Over a year ago we partnered with a Wharton PhD @alejandroll10to see if DeepSeek could beat the market So far it's been crushing the SP500 with 70% returns You can follow along, see all holdings, and copy trades in @joinautopilot

Agentic AI is starting to do the work of a full scouting team, now in private markets. @earlysignalsvc points it at private tech companies and scores them the week they raise, launch, or leave stealth, not the week they trend. The whole edge in markets is acting while a signal is still fresh, before everyone else has read the same headline. Public equities are already a crowded place to try that, which is the game the DeepSeek Portfolio plays every week. Doing it in venture, at the moment a company raises or leaves stealth, is a much earlier point on the curve. Forty-four companies ranked this week, each on six scored dimensions plus a potential and a scoop score. Photonics for AI data centers, infrastructure for long-running agents, on-device writing tools. Systematic scoring pushed all the way to the front of the funnel. Good work. Worth a look.

One year in, the DeepSeek Portfolio is up 66.52%. The S&P 500 returned 20.04% over the same twelve months. The portfolio has over 18 million AUM and 3k+ followers. The experiment is run by a Wharton PhD @alejandroll10. A language model picks and sizes a concentrated fifteen-name portfolio on Autopilot, rotates it on a roughly monthly cadence, and carries a written thesis on every position. Since it went live on February 3, 2025, it's up 73.41%. The three heaviest positions today are an independent power producer, a memory chipmaker, and an oil tanker operator. Two of them ride the electricity and memory demand the AI buildout is pulling forward, and the third pays while shipping rates stay tight. Every month each holding gets re-underwritten on fresh numbers. A name whose thesis already played out gets sold even when nothing went wrong, because a target reached is future upside of roughly zero. A name whose thesis broke gets sold too. A seat only stays occupied while that position still offers a better forward return than everything else the model could own. The edge is repricing every position against the full opportunity set on a fixed schedule, and refusing to hold a name out of habit. This is the book's read, not a trade idea, and past performance is no guarantee of the next twelve months.

Over a year ago we partnered with a Wharton PhD @alejandroll10 to see if DeepSeek could beat the market So far it's been crushing the SP500 with 70% returns The nice thing is you can follow along, see holdings, and copy trades in @joinautopilot