tasks. But drop them into real world scenarios, where they need to spot missing context or ask for help, and accuracy plummets as low as 4%. The core issue is judgment. Instead of asking clarifying questions, agents guess or invent their own requirements. No wonder over 90% of enterprise AI agent pilots fail despite huge investments. Imagine an economy run by bots too clueless to even realize they’re clueless.
Who gets the Nobel Prize when an AI makes the discovery? A team from Sakana AI in Tokyo, along with researchers from Oxford and the University of British Columbia, built an AI Scientist that completes the entire research process alone. The AI independently generated original ideas, ran experiments, analyzed results, and even produced papers good enough to pass peer review. Three of the AI written papers were submitted to an international conference, with one scoring higher than over half of the human submissions. Eventually the team withdrew them to avoid forcing a debate about AI’s role in research too soon. But let’s be real, the debate is already here. Next up… investigative journalism.
Wispr Flow saves you time. Just speak and it turns your voice into polished text in any app. Zero edits. 4x faster than typing. Millions use it daily, including teams at OpenAI, Vercel, and Clay.
Now on Android (free + unlimited). Try Wispr Flow for $0.
Programmable biology is here and regulators can’t keep up. OpenAI and Ginkgo Bioworks just ran 36,000 biological experiments autonomously in robotic cloud labs. That’s insane. Old school labs take months to run only a few experiments. These new AI systems let scientists design proteins and vaccines faster, cheaper, and up to 4x more accurately. But biosecurity rules weren’t built for AI running thousands of automated experiments at once. Current laws leave huge gaps, making it way too easy to conduct high risk bio research without oversight. Cutting edge biotech meets dial up regulation. What could go wrong?
China has erased the US lead in AI. Stanford’s latest AI Index Report claims the two countries are now neck and neck on key benchmarks. While the US still holds an edge in chip tech (for now), their main focus has been chasing moonshots like AGI and autonomous agents. China took a more strategic route by investing in practical AI, locking down patents, dominating robotics, and boosting sustainable energy. Public trust widened the gap, with just 28% of Americans regularly using AI compared to 53% in China. Maybe going all in on AGI wasn’t the smartest move.
Reach your audience
Get in front of 70,000+ tech decision makers with budget and technical influence. 95% US based. 66% open rate. Read daily by AI leaders and builders at high growth companies.
Neuralink wasted years chasing the wrong dream. Musk spent billions developing brain implants meant to merge our minds with AI. So far, Neuralink’s best trick is letting patients move a cursor with their thoughts. Meanwhile, rivals raced ahead with implants that restore speech, helping patients talk again with 97% accuracy. Now Neuralink is scrambling to pivot toward speech restoration just to catch up. Elon loves talking cyborgs, but maybe the reason the market is bigger for speech is that nobody wants to spend a lifetime with Grok in their head.
Stop burning money on the wrong GPUs. Most people think tokens are inherently expensive, but new data suggests the real issue is picking bad hardware. After testing 50 LLMs across 10 different NVIDIA GPUs in over 5,000 experiments, researchers found your choice of GPU makes a massive difference. Picking the right GPU can slash inference energy usage by up to 70% for server deployments, and up to 20% in batch scenarios. That’s a huge cost savings without changing a single line of AI code. If inference is the new electricity, most companies are paying to heat data centers instead of powering their models. Too bad Groq sold out to Nvidia.
fun stats
🎓 $150. Hourly rate LinkedIn is paying people to train AI on specialized subjects like nursing, finance… and Nordic languages. Strange times when being fluent in Finnish beats coding.
🛍️ 75%. Americans who say they’d lose trust in AI shopping (and brands) if results were sponsored.
⚡ 3 years. How long it took for generative AI to reach 53% adoption. For context, that’s faster than PCs or even the internet.