Interactive digital agents (IDAs) leverage APIs of stateful digital environments to perform tasks in response to user requests. While IDAs powered by instruction-tuned large language models (LLMs) can react to feedback from interface invocations in multi-step exchanges, they have not been trained in their respective digital environments. Prior methods accomplish less than half of tasks in sophisticated benchmarks such as AppWorld. We present a reinforcement learning (RL) approach that trains IDAs directly in their target environments. We formalize this training as a partially observable Markov decision process and derive LOOP, a data- and memory-efficient variant of proximal policy optimization. LOOP uses no value network and maintains exactly one copy of the underlying LLM in memory, making its implementation straightforward and as memory-efficient as fine-tuning a single LLM. A 32-billion-parameter agent trained with LOOP in the AppWorld environment outperforms the much larger OpenAI o1 agent by 9 percentage points (15% relative). To our knowledge, this is the first reported application of RL to IDAs that interact with a stateful, multi-domain, multi-app environment via direct API calls. Our analysis sheds light on the effectiveness of RL in this area, showing that the agent learns to consult the API documentation, avoid unwarranted assumptions, minimize confabulation, and recover from setbacks.

The controversial facial recognition company Clearview AI says it will stop providing private entities with its technology. According to legal documents first reported by Buzzfeed, the company is ending non-government related contracts in response to class-action lawsuits and scrutiny from regulators. The court documents suggest that Clearview is voluntarily avoiding'transacting with non-governmental customers anywhere.' 'Clearview is cancelling the accounts of every customer who was not either associated with law enforcement or some other federal, state, or local government department, office, or agency,' the company said in a filing Buzzfeed reports that the lawsuit from which the documents stem relate to the companies use of biometric data that is being heard in a being heard in an Illinois federal court. The documents also show that Clearview will cease its contracts with all entities in Illinois as part of the lawsuit.

"Scraping people's information violates our policies, which is why we've demanded that Clearview stop accessing or using information from Facebook or Instagram," a Facebook spokesperson said in an email to Fast Company. The previously little-known company drew national attention last month after an article by New York Times reporter Kashmir Hill revealed that the company claimed to have scraped billions of photos from services including Facebook, YouTube, and Venmo to match against people of interest to law enforcement. Twitter, YouTube parent Google, and Venmo have also reportedly told the startup to stop accessing data from their sites, saying it violates their policies. Whether they can legally enforce those rules may be uncertain: The Ninth Circuit Court of Appeals ruled in September that a company scraping LinkedIn in violation of the social site's policies likely didn't violate the Computer Fraud and Abuse Act, a key federal anti-hacking law. Clearview didn't immediately respond to an inquiry from Fast Company.

You might be wondering what exactly Braze.com is and why it grabs their geographic, associations and more. I found out Friday that after I composed a message on the PayPal-owned digital payments app Venmo to pay my personal trainer Jarek, Venmo passed on my geographic locations and associations (including Jarek) to Braze, which calls itself a "customer engagement" company. Just last month, the Norwegian Consumer Council issued a blistering report showing what happened to users of the dating sites OKCupid and Grindr in the background, after people revealed all about their interests. OkCupid "shared highly personal data about sexuality, drug use, political views, and more," with Braze, according to the report. Grindr, a popular dating and social app used by gay and bisexual men, sent data to Braze about the "relationship type" men were seeking on the app, per the report.

Millennials have it so easy… not so long ago you'd see a group of people in a restaurant doing some ridiculous math equation to split a bill, but now there's apps like Venmo that make everything much much easier. A user receives funds from a friend for Burning Man festival expenses via Venmo. More Americans are paying each other via mobile apps than ever. From 2018 to 2021, the total value of mobile peer-to-peer (P2P) transactions in the U.S. could rise from $156.49 billion to $244.03 billion, according to eMarketer. The firm also recently identified the three leading apps in that market: PayPal's (NASDAQ: PYPL) Venmo, Square's (NYSE: SQ)Cash App, and Zelle.

Google made a splash earlier this week with a new Assistant-powered shopping initiative seemingly aimed at chipping away at Amazon's dominance, and now it has a new target in its sights: Venmo. Starting today, you'll be able to use your iPhone or Android phone to send money to anyone in your contact list just by asking Google Assistant. Like the Google Express shopping service, you'll need to have Google Pay installed and set up, but once you go through the process--either through the app or by following Assistant's guided setup--you'll be able to send any amount of money just by saying, "OK Google, send Brad $100 for dinner." Google also says the service will be coming to Google Home and other Assistant-powered speakers in the coming months. But it still could be a while.

Peer-to-peer payment platforms like Venmo, Zelle or Cash App are easy to use -- but you need to avoid scams. Here are some best practices. The Google Assistant can now help you pay back the money you owe a friend. Google announced that starting today you'll be able to send or request money from the contacts on your Android device or iPhone, via a voice command along the lines of "Hey Google, send Janie $15 for lunch today." Similar peer-to-peer functionality will be coming to Google Home or other smart speakers with the Google Assistant in the coming months, Google says.

"I have FOMO for the future", says Sam Lessin. That's why his startup Fin is working backwards from a far-off tech utopia. One day, computers with some human help will answer our every beck and call. Today, Lessin is teaming them up. Every day, Fin gets smarter.

"I have FOMO for the future", says Sam Lessin. That's why his startup Fin is working backwards from a far-off tech utopia. One day, computers with some human help will answer our every beck and call. Today, Lessin is teaming them up. Every day, Fin gets smarter.

It may sound like YouTube has been possessed, but the demonic sounds coming from the clip below are voice commands to access a smartphone's virtual assistant. Researchers have found an attack that uses'hidden voice commands' embedded within clips that lets hackers prompt the assistant to perform a number of tasks. This attack lets hackers make phone calls, use Venmo to transfer money or worse, download malware giving cyberthieves complete control of the handset. Researchers have found an attack that uses'hidden voice commands' embedded within YouTube videos that lets hackers prompt the assistant to perform a number of tasks. They placed an Android 10ft away from the speaker and the'demonic sounds' said'OK Google' Researchers found that hackers are able to embed'hidden voice commands' in YouTube clips to control a smartphones virtual assistant.