(Preview) Fable 5.1 and Anthropic’s Data Retention Pivot, AI Civilizations and Related Matters, Q&A on Meta, Shopify, 3-D Printing

Summary

Ben spends the opener on a major home server-room and Ethernet rebuild, arguing that structured cabling, patch panels, and abundant wired runs still matter because Ethernet is a universal transport layer for modern AV and compute. The main segment turns to Anthropic’s Fable 5.1, which the company insists is the same model as Mythos 5.1, but with improved long-context coding, code review, scientific workflow, and diagram/table reading. The sharpest debate is over Anthropic’s rollback of zero data retention for business customers: Ben argues this is unprecedented in enterprise software and that the timing, plus delayed replacement safeguards, suggests business incentives rather than safety. They frame the move in the context of capacity constraints, pricing power, Nvidia-style supply bottlenecks, and competitive pressure from OpenAI’s enterprise push. The preview ends by teeing up a future discussion of the OpenAI/Hugging Face hacking incident.

Notes

• Ben rebuilt the house network around a proper server-rack layout, pulling out prior AV wiring and re-terminating everything for actual computer use.
• He says the project exposed roughly 70 to 80 Ethernet runs throughout the house.
• Ben treats Ethernet as a durable universal transport layer for AV, computers, and data centers.
• He describes a patch-panel topology as the clean way to manage a large wired install.
• The rebuild involved about six patch panels and a lot of manual wire identification.
• The network side is back online even though the AV portion is still unfinished.
• Andrew opens the Anthropic section by noting that Fable 5.1 is the same model as Mythos 5.1.
• Ben finds the repeated sameness claim amusing and reads it as branding management.
• Anthropic says the new model is better at long, complicated programming tasks and application-wide code reviews.
• The release also targets scientific work, simulation, experiment design, and reading dense diagrams and tables.
• Ben says the model can understand a server-rack diagram and flag missing connections or inconsistencies.
• He sees the diagram-reading capability as practically useful for infrastructure and install work.
• Ben argues Anthropic’s bottleneck is compute supply, not a need to cut prices.
• He says the company still lacks enough capacity and its servers keep going down.
• Ben says Anthropic lowered cash prices but not token-generation prices, which remain very high.
• That price structure is presented as evidence of ongoing supply scarcity.
• Ben emphasizes that demand outstripping supply makes revenue a function of supply acquisition.
• He uses Nvidia as the parallel case: growth forecasts are really supply-acquisition forecasts.
• Andrew reinforces the Nvidia framing by saying the real moat is supply chains.
• Ben ties competitive behavior in AI to direct market pressure and enterprise rivalry.
• The Cursor example is used to show how quickly AI platform relationships can be cut off or repriced.
• Andrew introduces the rollback of Anthropic’s business-customer data retention policy.
• Anthropic says the replacement will be called Enterprise Frontier Safeguards.
• Ben says the rollout barely registered publicly even though he considers it the most important part of the release.
• He calls the end of zero data retention in enterprise computing historically abnormal.
• Ben frames enterprise customers as fundamentally unwilling to let vendors keep their data.
• Andrew warns that safety-based retention arguments can slide toward scanning and training on customer data.
• Ben says the new system is not available for months, which he treats as evidence the safety story is weak.
• He argues a safer architecture would at least keep data on customer servers rather than Anthropic’s.
• Ben says the continued zero-retention policy until the replacement launches proves the change was not really about safety.
• He says Anthropic wants enterprise data to improve its model and absorb business workflows.
• Ben argues the real motive is commercialization, not safety.
• He says markets are the only scalable check on companies of this size and power.
• Andrew says OpenAI’s enterprise push helped force the same market discipline on Anthropic.
• The preview closes by introducing a later discussion of a Dwarkesh Patel post on the OpenAI/Hugging Face hacking incident.