Happy Friday! Tomorrow is Valentine’s Day, and nothing says romance like an AI love poem, right? Tempting as it may be to let ChatGPT churn out a quick love letter, don’t. Research says you’ll end up feeling guilty. Apparently, emotional authenticity still matters. Scientists call this awkward feeling a “source-credit discrepancy” when you take credit for someone (or something) else’s words. The fix is simple. Let AI handle the brainstorm, then rewrite the message. Your Valentine and your conscience will thank you.
How you know we’re not AI agents: We’re taking Monday off for some well deserved rest. Enjoy the weekend and we’ll see you back here on Tuesday.
The race is on to standardize AI agents. Right now, businesses juggle multiple AI tools from different providers, each living in its own silo and speaking its own language. OpenAI thinks the answer is Frontier, a unified interface built into ChatGPT designed to manage interactions between agents and business applications. But traditional software firms aren’t convinced. They argue managing AI agents demands deep expertise in messy real world workflows like invoicing, budgeting, deployment, and accountability. And honestly, they have a point. Whoever cracks the code to standardize agent interactions could dominate enterprise software for decades. [The Information]
AI is getting really good at making predictions. Elite prediction competitions like Metaculus traditionally measure people’s forecasting skills on everything from elections to pop culture. At first, AI struggled badly, but lately, things have changed… fast. Mantic’s AI jumped from 8th place to 4th, beating the collective wisdom of top human forecasters. Built from specialized LLMs, these bots spot patterns across politics, entertainment, and economic trends by crunching data in ways people can’t match. They predict a 95% chance that AI will dominate prediction competitions by 2030. [The Atlantic]
Nebius Token Factory is how teams run large open-source AI at production speed. Deploy models like Llama, Qwen, and DeepSeek with enterprise-grade reliability.
- Predictable inference, even under heavy traffic
- Easy post-training controls
- Transparent, optimized pricing that scales
- Built for serious, large-scale AI
Anthropic and OpenAI are taking their rivalry to Washington. Anthropic just gave $20 million to a new super PAC called Public First Action in a bid to counter OpenAI’s political influence. Anthropic says the next few years of AI policy will reshape society, so it’s funding candidates who support stronger safety rules. OpenAI backs rival super PACs like Leading the Future, which push the kind of deregulation the Trump administration loves. Maybe instead of fighting over regulation, they should focus on ensuring AGI benefits all of humanity. [NYTimes]
Brain cells could soon replace silicon chips. Two neurosurgeons have raised $25 million betting that within 5-10 years data centers will run on living neurons billions of times more efficient than silicon. It started as a wild experiment feeding stock market data into actual brain cells on an electrode grid. The neurons didn’t just respond – they recognized patterns. Now their startup, The Biological Computing Co, uses these neural networks for practical tasks like sharpening images and speeding up video. No word yet on what they feed them. [Fortune]
World models might be the coolest tech nobody can afford. Robotics startups rave about virtual universes so realistic they can compress years of robot training into hours. Gamers love building digital worlds with physics lifted straight from reality. But there’s one major glitch. These simulations are brutally expensive because every user needs dedicated AI hardware. Odyssey’s basic model alone consumes an entire H200 chip worth roughly $40k. Upgrade to Pro and you’ll need a dedicated server rack, pushing hardware costs deep into six figures. Why so expensive? Because perfect digital reality demands billions of parameters updating every few milliseconds. Until someone solves that problem, world models will keep bankrupting startups. [The Information]
What if you could test decisions on a virtual audience before rolling them out in real life? That’s what Simile.ai is doing. The AI startup has raised $100 million to build hyper-realistic digital versions of real people using deep interviews, historical data, and decades of behavioral research. Think of it as a flight simulator for human decisions. Companies can safely rehearse earnings calls, courtroom strategies, or major strategic moves before committing. Because our future is too important for trial and error. [Bloomberg, Reddit]
fun stats
🚘 1 million. Weekly paid robotaxi trips Waymo aims for in 2026, over twice their current 400k rides per week.
👨💻 $380 billion. Anthropic’s valuation after raising $30 billion. Claude Code business subscriptions quadrupled since early 2026, and enterprise deals worth $100k+ jumped 7x in the past year.
🎤 0. Lines of code written by Spotify’s top developers since December. The CEO says Claude Code does the heavy lifting as they rapidly crank out new AI features.
🌜 $1.1 million. What European startup Deep Space Energy raised in pre-seed to turn nuclear waste into electricity on the moon. This could speed up moon missions by 5 years and help power the growing moon economy.