Luigi Mangione, the man suspected of killing the chief executive of UnitedHealthcare, underwent surgery and was reported missing in San Francisco before the shooting. Brian Thompson, 50, CEO of the healthcare insurance giant, was gunned down last week in Midtown Manhattan, spawning a five-day manhunt that eventually led to Mangione's arrest at a McDonald's restaurant in Altoona, Pa. Questions about Mangione's alleged motives and background have swirled in the media since his arrest Monday. As prosecutors worked to bring him to New York to face charges, new details emerged about his life and his capture. The 26-year-old Ivy League graduate from a prominent Maryland real estate family was charged with murder hours after his arrest.

NEW YORK – The masked gunman wanted in connection with the ambush shooting death of UnitedHealthcare CEO Brian Thompson Wednesday morning fled the crime scene using various modes of transportation before police believe he got on a bus out of the Big Apple, authorities told Fox News. Police traced his route from the crime scene near 54th Street and Sixth Avenue up to Central Park, which he exited at 77th Street and Central Park West, Chief of Detectives Joseph Kenny told Fox News Friday. Kenny's boss, new NYPD Commissioner Jessica Tisch, added that investigators have picked up an abundance of video and digital evidence in addition to physical evidence they hope can lead them to the killer. "We actually have a tremendous amount of forensic evidence in this case that we've collected- DNA evidence, fingerprint evidence, which is all at the lab now being processed," she told Fox News Friday. This undated photo provided by UnitedHealth Group shows UnitedHealthcare chief executive officer Brian Thompson.

A controversial AI program used to deny elderly people health coverage is now at the center of questions about the shooting of the UnitedHealthcare CEO. Brian Thompson, 50 was gunned down Wednesday outside a Hilton in Midtown Manhattan in what police have described as a'brazen' and'targeted' attack. The killer is still on the loose and the motive is not yet known - but a former-FBI agent told Newsweek that he may have been denied health coverage. UnitedHealthcare became the largest denier of insurance plans in 2023, dismissing one in every three claims. It has now emerged that during the years before that, the company implemented AI software that had a 90 percent denial rate.

We introduce a new, extensive multidimensional quality metrics (MQM) annotated dataset covering 11 language pairs in the biomedical domain. We use this dataset to investigate whether machine translation (MT) metrics which are fine-tuned on human-generated MT quality judgements are robust to domain shifts between training and inference. We find that fine-tuned metrics exhibit a substantial performance drop in the unseen domain scenario relative to metrics that rely on the surface form, as well as pre-trained metrics which are not fine-tuned on MT quality judgments.

We propose the on-the-fly ensembling of a machine translation model with an LLM, prompted on the same task and input. We perform experiments on 4 language pairs (both directions) with varying data amounts. We find that a slightly weaker-at-translation LLM can improve translations of a NMT model, and ensembling with an LLM can produce better translations than ensembling two stronger MT models. We combine our method with various techniques from LLM prompting, such as in context learning and translation context.

We show that content on the web is often translated into many languages, and the low quality of these multi-way translations indicates they were likely created using Machine Translation (MT). Multi-way parallel, machine generated content not only dominates the translations in lower resource languages; it also constitutes a large fraction of the total web content in those languages. We also find evidence of a selection bias in the type of content which is translated into many languages, consistent with low quality English content being translated en masse into many lower resource languages, via MT. Our work raises serious concerns about training models such as multilingual large language models on both monolingual and bilingual data scraped from the web.