Mark Zuckerberg’s new goal is creating artificial general intelligence

Original Article ↗ Hacker News Discussion ↗

Meta’s shift from its metaverse ambitions to a declared long‑term goal of building and “responsibly” open-sourcing artificial general intelligence is prompting scrutiny of both its strategy and motives. Commenters note Meta’s massive GPU investments, strong AI research track record, and quasi–open source LLaMA models, but question whether branding future systems as AGI is realistic or mainly marketing. Many see the move as an attempt to commoditize core AI tech to undercut closed rivals like OpenAI and Google, while raising unresolved concerns about safety, bias, data usage, and the societal impact of powerful, widely available models.

Meta’s AGI Pivot and Mission

  • Thread notes a new “long‑term vision” to build and (responsibly) open‑source AGI, with internal AI groups being merged.
  • Some see this as goal‑post shifting from past “big bets” (metaverse, crypto, curing all disease), others as normal corporate pivoting enabled by huge cash flow.
  • Several commenters think the headline overplays “AGI” and see this mainly as investor narrative and talent magnet.

GPU Stockpile and Hardware Economics

  • Meta reportedly targeting ~340k H100s and ~600k “H100‑equivalents”; rough retail value discussed around $15B.
  • Debate over whether Meta gets major discounts vs GPUs being supply‑constrained; consensus that even with discounts it’s a massive capex.
  • Some call this “rapidly depreciating assets,” others note all compute depreciates and expect a large second‑hand market later.

Open Models, LLaMA, and “Open Source”

  • Many see Meta as the main driver of open(-ish) LLMs via LLaMA, enabling local models on consumer hardware.
  • Others argue the licenses are restrictive and calling them “open source” dilutes the term; “source‑available with conditions” is suggested as more accurate.
  • Supporters say training is the expensive part; releasing weights is analogous to open‑sourcing a heavily funded internal framework like React.

Business Strategy and “Commoditize Your Complement”

  • Widely discussed view: Meta can’t easily win closed‑source enterprise AI vs OpenAI/Google, so it tries to commoditize base models.
  • By releasing strong free models, Meta pressures competitors’ pricing, spreads Meta‑based stacks, and potentially strengthens its own platforms (social graph, VR).

Metaverse, VR/AR, and Data

  • Skepticism that metaverse vision has real traction; some see VR as a niche “gadget” with very low retention, citing past reports.
  • Others defend Quest 2/3 and Ray‑Ban smart glasses as commercially solid and technically impressive, especially with younger users.
  • One line of argument: Meta’s huge multimodal, egocentric, and behavioral datasets (social apps, headsets, glasses) may be a unique AGI asset, though data quality (vs textbooks, code, etc.) is questioned.

AGI, Scaling, and Risks

  • Disagreement on how close current LLMs are to AGI and what “AGI” should mean.
  • Scaling laws are cited as real for loss, but mapping lower loss to new capabilities is seen as uncertain.
  • Strong concern over alignment and “responsible” release: worries about misuse, mass misinformation, and the difficulty of constraining a powerful general system, especially if truly open.