Get this creature out of power by any means necessary.

Very amusing that banks, most of which are too stupid security wise to embrace TOTP instead of SMS short codes, are first partners for the Mythos fabulous security AI. Mythos is scary because it is so good at finding security exploits that those that control it can literally crack into most anything in our commmunication and computational environment. Anything. I am sure NSA and its equivalents are salivating. On April 7, 2026, Bloomberg reported that OpenAI, Google, and Anthropic have begun sharing intelligence through the Frontier Model Forum to detect and combat adversarial distillation -- the technique by which Chinese AI competitors extract structured knowledge from proprietary U.S. frontier models to train competitive smaller models, effectively compressing years of research investment into weeks of supervised fine-tuning (Bloomberg / Gadgets 360). Distillation in its authorized form is a well-established compression technique: a smaller "student" model is trained on the output distribution of a larger "teacher" model, learning to replicate its behavior at reduced computational cost. In adversarial distillation, this is done without authorization and at scale, systematically querying frontier APIs for specific prompt-response pairs designed to maximize knowledge transfer. OpenAI first accused DeepSeek of this practice following the DeepSeek-R1 release in early 2025, which showed remarkable reasoning capability at a training cost that was widely considered implausibly low for independent development. The three companies characterize these extraction attempts as both a commercial theft (estimated at billions in annual profit impact) and a national security risk, since the models being distilled are trained under export-controlled hardware (Bloomberg / Gadgets 360).

Looks to a lot like Big AI protecting its ass from legitimate bright use to build the next generation of AI. Or don't they believe in recursive improvement unless they control. In any case hard to distinguish from manufactured scarcity in my book. On April 7, 2026, Bloomberg reported that OpenAI, Google, and Anthropic have begun sharing intelligence through the Frontier Model Forum to detect and combat adversarial distillation -- the technique by which Chinese AI competitors extract structured knowledge from proprietary U.S. frontier models to train competitive smaller models, effectively compressing years of research investment into weeks of supervised fine-tuning (Bloomberg / Gadgets 360). Distillation in its authorized form is a well-established compression technique: a smaller "student" model is trained on the output distribution of a larger "teacher" model, learning to replicate its behavior at reduced computational cost. In adversarial distillation, this is done without authorization and at scale, systematically querying frontier APIs for specific prompt-response pairs designed to maximize knowledge transfer. OpenAI first accused DeepSeek of this practice following the DeepSeek-R1 release in early 2025, which showed remarkable reasoning capability at a training cost that was widely considered implausibly low for independent development. The three companies characterize these extraction attempts as both a commercial theft (estimated at billions in annual profit impact) and a national security risk, since the models being distilled are trained under export-controlled hardware (Bloomberg / Gadgets 360).

Gemma 4 run under nano-gpt (no KYC, pay with cripto) is freaking amazing. I have better interactions with it for general chat than with Claude AI. I am very impressed. And nano-gpt is anonymous! What is not to like?

When one house of Congress won't even vote on such a thing it is not a functioning democracy. House Republicans poised to reject Senate-passed DHS funding bill, likely extending standoff House conservatives oppose Senate bill funding DHS except ICE, risking continued airport delays as over 11% of TSA workers miss shifts amid a 42-day funding stalemate. On Friday, Speaker Mike Johnson and House Republican leadership declared they will not vote on the Senate's DHS funding bill, which excludes funding for Immigration and Customs Enforcement and Customs and Border Protection. Senators approved the partial funding package by unanimous consent early Friday before starting a two-week recess, leaving the House to resolve the 42-day shutdown that has caused significant airport delays and staff resignations. House Conference Chair Lisa McClain called the plan "garbage," demanding the restoration of ICE and Border Patrol funding alongside voter identification provisions before any House vote occurs. President Donald Trump announced plans to pay TSA agents using 2025 tax bill funds to stem airport "Chaos at the Airports," though the stalemate persists amid the funding impasse. Sixty-Day funding measures are being considered as the legislative impasse leaves the department's operations uncertain, with federal workers facing potential continued financial hardship during the ongoing government shutdown.

Supreme Court Blocks Music Industry's Push to Cut Millions Off the Internet Over Piracy Claims ============================================================================================== The Supreme Court has unanimously refused to let the recording industry turn internet providers into copyright enforcers with the power to cut millions of people off from modern life. The ruling, handed down Wednesday in Cox Communications v. Sony Music Entertainment, kills a legal theory that would have given ISPs one rational choice when they received a copyright complaint: sever the connection first, figure out the truth later.