'Internet Compression as a Service': are LLMs erasing the long tail of knowledge?
Conscious-content42 · reddit · 2026-08-26
A Reddit user proposes ICaaS (Internet compression as a service) as a framing: model providers hoover up and compress the internet into single responses, potentially accelerating the loss of the long tail of perspectives — an effect that compounds as AI output flows back into training data.
The author isn't anti-model (they use them daily for engineering, finance, cooking) but asks the community how to support long-tail knowledge against this compaction. Internet Archive and Wikipedia help but carry their own biases about what gets collectively remembered; some users have even returned to older models as less contaminated by aggregated AI responses.
More from AGI Musings
- Terence Tao: Autoformalization still costly but rapidly improving — littmath · 2026-08-27
- Terence Tao: Use AI to produce better work, not just more PDFs — littmath · 2026-08-27
- Teaching students to keep their judgment when working with AI: the real frontier of AI education — DrKavner · 2026-08-27
- Phantom traffic jam experiment offers a grim model for AI progress — robleclerc · 2026-08-27
- Who can overrule your AI when it's confidently wrong? — YvesMulkers · 2026-08-27
- Brands are getting impressions in AI answers without the click — nikvassev · 2026-08-27