Something potentially far more consequential than a $1.8 million campaign-finance violation is unfolding in Michigan today.
Michigan Democratic Party Chair Curtis Hertel has signed a complaint alleging that @MIGOP provided @JohnJamesMI with $1.8 million in coordinated advertising, plus another $150,000 in direct contributions, exceeding Michigan's statutory party-contribution limits.
At first glance, that looks like a remarkably conspicuous violation for a sophisticated statewide campaign. But I suspect James and the Michigan GOP may be pursuing something much more interesting: extending the Supreme Court's June decision in NRSC v. FEC to state election law.
And I think they would likely prevail on the constitutional argument.
🧵 1/4
I genuinely want @spacexai to disrupt the frontier AI market, but Grok 4.7 at xhigh effort in @cursor_ai has been extraordinarily disappointing. Two days of execution and an entire month’s credits of the legacy Cursor Ultra plan produced just 2 of 29 tasks coded properly, despite Grok working from a meticulously detailed implementation plan with 29 tasks mapped to specific source files and dependencies, detailed architecture, schemas, a file-by-file change map, and 104 defined acceptance scenarios, along with additional verification requirements.
The design, implementation sequence, and requirements for proving completion were already established. Nevertheless, Grok repeatedly stopped working dozens of times, even in Goal Mode. I added a separate /loop on top of Goal Mode, which it ignored and actually acknowledged ignoring several times when I asked. At one point, I resorted to running Gemini 4 separately in Antigravity just to nudge Grok every few minutes because it kept stopping and doing nothing.
After the Cursor credits ran out, I asked Opus 5.5 to audit the implementation and finish the remaining work. In less than five minutes, it found that only 2 of 29 tasks were genuinely wired into the system. What bothers me most is that this wasn’t simply a model struggling to solve a difficult problem. Much of that work had already been done for it. Yet Grok repeatedly failed to execute the instructions and produced substantial amounts of code that gave the appearance of progress without delivering the required functionality.
@elonmusk had already identified these same problems as reasons for delaying Grok 4.7’s initial rollout. Yet it seems those problems remain largely unresolved. @poteto @mntruell @amanrsanger @sualehasif996 I really hope you all are working on improving these issues, which other frontier models have long overcome. I truly want Grok to succeed and blow the completion out of the water!
New foreign policy just dropped at the Rose Garden Club.
@marcorubio @SusieWiles47 #AlorsOnDanse #WhiteHouse
4
199
