Unverified50% confidenceOpinionExact time
Meta's social nature makes the types of data it collects richer; posts, comments, images, and public profile information can all become training material for multimodal models.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedMeta在Llama系列最新迭代中开始引入信息密度作为训练目标之一68% similarUnverifiedThe manipulation method is not to directly train LLMs, but to flood the internet with SEO blog content so that data used for future LLM training is saturated with specific narratives.68% similarUnverifiedThe vast majority of large model training data comes from human publicly available knowledge such as web text, books, code, and forum discussions.67% similarUnverifiedMeta's new terms text shows that Meta can use external content for 'development, improvement, research' as well as 'training', where the word 'training' points directly to AI model training scenarios.65% similarUnverifiedMeta has open-sourced the fairscale distributed training library and the torchtune model fine-tuning tool.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1056797API
curl https://kongchang.com/api/v1/knowledge/claims/1056797MCP
get_claim(id=1056797)