Bytes: What’s new in tech

DOJ preps for Google antitrust suit

The Justice Department is likely to reject concessions offered by Google parent Alphabet to stave off an antitrust trial, said Leah Nylen in Bloomberg, and “the agency is poised to file a lawsuit” over Google’s practices in the advertising technology market, according to people familiar with the matter. Last week, Google “proposed splitting part of its business that auctions and places ads on websites and apps into a separate company” to address regulators’ concerns. However, that separate company, which on its own “could possibly be valued at tens of billions of dollars,” would still “remain under the Alphabet umbrella.” That’s hardly a compromise, analysts say, as it “wouldn’t eliminate any of its monopoly advantages” as both a leading auction house and bidder for ads.

A crypto disappearing act

The co-founders of failed crypto hedge fund Three Arrows are on the lam, said MacKenzie Sigalos in CNBC.com. The whereabouts of Zhu Su and Kyle Davies were unknown last week after creditors attempted to reach them “ahead of a hearing to discuss next steps in the company’s liquidation process.” Creditors say there is now a real risk the founders could disappear with investors’ money. As recently as March, Three Arrows managed about

$10 billion in assets. But the plunge in the digital-asset market in recent months swiftly wiped out the company. Creditors have been “trying to determine what assets remain” in Three Arrows, but when liquidators arrived at the company’s Singapore offices in June, they had apparently been vacant for some time.

An open-source AI model

More than 1,000 volunteers collaborated on an open-access large language model “that performs as well as other leading models,” said Melissa Heikkilä in MIT Technology Review. The new model, BLOOM, can understand 46 natural languages, including 20 African languages, plus 13 programming languages. Users “can type in requests for BLOOM to do tasks like writing recipes or poems, translating texts, or writing code.” Like other deep-learning algorithms under development at OpenAI or Google, BLOOM is “trained on massive amounts of data” by scraping the internet, often sweeping up bigotry and bias into the training set. BLOOM attempts to mitigate this by putting its code “out in the open,” so researchers can “interrogate the model’s strengths and weaknesses.”