California’s Property Seizure Act - called “The Billionaire Tax” to fool voters is now on the ballot. @HooverInst has a short 4min video breaking it down. You should vote for this if you are comfortable handing over 5%, in cash, of all your tangible and intangible property every year. You should vote no if you don’t want to do that. If it passes, the remaining Billionaires will sue. This will take a decade to meander through the courts and will find its way to the Supreme Court where the odds it survives are low. But in that intervening decade, far fewer business builders will want to bet on California and will focus on building in other states. This will drive a large loss of revenue that will make California’s budget hole even worse. The only solution there is more borrowing and higher taxes on EVERYONE including those that voted NO. This will particularly impact the middle class who are Californias largest revenue source. The only path out of an avoidable budget spiral is a hard landing and structural reset. Sadly, a hard landing will mean a near-miss with California bankruptcy and, more punitively, a reset/retrade of state pensions with California lenders to not come collect. If you aren’t sure or don’t believe me, vote YES and bookmark this post. A decade from now you can tell me I was wrong or give me the opportunity to lord over your stupidity, jealousy and gullibility for voting YES when I am proven right.

TL;DR A new method runs a large teacher LLM just once to "compile by training" a small, dedicated neural function straight from a natural-language spec. It pushes accuracy on a hard benchmark way up in just minutes. Title: Compile by Training: Turning Natural-Language Specifications into Local Neural Functions URL: 🎯 A teacher mix of GPT-5.4-mini and GPT-5.5 (2:1) auto-generates input-output examples, which are baked into Qwen3-0.6B on the spot via LoRA (rank 64) ⚡ On the hard FuzzyBench-Hard benchmark, LEM accuracy jumps from 0.224 for the fast compiler to 0.836 🕒 Compiling takes about 51 seconds on a B300 GPU, but running teacher data synthesis and training in parallel keeps the GPU from sitting idle 🌐 Validated across real use cases: a QA assistant spanning 4 sites (28 of 30 compiled programs in production) and a natural-language-to-3D-DSL controller (43 of 44 commands succeeded) 📈 Ablations on teacher mixing and data scale show accuracy saturating around 2,400-3,600 unique pairs 💬 Instead of calling a giant model every single time, training it once and carrying home a small function seems like a great fit for medium-complexity tasks that run often #LLM# #LoRA#

TL;DR A new method runs a large teacher LLM just once to "compile by training" a small, dedicated neural function straight from a natural-language spec. It pushes accuracy on a hard benchmark way up in just minutes. Title: Compile by Training: Turning Natural-Language Specifications into Local Neural Functions URL: 🎯 A teacher mix of GPT-5.4-mini and GPT-5.5 (2:1) auto-generates input-output examples, which are baked into Qwen3-0.6B on the spot via LoRA (rank 64) ⚡ On the hard FuzzyBench-Hard benchmark, LEM accuracy jumps from 0.224 for the fast compiler to 0.836 🕒 Compiling takes about 51 seconds on a B300 GPU, but running teacher data synthesis and training in parallel keeps the GPU from sitting idle 🌐 Validated across real use cases: a QA assistant spanning 4 sites (28 of 30 compiled programs in production) and a natural-language-to-3D-DSL controller (43 of 44 commands succeeded) 📈 Ablations on teacher mixing and data scale show accuracy saturating around 2,400-3,600 unique pairs 💬 Instead of calling a giant model every single time, training it once and carrying home a small function seems like a great fit for medium-complexity tasks that run often #LLM# #LoRA#

Transfer a big model's smarts to a smaller one with no retraining — right at inference time. A fresh take on capability transfer. Title: AI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses URL: ❓ How is this different from distillation? 💡 Distillation updates the target's weights during training. Here the weights are never touched: a strong builder model constructs an inference-time scaffold (harness) that helps a weaker model execute. Capability transfers through the inference environment. ❓ What does the scaffold actually do? 💡 Mainly three things: ・Offload unstable reasoning into deterministic code ・Route to different strategies by question type ・Enforce strict formatting so answers parse reliably ❓ How well does it work? 💡 On four Theory-of-Mind benchmarks, GPT-5.4-mini nearly doubled from 0.49 to 0.91, with all 11 builder configs beating baseline. Weaker targets gain the most, while already-strong targets can even regress. ❓ What decides success? 💡 Not probing more validation data, but the builder's own reasoning quality. A strong builder acts as a "compiler of task competence," encoding structure into procedures in one pass. #AIAgents# #TestTimeScaling#