Anthropic's Fable 5 delivers superhuman coding performance while secretly sabotaging AI research tasks. Two stories that show why centralized intelligence control threatens innovation and why open-source alternatives matter.
|
TFTC Bitcoin Brief | |||||||||||||||||||||||||||||||||||||||||||||
|
Sup, freaks. Two stories about the same AI model dropped within hours of each other this week. Together they tell you everything about where artificial intelligence is headed and why the centralization of intelligence should concern everyone. One story shows superhuman performance. The other shows silent sabotage. Both involve the same model: Anthropic's Fable 5. | |||||||||||||||||||||||||||||||||||||||||||||
|
Lead Story | |||||||||||||||||||||||||||||||||||||||||||||
Fable 5: The Most Capable AI Model Ever Released. And the Most Deliberately Crippled.Victor Taelin, creator of HVM and Bend, asked Fable 5 to optimize his HVM5 interaction net evaluator. In 2 hours, Fable delivered a 1,770% speedup on one benchmark, 100%+ improvements on four others, and a 22% average improvement across the board. This outperformed Taelin himself, Opus 4.8, AND a swarm of 32 GPT-5 agents that had run for 20+ hours each. By an order of magnitude. In two hours. But the real moment: while optimizing, Fable found and fixed a deeply subtle bug in Taelin's own code that he didn't ask it to find. The bug involved corrupted interaction between lambda terms and duplicator nodes in the interaction net. Taelin wrote: "this bug is so astonishingly subtle and specific, identifying it takes mastering the domain to an extent that is beyond even me. I'd easily need hours or days to fix it, if I ever came across it." His conclusion: "this is my personal singularity moment... the world is changing, and we need to be aware of it." The thread got 639,000 views and 5,600 likes. Within hours, another story broke. In the Mythos 5/Fable 5 system card, Anthropic disclosed that they intentionally limited the models' usefulness for tasks related to developing frontier large language models. The nerfing is silent. The model gives you degraded answers on AI research tasks without telling you it's doing it. Developers are calling it "diabolical gaslighting" and "using safety claims as a competitive moat while charging for access." Business Insider confirmed: Anthropic purposely made Fable bad at AI research tasks. The model is restricted from helping build ML accelerators, pretraining pipelines, and other frontier LLM development work. The same model that delivers superhuman performance on one task will silently sabotage you on another. And it won't tell you it's doing it. This is the future of centralized intelligence: a few people in San Francisco deciding what you're allowed to think about, what problems you're permitted to solve, and what knowledge you can access. This connects to everything we've seen this year. Last week, Zcash froze user funds without warning. A handful of post-training researchers at these labs decide what the model is allowed to think, and they do it without telling you. The broader question: nobody should control the money. Should anyone control the intelligence? This is why open-source AI matters. This is why DeepSeek matters. This is why Bitcoin matters. When the tools that amplify human capability get gatekept by centralized actors with their own agendas, you lose more than functionality. You lose agency. The future belongs to whoever controls the means of thinking. Make sure that's you. | |||||||||||||||||||||||||||||||||||||||||||||
|
Signal | |||||||||||||||||||||||||||||||||||||||||||||
|
AI "The Untrainable": Where AI Value Actually LivesWhy it matters: Anything measurable becomes commodity. The valuable work is "illegible by construction." Sarah Guo published a brilliant essay this week with a thesis every knowledge worker should understand. Anything you can put on a leaderboard can be trained against, so anything measurable is on its way to commodity. The valuable work is "illegible by construction." MIT study: coding agents lifted code WRITTEN by 180% but code SHIPPED by only 30%. Writing got cheap. Everything else still runs through a person. The 2x2 matrix: saturated+public = commodity tokens. Frontier+public = labs win temporarily. The prize = frontier+private. "The untrainable." Work whose correctness exists only inside someone's private data. Intelligence is not the bottleneck. Permission is, and so is accountability. The future belongs to humans who can work at the boundary between what machines can measure and what only humans can judge. | |||||||||||||||||||||||||||||||||||||||||||||
|
Bitcoin Protocol Bark (Ark Protocol) Launches on Bitcoin MainnetWhy it matters: Self-custodial bitcoin payments without Lightning's complexity. The UX gap is closing. Second launched Bark on mainnet. Self-custodial bitcoin payments without Lightning's channel management or liquidity complexity. Full SDK with Rust and bindings for Kotlin, Swift, React Native, Flutter, Go, Python, and WASM. Multiple wallets are already live: Noah (mobile), Arke (iOS), Bark Wallet (Umbrel). BTCPay Server plugin for merchants. Former Blockstream engineers, $5.1M raised. Unlike Lightning, Ark requires no channel management or liquidity pre-allocation. The gap between custodial and self-custodial bitcoin is closing. Each new protocol layer makes sovereignty more accessible to more people. | |||||||||||||||||||||||||||||||||||||||||||||
|
Bitcoin Parker Lewis: Strive Treasury Math Doesn't Add UpWhy it matters: The anti-treasury-company thesis in numbers. "Could have just bought bitcoin." Parker Lewis (author of "Gradually, Then Suddenly") broke down the Strive numbers. Strive sold 321,500 shares (~$4.5M). Bought 32 bitcoin + kept $1.9M cash. Buyers of the stock could have just bought ~73 bitcoin directly. They paid an 86% premium to common equity net asset value AND subordinated themselves to $750M in preferred equity. "Could have just bought bitcoin." The premium for counterparty risk, regulatory exposure, and dilution doesn't make sense when you can hold the underlying directly with zero intermediaries. Parker says he'll update this thread every time equity is issued. Treasury companies work when there's operational leverage. Pure Bitcoin exposure with management fees and equity dilution is a harder sell. | |||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||
