🔓 Open weights
camp_open_weights
Capability concentrated in a few closed labs is the structural harm. Model weights should distribute by default; gatekeeping via closed APIs is a worse failure mode than any misuse risk that open release introduces. Cross-cuts both the pause camp (rejects moratoria) and the closed-lab camp (rejects release gating).
Agents
- · Meta AI (Llama family)
- · Mistral AI
- · Hugging Face
- · EleutherAI
- · Yann LeCun (public advocacy)
- · a16z open-source AI advocacy
- · Together AI, Perplexity, and adjacent open-stack builders
Descriptive claims held (19)
- AI capability is accelerating along compute, data, and algorithmic axes.
- Algorithmic progress roughly halves the compute required to reach a fixed langu…
- Altman frames the deployment choice around AGI as a binary between individual e…
- China's Global AI Governance Initiative formally opposes 'drawing ideological l…
- Frontier AI performance scales with compute and capex.
- DeepSeek is reported to operate a fleet of approximately 50,000 Hopper-generati…
- DeepSeek-V3's training cost is on the expected algorithmic-efficiency cost-redu…
- Amortized hardware and energy cost of flagship training runs has grown ~2.4x an…
- GPU cloud compute is commoditizing toward per-hour metered pricing with low pri…
- Over 90% of leading-edge (<10nm, effectively 100% of <5nm) logic fabrication ca…
- In July 2024 Meta released Llama 3.1 405B as what it describes as the first fro…
- Meta claims developers can run inference on Llama 3.1 405B on their own infrast…
- At the Llama 3.1 launch, Meta coordinated partner support spanning cloud provid…
- Meta's stated business model does not depend on selling access to AI models, wh…
- Meta has publicly endorsed the use of its open-source Llama models for US natio…
- The Paris AI Action Summit Statement explicitly foregrounds 'open AI models in…
- Reinforcement-learning-based reasoning training (the second-stage paradigm behi…
- The US currently leads China in frontier AI by roughly 6-18 months.
- Zuckerberg argues that closing US frontier models to prevent Chinese access wil…