Zuckerberg described his vision for AI development in an essay
Meta CEO Mark Zuckerberg published an essay describing the company's vision for the artificial intelligence race, its importance to society, and how authorities should regulate and promote this technology. The Wall Street Journal recounts the key points.
Meta is currently lagging in the AI race and is trying to catch up with Anthropic and OpenAI, which possess the most advanced artificial intelligence models.
Open Source
Meta is resuming the release of models with open parameters (available for download), a practice it temporarily abandoned after last year's reorganization of its AI division. Zuckerberg also defends the practice of distillation (where developers learn from other models through queries) — he considers it a normal form of learning from what can be observed.
More money for cities near data centers
The company is launching the "$1 Billion Future Is For Everyone Fund" to support local communities near data centers — partly in response to growing dissatisfaction from activists with such projects.
Zuckerberg also wrote that one of the US's disadvantages in its struggle with China to develop superintelligent computer systems is the difficulty of creating infrastructure within the US. This year, Meta plans to invest up to $145 billion in capital expenditures, mainly for building data processing centers, and $600 billion by 2028.
Early government involvement in model development
Zuckerberg criticizes the current 30-day voluntary review period for new models as too long given the pace of AI development. Instead of strict regulation, he proposes close cooperation between laboratories and US authorities, including the transfer of intermediate training checkpoints even before completion.
More internal control over releases, including at Meta
Meta is introducing a new governance structure: the board of directors gains the right to assess whether new models meet approved safety criteria. Zuckerberg states that he does not believe it is right for a single person to decide on the deployment of superintelligence, and calls on other companies and the entire industry to adopt this approach.
Users must define the form of AI
Developing the concept of "personal superintelligence," Zuckerberg insists that users, not the model developer, must decide which values to embed in the system. He also promises a fully private mode for personal AI agents, to whose data even Meta itself will not have access.
Now reading
Flew NATO personnel to Afghanistan and dropped humanitarian aid in Sudan, then ended up in prison due to a denunciation from a friend in Ukraine 1
Flew NATO personnel to Afghanistan and dropped humanitarian aid in Sudan, then ended up in prison due to a denunciation from a friend in Ukraine
"I can't take these cadences anymore." The wife of a Belarusian long-haul truck driver asked where her husband could find a job to be closer to home 55
Comments
Вельмі сумнеўнае абяцанне, з улікам таго, што праграмы і сайты Meta, у першую чаргу facebook, максімальна і бессаромна адсочваюць карыстальнікаў.
Настолькі, што расейскія праграмы і сайты толькі даганяюць.
Так што не веру, што будзе нейкі персанальны рэжым, у які яны не залезуць для сваёй карысці.