Alexander Mashrabov

Alexander Mashrabov was one of ToTheMoon's hosts in 2024 and 2025 and later stays in the conversation as someone the other hosts cite. Episode 156 introduces him to viewers as the owner of Higgsfield and a former Snap director who placed a serious bet on Google — at one point more serious than on OpenAI or Anthropic. He brings a practitioner's view: someone who deploys models himself and reads their licenses and accuracy figures.

The episodes he takes part in are about verification rather than announcements. In episode 044, after DeepSeek's surge, the question is where a model's output can be trusted: Deep Research is examined as a tool that searches on its own and asks clarifying questions, scoring 26.6 percent against 3.3 for 4o on the test discussed. In episode 047 the Gandalf game shows eight layers of protection talked past in plain conversation — a template for attacks on agents with access to mail and payments. Episode extra03, on open-model licenses — MIT for DeepSeek, Llama's own terms — cites his words that Snapchat would gladly use Llama but its license does not allow it, and concludes that what matters is not the words open source but specific rights and liabilities.

The other hosts cite his words too: his remark about detecting model-written text (episode 062), his point that independent parallel teams work on different parts of OpenAI's product (episode 092), and his observation about data centers in America and China with Europe lagging behind (episode 075). In episode 088 the host announces a special episode with him, and in episode 076 he explains Meta's bet on the next mass consumer device himself.