☕ Which AI model is the best?

Good morning!

Some good news for your Tuesday: a rubber duck race known as the ‘Chicago Ducky Derby’ has raised almost $400,000 for Special Olympics Illinois. The not-for-profit aims to “transform the lives of people with intellectual disabilities,” empowering them “to realise their full potential in sports and in life.”

Its annual Ducky Derby attracted more than 80,000 participants this year, who each paid $10 to ‘adopt’ a rubber duck. Special Olympics Illinois thanked everyone who “adopted a duck, showed up, cheered loud, and helped us celebrate 20 years of racing for inclusion.”

I’ve got 30 seconds

Some headlines from this morning:

• President Trump deployed 800 National Guard troops to Washington, D.C. and took control of the Capitol's police force overnight, saying the city was plagued by "bloodthirsty criminals" and "roving mobs of wild youth." D.C. Mayor Muriel Bowser acknowledged the president's authority to take control for 30 days but called the actions "unsettling and unprecedented." Critics noted that D.C. crime rates are at a 30-year low after peaking in 2023, calling the takeover unlawful. The deployment came after a DOGE employee was allegedly beaten during an attempted carjacking last week.

• The U.S. and China extended the suspension of mutual higher tariffs for another 90 days on Monday, just hours before tariffs of up to 145% were set to be reimposed. In a joint statement, Washington and Beijing said the current 30% tariffs on Chinese products and 10% on U.S. products would remain while negotiations continue. The White House said negotiations would focus on "remedying trade imbalances" and "unfair trade practices." The extension prevents an escalation in the ongoing trade dispute between the world's two largest economies.

• The Office of the UN High Commissioner for Human Rights condemned an Israeli strike that targeted a designated tent for journalists, killing seven people. The UN Human Rights Office called the strikes a "grave breach of international humanitarian law." Among the dead was prominent Al-Jazeera reporter Anas al-Sharif, whom Israel said they targeted because of his alleged connection to Hamas. UN Secretary-General Antonio Guterres called for an independent investigation and noted that over 240 journalists have been killed during the Israel-Gaza war.

• Taylor Swift announced her twelfth album, "Life of a Showgirl," is available for pre-order during an appearance on her boyfriend Travis Kelce's podcast. The announcement follows last year's "Tortured Poets Department," which broke Spotify and other streaming records. Swift concluded a 152-city tour last year that drew over 1.2 million people in the UK alone and generated billions in economic activity worldwide. Her website says pre-orders will ship before October 13, but no official release date has been confirmed.

I’ve got 1 minute

A gunman fixated on vaccine conspiracy theories killed a police officer and fired 180 rounds at the CDC headquarters in Atlanta on Friday, marking the deadliest attack on America's top health agency. The shooting highlights escalating threats against public health officials following years of COVID-19 misinformation. Here's what you need to know.

What happened?

Patrick Joseph White approached the Centers for Disease Control and Prevention campus armed with five guns before security turned him away. He then retreated to a pharmacy across the street and opened fire on the sprawling government complex.

White smashed over 150 windows across four CDC buildings during the attack. A DeKalb County police officer was killed responding to the scene. The officer leaves behind a pregnant wife and two children.

Police found White dead on the pharmacy's second floor after a shootout. Officials haven't determined whether he was killed by police gunfire or took his own life.

Shooter’s motivation

Law enforcement officials told The New York Times that White blamed the COVID-19 vaccine for his physical and mental health problems, including depression. Investigators believe he targeted the CDC because it has become central to vaccine conspiracy theories that portray the public health agency as harmful to Americans.

The attack represents the most violent escalation of threats against the CDC, which has faced increasing hostility since the pandemic began in 2020.

Reaction

Health and Human Services Secretary Robert F. Kennedy Jr., who oversees the CDC, condemned the shooting. "No one should face violence while working to protect the health of others," Kennedy said.

However, CDC employees speaking to U.S. media blamed Kennedy for contributing to anti-vaccine sentiment during his previous career as a prominent vaccine skeptic.

"The hatred RFK and his allies have spent their lives stoking puts a target on the backs of anyone in public health," a senior CDC employee told MSNBC.

The FBI will continue to investigate the attack, while the CDC reviews security protocols at its Atlanta headquarters. The shooting comes as health officials nationwide report increased threats and harassment since the pandemic.

I’ve got 2 minutes

Last week, OpenAI released GPT-5, claiming its latest AI model operates at the level of a PhD student in math, coding, and science. The company says it sets new performance records, but experts warn these benchmark claims may be misleading as AI companies compete for dominance. Here's what you need to know.

GPT-5

GPT-5 reportedly achieved breakthrough scores on standardized AI tests measuring mathematical reasoning, coding ability, and multimodal understanding across images, text, and code. OpenAI CEO Sam Altman called it "a significant leap in intelligence over all our previous models" and "a significant step" toward Artificial General Intelligence - AI that can perform any intellectual task humans can do.

The company highlighted GPT-5's performance on medical exams, legal tests, and complex problem-solving benchmarks typically used to evaluate AI capabilities.

Disputed claims

AI researchers say benchmark testing has become unreliable as companies manipulate results to boost marketing claims. "Developers cherry-pick tests that favor their model's strengths while hiding weaknesses," said Dr. Emily Bender, a computational linguistics professor at the University of Washington.

Three main issues plague AI testing: Companies can train models specifically on test questions (like having exam answers beforehand), they highlight favorable results while downplaying poor performance in other areas, and many current tests are too easy for advanced AI systems to meaningfully differentiate between them.

The landscape

Every major AI company has made similar "best ever" claims recently. Elon Musk's xAI claimed Grok-4 could "reason at PhD level" last month. Meta said Llama-4 outperformed earlier ChatGPT models in April. Google claimed Gemini was "first to outperform humans" in December.

"The problem isn't that these models aren't improving—they are," said Timnit Gebru, founder of the Distributed AI Research Institute. "It's that we lack reliable ways to measure and compare that improvement."

So… which model is the best?

The "best" AI model depends entirely on your needs. Some excel at creative writing, others at mathematical reasoning or coding assistance. Specialized benchmark tests now measure everything from cost-effectiveness to which AI is most likely to report criminal behavior to authorities.

Industry experts recommend trying multiple models for your specific use case rather than relying on company performance claims. Independent testing organizations are working to develop more reliable evaluation methods as AI capabilities rapidly advance.

TDU asks