What happened
More than 1,000 people have died of Ebola in Congo, in the fastest-growing outbreak on record, African health officials said yesterday. Congo’s Ministry of Health says 2,473 cases have been reported since this outbreak was declared May 15. There are no approved vaccines or treatments for the Bundibugyo virus responsible for the outbreak.

Who said what
“These are people dying” because “we don’t have vaccines, we don’t have medicine, we don’t have funding,” Dr. Jean Kaseya, head of the Africa Centres for Disease Control and Prevention, said at a health summit in Ghana yesterday. In the worst outbreak to date, which killed more than 11,000 people between 2013 and 2016, it “took about eight months from the first case to reach 1,000 deaths,” The Associated Press said. The current outbreak in Congo appears to be “spreading faster than health officials can track it.”

What next?
An Ebola vaccine from Merck licensed to target the Zaire virus could also offer some protection against the Bundibugyo species, according to a paper published yesterday in the New England Journal of Medicine. The World Health Organization has “expressed lukewarm support” for testing the vaccine in Congo, Stat News said, but Bundibugyo-specific vaccines are “still months away” from clinical trials.  

What happened
French professional model recruiter Daniel Siad, who introduced Jeffrey Epstein to young women and teenage girls for about a decade, was found dead in his home near Paris, French prosecutors said yesterday. Siad’s name “appears more than 2,000 times” in the Justice Department’s Epstein files, which also document Epstein paying him “tens of thousands of dollars,” CNN said.

Who said what
Siad, 69, had been “the subject of special investigative techniques,” including phone taps, as part of a wide-ranging, ongoing investigation of Epstein-related human trafficking and tax fraud, Paris prosecutor Laure Beccuau said in a statement. But nothing had “produced evidence justifying his immediate arrest.” 

The Epstein files exposed the sex offender’s “cozy relationships with high-profile and powerful” people, The Washington Post said, but also a “network of lesser-known associates — including those from the modeling industry — who offered to bring women into his orbit,” often via “crude and sexualized” emails in which the “purpose of the introduction was not explicitly stated.” Siad is the “second man publicly linked to Epstein who has died in France,” Reuters said, after modeling agent Jean-Luc Brunel, “found hanged in his prison cell in 2022.”

What next?
Siad’s lawyer, Ménya Arab-Tigrine, told Reuters her client died of a heart attack, though “we must await the autopsy results to confirm whether his body simply gave out” amid the “grueling media spotlight.” 

If you want a friendlier tone when using Claude, you may want to switch your language settings, according to an analysis by the chatbot’s creator, Anthropic. The company said its study of AI conversations found that Claude “emphasizes different values” when producing responses in different languages. “We do not imply that Claude intrinsically holds values,” Anthropic added, but just “refer to the values reflected by Claude’s behavior and outputs.”

Anthropic compared data from more than 300,000 anonymized conversations in 20 languages conducted over two weeks in May. The study used the technique of “dimensionality reduction” to measure the responses across four separate axes: warmth versus rigor, depth versus brevity, candour versus execution, and deference versus caution.

Arabic responses had “greater warmth,” containing more “polite phrasing, humor, playfulness and affirmation,” said The Decoder. On the candor versus execution axis, Dutch responses showed a more “honest appraisal of potential shortcomings,” said The Register.

The study also showed differences across Claude’s models. Sonnet 4.6, the default free model, is more affirming toward users’ ideas and “offers comfort without passing judgment,” The Decoder said, while Opus 4.7, the premium paid model, “openly critiques” and “questions assumptions.”

Whether these differences “represent desirable adaptation to different speech communities or unintended training effects remains an open question,” said The Decoder. Anthropic said it will continue “making sense of this variation” and use what it learns to “improve Claude’s behavior.”