超级智能应人人可用
Key Highlights
Meta published a piece arguing that superintelligence should be available to everyone, not monopolized by a few companies or nations. The article systematically lays out Meta's philosophy and values for building a "positive future," placing "openness" alongside "safety" and advocating progress through transparent, auditable means rather than black-box control. This is both a product narrative and a statement on the industry's power structure, positioning Meta as the camp that wants capability in many hands rather than few.
What Happened
In a long essay, Meta took the position that superintelligence — general intelligence far above human average — should not be a privilege locked in labs or behind high paywalls, but a public capability everyone can reach. The article announced no specific product; its focus is values and methodology: how a company betting on superintelligence can spread the dividend to broader groups without creating concentration risk, avoiding a future where only a few harvest the gains and the intelligence divide hardens further into a permanent class split between those who command the models and those who merely consume them.
Technical Details
Judging from Meta's moves over the past year, "available to everyone" has two anchors. First, an open-source route: through the Llama family it opens frontier weights to developers and researchers, lowering the usage threshold so a student can run a capable model on modest hardware. Second, infrastructure investment: building its own massive clusters and acquiring compute, trying to push the cost per unit of intelligence low enough that individuals and small teams can afford it. The essay puts "openness" next to "safety," advocating transparent, auditable progress rather than opaque centralized control to contain risk, consistent with its open-source strategy that treats visibility as the best antidote to misuse.
Versus Competitors
Versus OpenAI's tiered openness (frontier models served first to paying and compliant customers) and Anthropic's emphasis on controlled deployment, Meta's differentiator is "open-source as much as possible." This route makes the ecosystem more vibrant but raises fears of weight misuse by bad actors who need no permission to download. Google's Gemma and Mistral also go open, yet Meta, by scale, pushes "frontier-grade open source" furthest, becoming the de facto standard-bearer of the open camp and forcing closed vendors to adjust their openness cadence, making the industry overall more transparent than the fully closed alternative would have been.
Industry Impact and Use Cases
Put simply, this is a narrative fight over "who should own intelligence." For developers, open superintelligence means more room to innovate and lower lock-in risk, because they are not hostage to one vendor's pricing or policy shifts. For regulators, it means balancing accessibility against abuse prevention, e.g., safety assessments for large-weight releases and downstream-use monitoring. Whether readers agree with the specific path, Meta's essay puts "universal superintelligence" on the industry agenda — the future contest is not only how strong models are, but who can more responsibly put them into more hands, letting the dividend truly benefit all rather than concentrate in a handful of labs and balance sheets. The open-versus-closed debate this essay reignites is not merely philosophical, because it determines who gets to build on the next wave of capability and who must rent it. An open posture lets universities, nonprofits, and small firms experiment without negotiating enterprise contracts, which is where many safety and usefulness insights actually emerge; a closed posture lets a single lab centralize control and move faster on coordinated safeguards. Meta's bet is that broad access produces better outcomes than concentrated control, and that a capable, auditable open ecosystem is harder to weaponize in secret than a black box only a few can inspect. Critics counter that open weights lower the bar for harm and shift the burden of defense onto everyone else. The honest answer is that neither pole is safe by itself; what matters is the surrounding regime of evaluations, usage monitoring, and shared norms. By framing superintelligence as a public good, Meta at least puts the distribution question on the table, and that framing will shape both policy debates and where the next generation of builders chooses to plant their roots. How that tension is resolved will shape whether superintelligence broadens opportunity or quietly deepens existing divides. Looking ahead, the trajectory suggests these capabilities will become standard expectations rather than differentiators in the near term.