A 6,500-word AI manifesto argues for broader access to AI models as essential to the industry’s future, with emphasis on reducing concentrated power in AI leadership and empowering individuals through what the author calls a form of personal superintelligence.

Policy recommendations urge the U.S. government to support safety testing collaborations with AI firms and warn that government delays could hand an advantage to China in AI development.

The manifesto envisions widely accessible AI that lowers barriers for independent developers and small businesses by moving beyond expensive enterprise contracts.

Muse Spark, Meta’s paid, closed model introduced last July, will continue operating in a closed format.

Open-weight models publish trained weights but typically withhold training data, code, architecture details, and full methodologies, trading transparency for flexibility.

Broader themes include U.S. domestic politics, foreign policy tensions, COVID-19 and vaccines, and ongoing debates among media figures about AI policy.

Meta plans to roll out Watermelon, a more powerful internal model, in the coming months, though it remains to be seen whether it will be open or closed.

For developers and users, the practical impact is broader access to powerful models without costly enterprise contracts, enabling more players to participate.

Current market dynamics show many American firms already using cheaper Chinese open-source models, prompting policy focus on enabling domestic open-source development.

Meta positions itself as a leading participant in the AI race by investing in models and integrating AI across products used by billions, while noting that regulation could affect pace and deployment.

Muse Spark lags behind top models in benchmarks for coding, reasoning, and writing, positioning it as a cost-effective alternative rather than a market leader.

Audience targets include regulators favoring open source, developers pushing open-source through Muse Glimmer, and competitors who could be pressured by an open-source advantage.