Keep pulling the thread on Alex Wang.
Meta expects to release its next, more powerful AI model in the coming months.
Anthropic's Claude Code has overtaken OpenAI's ChatGPT in revenue.
Meta's internal testing of the MuseSpark model triggered safety guardrails related to biological, chemical, cyber, and loss-of-control risks.
The current version of MuseSpark is not suitable for open-sourcing because it triggered internal safety checks.
Meta believes that physical superintelligence is a critical capability that must be developed within years, following closely after digital superintelligence.
Meta expects its upcoming AI models to be state-of-the-art in its key areas of focus.
Meta's brain prediction models can achieve good zero-shot generalization, enabling them to predict a new person's brain response to stimuli like images, videos, or audio without prior training on that individual's brain data.
Meta is currently developing AI models that are superior to MuseSpark.
The next phase of AI technology is primarily constrained by compute, creating a stratification where companies with large compute resources can build things that others cannot.
A fundamental problem with Meta's prior AI efforts was a lack of organizational conviction that superintelligence is a near-term, realistic goal.
Meta Superintelligence Labs' strategy to achieve high research velocity is based on higher compute per researcher, high talent density, and making ambitious research bets.
Meta's AI strategy is focused on building powerful agents to empower its billions of consumers and hundreds of millions of business customers, creating a new "agentic ecosystem."