activescott's Notes

ArduPilot provides a comprehensive suite of tools suitable for almost any vehicle and application. As an open source project, it is constantly evolving based on rapid feedback from a large community of users. The Development Team works with the community and commercial partners to add functionality to ArduPilot that benefits everyone. Although ArduPilot does not manufacture any hardware, ArduPilot firmware works on a wide variety of different hardware to control unmanned vehicles of all types. Coupled with ground control software, unmanned vehicles running ArduPilot can have advanced functionality including real-time communication with operators.

Installed in over 1,000,000 vehicles world-wide, and with advanced data-logging, analysis and simulation tools, ArduPilot is a deeply tested and trusted autopilot system.

The software suite is installed in vehicles from many manufacturers

function assertNever(x: never): never { throw new Error(Unexpected value: ${x}); }

In 2024, the Republicans narrowly retained control of the House with 220 seats to the Democrats' 215, the slimmest majority since 1930. That razor-thin margin is why the redistricting battle matters so much.

Every 10 years after the US census, the country’s 435 House seats are reapportioned based on population shifts, triggering a nationwide redrawing of congressional districts. The number of seats in the House has been 435 since 1912, and the only exception was made in 1929 to allow the admission of Alaska and Hawaii into the US.

The most recent census was in 2020, and states completed their redistricting by 2022. Since 2024, however, several states have redrawn their congressional maps again - some successfully, some blocked by the courts - with accusations of gerrymandering.

Gerrymandering is the practice of drawing electoral district boundaries to favour one political party over another. The tactic exists in many countries with district-based voting systems but is most closely associated with the US.

Florida’s new congressional map is expected to strengthen Republican control of the state’s 28 House seats and could help the party gain up to four additional Republican-leaning districts before the midterms.

Texas Republicans redrew congressional districts before the midterms as Democrats fled the state, leaving lawmakers missing during key votes. The new map added five more seats for the Texas Republicans.

Missouri Republicans redrew congressional maps before 2026, aiming to gain one extra House seat for their party.

In October, the North Carolina Senate approved a new congressional map that is expected to make one more US House seat Republican-leaning by reshaping several previously competitive or Democratic-leaning districts.

California voters approved a new Democratic-backed map under Proposition 50, known as the Election Rigging Response Act, in a 2025 special election.

The new boundaries are designed to help Democrats protect and potentially expand their existing 43-seat majority in the state.

Ohio was once considered a major presidential swing state, but Republicans now hold 10 of its 15 House seats, and the redraw is expected to further strengthen the party’s advantage.

The lawsuit also accused the Trump administration of trying to intimidate those who speak out against Israeli rights abuses.

Albanese is not alone in facing economic penalties for her work. Since taking office for a second term, Trump is estimated to have issued sanctions against nine ICC judges, as well as prosecutors for the court.

The judges and prosecutors were reportedly involved in probes into abuses by US and Israeli forces.

On May 13, US District Judge Richard Leon, an appointee of former President George W Bush, ruled in favour of the Albanese family’s lawsuit, granting a temporary injunction against the sanctions.

Leon found that the Trump administration had used the penalties to curtail Albanese’s constitutionally protected speech. He also stated that Albanese could not be blamed for the ICC’s actions.

“It is undisputed that her recommendations have no binding effect on the ICC’s actions,” Leon wrote. “They are nothing more than her opinion.”

As a result of the ruling, Albanese was removed from the sanctions list this month.

But the Trump administration appealed Leon’s order. It also said it would restore her to the sanctions list as soon as it was able, though it is unclear what prompted Wednesday’s change.

...these systems sit squarely in the center of a recent high-stakes battle between the US government and AI startup Anthropic. Anthropic is seeking to preserve two “red lines”: bans on domestic mass surveillance and on weapons that can identify, track, and kill targets with zero human involvement. Since the start of the year, it’s emerged as the only military AI contractor to place meaningful limits on what experts call one of the final frontiers of AI warfare.

At the center of the debates is DOD Directive 3000.09, one of the only policies governing the use of lethal autonomous weapons. Originally written in 2012, it defines such a system as one that, “once activated, can select and engage targets without further intervention by an operator.” And it decrees that both fully autonomous and semi-autonomous weapons be designed to allow humans to “exercise appropriate levels” of judgment over the use of force.

The directive set up the “first policy on the use of autonomy in warfare,” said Hamza Chaudhry, who leads AI and national security at the Future of Life Institute.

Depending on how you interpret the definition, however, certain missile defense programs may have crossed that line decades ago. Take the Phalanx CIWS, for instance. It’s an automated weapon system resembling a very large gun, built to defend naval vessels from incoming missile attacks. That type of system wouldn’t work if there were a human in the loop, since it has to respond in milliseconds.

The difference, some experts say, is that these systems operate solely in a defense-only, fixed environment. They’re engaging, this interpretation goes, but not deciding — just reacting to an incoming threat. “The ‘and’ is doing a lot of work inside of that statute — we have systems that can decide and systems that can engage but you can’t have a system that does both,” Reddie said.

They're also "killing" missiles not humans.

Google employees argued their company should take a stand — and it did, choosing not to renew its contract amid the controversy in mid-2018. But Amazon and Microsoft quickly swooped in to pick up tens of millions of dollars in contracts for the same work. Palantir soon took over, and Project Maven became the Maven Smart System (MSS), which not only allows for object detection and tracking but also analyzing surveillance data on a large scale.

The sheer volume of targets could make any meaningful human supervision difficult, said Shoker. “What we know about MSS is that it reduces the number of human beings in the targeting cycle — and that’s actually by design.”

While Anthropic might have been all right reducing human intervention, it’s pushed back against setting it to zero. As Google found with Project Maven, though, competitors are more than willing to fill the gap.

OpenAI quickly signed onto the terms Anthropic had spurned. And in the months after snubbing Anthropic, the Department of Defense signed deals with eight companies to deploy their AI on classified networks: Google, Microsoft, Amazon Web Services, Nvidia, OpenAI, Reflection, Oracle, and SpaceX.

Silicon Valley executives are aggressively pushing back against employee organizing and speaking out, including by using AI to identify leakers. And many tech workers already fear for their jobs in an era when AI is set to replace entry-level roles at their own firms.

Anthropic CEO Dario Amodei has held firm on mass surveillance for Americans, but he’s demonstrated no problem with — and in fact expressed his support for — such surveillance for everyone else.

Anthropic’s “very narrow” red lines “do not go far enough to protect human rights or to comply with international law,” said Tech Justice Law’s Batt. “Anthropic specifically talks about mass domestic surveillance of US persons as posing grave civil liberties concerns, but the same civil liberties concerns apply with equal force to non-US persons,” she added.

In a blog post, he said that “fully autonomous weapons (those that take humans out of the loop entirely and automate selecting and engaging targets) may prove critical for our national defense.” Amodei even said he was happy to “work directly with the Department of War on R&D to improve the reliability of these systems” and speed up the timeline for the company’s help in deploying them.

A panel of federal judges blocked Alabama from using Republican-supported congressional district maps that would dilute the votes of Black people in November’s midterm elections.

The ruling in U.S. District Court in Birmingham, Alabama, which found that the maps “intentionally discriminated based on race,” sets the stage for the U.S. Supreme Court to determine whether the maps, which were first proposed in 2023, can be used by Alabama this year.

Earlier this month, the Virginia Supreme Court blocked maps for Democratic-leaning congressional districts in that state, which had been approved in a statewide referendum in April.

Republicans last year began a series of congressional redistrictings in an effort to retain their ultra-thin majority in the House. Florida Gov. Ron DeSantis on May 4 signed a law creating a new congressional map that is projected to help Republicans add control of four House districts from the state.

Two judges on the panel were appointed by President Donald Trump: Anna Manasco and Terry Moorer. The third judge, Stanley Marcus, was first nominated to a federal district court by President Ronald Reagan and then was nominated to the 11th Circuit U.S. Court of Appeals, where he currently sits, by President Bill Clinton.

“We do not lightly intrude in state affairs, but our previous review of the undisputed evidence left us in no doubt that Alabama’s legislatively enacted plan (the ’2023 Plan’) intentionally discriminated based on race in violation of the Constitution,” the panel said. “Our re-examination in light of Callais yields the same conclusion. We again cannot understand the 2023 Plan as anything other than intentionally discriminatory.”

a presentation that the project’s developer gave to the Washington State Parks Commission on Oct. 18, 2023. For the full presentation, see here. (Westport Golf Links, Inc.)

In the early 2000s, Mox Chehalis LLC had plans to develop land owned by the Port of Grays Harbor into a golf resort. 

The plans, which included a hotel and convention center, were met with resistance from Friends of Grays Harbor and other environmentalists, triggering a legal battle that dragged on for about six years.

In 2007, Friends of Grays Harbor and Mox Chehalis LLC signed a settlement agreement that limited the ability of the current and future owners to fill wetlands on the property for golf course development.

Further inspection of the site by the U.S. Army Corps of Engineers found that work carried out by Mox Chehalis included filling wetlands in violation of the Clean Water Act.

To avoid future violations of the federal law, the Army Corps and Mox Chehalis formed a land covenant that, among other limitations, prohibited filling wetlands, clearing vegetation, changing the topography, and constructing buildings on about 114 acres of the land.

Washington purchased the property from Mox Chehalis and J.D. Financial Corp, which shared common ownership, in 2015 using a grant from the State Recreation and Conservation Office.

The environmental groups now suing to halt the project argue the state is subject to the restrictions on development in the settlement agreement and land covenant. 

And they say that Washington law prohibits the state from converting the property into a golf course because of how it was acquired with a Recreation and Conservation Office grant.

In March 1919, Benito Mussolini founded the first Italian Fasces of Combat (FIC) at the beginning of the so-called Red Biennium, a two-year long social conflict between the Italian Socialist Party (PSI) and the liberal and conservative ruling class. Mussolini's Fascists suffered a defeat in the election of November 1919, winning no seats in the Italian parliament.

During the "two red years", there were numerous strikes, protests against rises in the cost of living, occupations of factories and land by industrial workers or agricultural laborers, and other types of clashes between socialists on one side and landowners and business owners on the other side.

Local elites felt themselves vulnerable and established an alliance with the small Fascist movement, which contained many veterans of World War I and had a reputation for violence, in the hope of using Fascist paramilitary squads to destroy socialist organizations.

Since 1919, Fascist militias, known as squadristi or "Blackshirts" due to their uniforms, had frequently attacked socialist politicians and militants. In August 1920, the Blackshirt militia was used to break the general strike which originated at the Alfa Romeo factory in Milan

Local elections in 1920 were won by the socialists in many towns, cities and villages across Italy, and in response Fascist militias attacked union organizers and municipal administrators, making it difficult for local governments to function.

A local deputy from the town of Budrio sent a telegram to the prime minister in October 1921 to report that the Fascists had effectively taken over, that "unions and socialist clubs [were] ordered to dissolve themselves within 48 hours or face physical destruction" and that the "life of the town is paralysed, authorities impotent".

within the National Blocs of Giovanni Giolitti, an anti-socialist coalition of liberals, conservatives and fascists. The Fascists won 35 seats and Mussolini was elected in the Parliament for the first time.

After a few weeks, Mussolini withdrew his support for Giolitti and his Italian Liberal Party (Partito Liberale Italiano, PLI) and attempted to work out a temporary truce with the Socialists by signing the so-called "Pact of Pacification" in the summer of 1921.

the Pact with the Socialists was nullified during the Third Fascist Congress on 7–10 November 1921, during which Mussolini promoted a nationalist program and renamed his movement National Fascist Party (PNF), which enrolled 320,000 members by late 1921.

In August 1922, an anti-fascist general strike was organized throughout the country by the socialists. Mussolini declared that the Fascists would suppress the strike themselves if the government did not immediately intervene to stop it, which enabled him to position the Fascist Party as a defender of law and order.[13] On 2 August, in Ancona, Fascist squads moved in from the countryside and razed all buildings occupied by socialists.[13] This was then repeated in Genoa and other cities.[13]

Then, with the support of local business owners, they took over local government and expelled the elected socialist administration from the town hall

The Italian national government in Rome did nothing to react to these developments, and its inaction prompted Mussolini to plan a march on Rome.[13] From their new power base in Milan, the Fascists gathered the financial support of large companies who were determined to fight against "strikes, bolshevism and nationalization".

Also a few days before the march, Mussolini consulted with the U.S. Ambassador Richard Washburn Child about whether the U.S. government would object to Fascist participation in a future Italian government and Child gave him American support.

Squadrismo (Italian: [skwaˈdrizmo]) was the movement of squadre d'azione (English: action squads), the fascist militias that were organised outside the authority of the Italian state and led by local leaders called ras (a noble Ethiopian title). The militia originally consisted of farmers and middle-class people, who created their own defence from revolutionary socialists. Squadrismo became an important asset for the rise of the National Fascist Party, led by Benito Mussolini, and systematically used violence to eliminate any political parties that were opposed to Italian fascism.

The violence was not only an instrument in politics but also a vital component of squadrismo identity, which made it difficult for the movement to be tamed. That was shown in the various attempts by Mussolini to control squadrismo violence with the Pact of Pacification and later the Consolidated Public Safety Act. Squadrismo, which ultimately became the Blackshirts, served as a source of inspiration for Adolf Hitler's Sturmabteilung.

The Blackshirts, formally established as the Squadrismo in 1919, comprised numerous disgruntled demobilized soldiers. It was given the task of leading fights against their bitter enemies—the Socialists.

The Department of Justice is acknowledging it has removed from its website news releases about criminal cases related to the 6 January 2021 Capitol attack, calling the information about the prosecutions “partisan propaganda”.

The purge of news releases documenting criminal charges, convictions and sentencings is the latest step by the Trump administration to dramatically rewrite the history of the assault on the US Capitol, when hundreds of supporters of Donald Trump stormed the building in an effort to halt the congressional certification of his 2020 election loss to Democrat Joe Biden.

Trump, on his first day back in office in January 2025 , pardoned, commuted the prison sentences or vowed to dismiss the cases of all of the 1,500-plus people charged with crimes during the Capitol assault, including those convicted of attacking officers with makeshift weapons such as flagpoles, a hockey stick and a crutch.

Trump dismisses $10bn suit against IRS and creates $1.7bn ‘anti-weaponization’ fund Read more On Monday, the justice department announced the creation of a $1.776bn fund meant to compensate Trump allies who feel they were unjustly investigated and prosecuted. The acting attorney general, Todd Blanche, has not ruled out that rioters convicted of violence will be eligible for payouts, prompting bipartisan anger in Congress.

After a journalist on Friday observed on the social media platform X that the justice department was “quietly” removing news releases on its website that were related to the January 6 attack, including about a Texas man who pleaded guilty to assault and also faced separate state charges of soliciting a minor, the department responded through its “rapid response” account that there was “nothing ‘quiet’ about it”.

Among the releases removed from the site were those concerning seditious conspiracy cases against members of the Proud Boys and the Oath Keepers, far-right extremist groups. The justice department, in an unopposed motion last month, asked a federal appeals court to vacate those seditious conspiracy convictions, a request that was granted on Thursday. The department on Friday moved to dismiss the cases against the group members.

SealQA, a new challenge benchmark for evaluating SEarch-Augmented Language models on fact-seeking questions where web search yields conflicting, noisy, or unhelpful results. SealQA comes in three flavors: (1) Seal-0 (main) and (2) Seal-Hard, which assess factual accuracy and reasoning capabilities, with Seal-0 focusing on the most challenging questions where chat models (e.g., GPT-4.1) typically achieve near-zero accuracy; and (3) LongSeal, which extends SealQA to test long-context, multi-document reasoning in "needle-in-a-haystack" settings. Our evaluation reveals critical limitations in current models: Even frontier LLMs perform poorly across all SealQA flavors. On Seal-0, frontier agentic models equipped with tools like o3 and o4-mini achieve only 17.1% and 6.3% accuracy, respectively, at their best reasoning efforts. We find that advanced reasoning models such as DeepSeek-R1-671B and o3-mini are highly vulnerable to noisy search results. Notably, increasing test-time compute does not yield reliable gains across o3-mini, o4-mini, and o3, with performance often plateauing or even declining early.

benchmarks are not keeping pace in difficulty: LLMs now achieve over 90% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam, a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. The dataset consists of 2,500 challenging questions across over a hundred subjects. We publicly release these questions, while maintaining a private test set of held out questions to assess model overfitting.

Accuracy. All frontier models achieve low accuracy on Humanity's Last Exam, highlighting significant room for improvement in narrowing the gap between current LLMs and expert-level academic capabilities on closed-ended questions.

Hahaha current state of the art is Gemini 3 Pro w/ 38.3:

Given the rapid pace of AI development, it is plausible that models could exceed 50% accuracy on HLE by the end of 2025.

Data and code for our paper FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation.

We believe that human evaluators possess the expertise and common sense required to detect issues like hallucinations, making them more reliable than automated evaluation metrics for assessing LLMs' factuality. However, researchers have the flexibility to adjust their evaluation methods if human evaluation proves challenging. An easily implemented alternative is to use standard metrics like F1/exact match or recall, which assess the overlap between the model response and the ground truth answer(s) (e.g., see You.com's recent blog where they report FreshQA recall). Researchers can also use LLM-based automatic evaluation metrics such as FactScore or our FreshEval metric below.

To facilitate future evaluations, we have developed FreshEval, a simple automatic metric that uses few-shot in-context learning to teach an LLM to judge model responses, which achieved high agreement with human raters (see Appendix B in our paper for details).

To use FreshEval under a specific evaluation mode (Relaxed or Strict), please follow the instructions below:

Make a copy of our latest data spreadsheet and store it in your Google Drive with a new filename (e.g., fresheval_relaxed or fresheval_strict).
Insert 3 new columns D, E, F in the new spreadsheet for model responses, evaluation rating, evaluation explanation, respectively and save your model's responses in column D (see our sample evaluation spreadsheet below).
Run the associated FreshEval notebook with the evaluation mode. Note that for demonstration purposes, we evaluated only the first 10 model responses. You can adjust the number as needed.

Note: Currently, we recommend gpt-4-1106-preview over gpt-4-0125-preview for FreshEval as it yielded slightly better agreement with human annotations in our small-scale evaluation.

we perform a detailed study of the factuality of LLM-generated text in the context of answering questions that test current world knowledge. Specifically, we introduce FreshQA, a novel dynamic QA benchmark encompassing a diverse range of question and answer types, including questions that require fast-changing world knowledge as well as questions with false premises that need to be debunked.

Through human evaluations involving more than 50K judgments, we shed light on limitations of these models and demonstrate significant room for improvement: for instance, all models (regardless of model size) struggle on questions that involve fast-changing knowledge and false premises.

we present FreshPrompt, a simple few-shot prompting method that substantially boosts the performance of an LLM on FreshQA by incorporating relevant and up-to-date information retrieved from a search engine into the prompt. Our experiments show that FreshPrompt outperforms both competing search engine-augmented prompting methods such as Self-Ask (Press et al., 2022) as well as commercial systems such as this http URL. Further analysis of FreshPrompt reveals that both the number of retrieved evidences and their order play a key role in influencing the correctness of LLM-generated answers. Additionally, instructing the LLM to generate concise and direct answers helps reduce hallucination compared to encouraging more verbose answers.

FRAMES offers a unified framework for assessing LLM performance in end-to-end RAG scenarios. Our dataset comprises challenging multi-hop questions requiring integration of information from multiple sources. Baseline results show that even state-of-the-art LLMs struggle, achieving 0.408 accuracy without retrieval. However, our proposed multi-step retrieval pipeline significantly improves accuracy to 0.66 (>50% improvement).