Google’s AI Co-Scientist solves a superbug conundrum that took a scientist years

Google has created a new AI research tool it calls “Co-Scientist”, which is designed to generate hypotheses for future scientific research projects. As part of the testing and validation of the tool, Google asked esteemed Professor Jose Penades, Chair of Microbiology at Imperial College London and a Fellow of the Royal Society, to give the AI a prompt to generate hypotheses about a problem in his field. He decided to ask it for hypotheses on mechanisms whereby some bacteria evolve into multi-drug resistant strains called “superbugs” that are exceedingly difficult to eradicate from a patient. Professor Penades and his team had been working on this problem for years, and their key research had not yet been published, so there was no chance of the AI cheating by looking up the answer. The Co-Scientist AI replicated their years of work in 2 days. Professor Penades confesses to an initial shock at this result, but is now excited to turbocharge his research with this powerful new tool, accelerating the process of generating fruitful hypotheses for study.

Google’s AI Co-Scientist replicates years of research in 2 days

Claude’s latest upgrade, 3.7 Sonnet, wows the AI crowd

The jockeying among top-tier AI companies to outdo competitors has produced an ever-shifting leaderboard. This week, it’s Anthropic’s Claude basking in the spotlight. Claude’s latest upgrade to its very popular Sonnet midsize model is topping the charts of benchmarks, equaling or besting the performance of competitors’ models that are larger. Claude has long been a favorite for users who want to use it to code simple computer applications from text requests. The new Claude leans into this, with even better computer coding ability. Claude 3.7 Sonnet also has the ability to drop into “reasoning” mode, where it tackles queries step by step and checks its work, allowing it to handle more detailed and complex requests. All in all, the consensus is that Claude is more than holding its own in the race for ever more intelligent AI.

Claude’s latest upgrade, 3.7 Sonnet, wows the AI crowd

Perplexity announces Comet, a new “Agentic” AI browser

AI search startup Perplexity wants to be Google for the AI Age. Their original product is an AI-enhanced search, which answers queries with AI-produced summaries of the relevant information from across the web, rather than a list of links (of which all the top spots are spammy “sponsored” links for sites that pay to play.) They upped the ante by releasing a more powerful research product with the exact same name as the competing Google offering, Deep Research. And now Perplexity is coming after Google’s Chrome browser, which currently has 65% of the global market. Details are sparse, but Perplexity’s new browser, named “Comet”, is reported to include “agentic” features, meaning at a minimum that it will perform more complex and detailed searches as needed, and may mean that the browser will perform tasks for you by interacting with the other apps on your computer. Someone is going to reinvent the browser with AI, and Perplexity is a leading contender.

Perplexity announces Comet a new “Agentic” AI browser

Nvidia helps launch platform to teach American Sign Language

AI chip-making giant Nvidia has launched a platform to teach American Sign Language (ASL), in collaboration with the American Society for Deaf Children and creative agency Hello Monday. ASL is the third most prevalent language in the US (after English and Spanish.) Yet most deaf children are born to hearing parents who don’t know how to sign. In order to help bridge the gap in communication, the Nvidia-led initiative, known as Signs, has created an online resource where native signers and expert translators can add video clips that demonstrate individual signs and explain their meaning with text and video. This repository will be used to train AI models that can understand signs and critique the signing of individual learners. The current version of signs contains 100 core words; the goal is to expand this number to at least 1000.

Nvidia helps launch platform to teach American Sign Language

Turkey’s elite translators are training the AI that will replace them

The modern nation of Turkey arose from the ashes of the shattered Ottoman Empire after World War I. Its first President, Mustafa Kemal Ataturk, initiated reforms that rapidly secularized and Westernized the new country. One consequence of this seismic cultural shift was the rise of a highly respected elite class of multilingual translators, who translated works in the various European languages into Turkish, and who assisted businesses and the government with communication to counterparts in the West. Now much of the day-to-day work of translating is performed by AI, and professional translators are used primarily either for a final review and edit of machine-translated text, or for training AI models to be even better at it, further reducing the work for human translators in the future. The AI training jobs pay well for now, up to 4 times the going rate for conventional translation, and so translators, particularly young professionals without an established position, are flocking to them. They typically work as contractors for AI outsourcing companies like Scale AI, which has come under criticism in the past for exploitive personnel policies in underdeveloped countries. Yigit Bener, former director of the Conference Interpreters Association of Turkey, believes that AI will eventually replace human translators in all but the most sensitive diplomatic meetings.

Turkey’s elite translators are training the AI that will replace them

The rise of AI boyfriends in China: a billion-dollar business in 2 years

Just 2 years ago, in January 2024 to be exact, a Shanghai-based game development company known as Paper Games released an AI-based relationship game called “Love and Deepspace.” The game uses AI and voice recognition technology to create highly engaging interactions between players and a virtual boyfriend. The game has become a cultural phenomenon in China, with over 6 million monthly users, almost all female. A virtual boyfriend, unlike a real one, is always available, responds quickly to phone calls and texts, and is always sympathetic and supportive. The game encourages in-game purchases of enhanced gameplay and interactions, and there are multiple reports of women spending thousands of dollars on their virtual boyfriend, a trend which has helped the game’s inventor become a billionaire.

The rise of AI boyfriends in China: a billion-dollar business in 2 years

China’s EV giants are mass producing cheap humanoid robots

This month’s MIT Technology Review magazine analyzes a growing trend in the humanoid robot space: the aggressive entry of multiple Chinese electric vehicle companies. China’s hyper-competitive EV sector has been winnowed down from over 480 companies to a mere 40, through a brutal process of mergers, acquisitions, and bankruptcies. The remaining giants are seeing profits squeezed , and are looking for higher-margin products to produce with their world-class engineering skills. Humanoid robots are having A Moment these days, with US startups raking in tons of VC money at dizzying valuations, even if they have few (or even no), existing customers. China’s EV manufacturers are diving in, seeing a chance to simultaneously give their stock a buzzy AI/humanoid robot boost, cut labor costs using their own robots to assemble their EVs, finance the new robot product line by having the EV operations as a captive customer, and deploy their engineering and manufacturing prowess in a lucrative new field. Fair warning to US robotics start-ups: never, ever underestimate China’s engineering. As reported earlier in this space, China graduates almost as many engineers each year (1.6 million) as all the working engineers in the US (2 million.) ‘Nuff said.

China’s EV giants are mass producing cheap humanoid robots

Uber Eats rolls out food delivery robots in Jersey City

Uber is aggressively branching out to bring its ride-hailing technology to autonomous vehicles. Last fall they announced a partnership with Google’s robo-taxi company Waymo, and last week they revealed a partnership with Avride, a spinout from the Russian technology company Yandex. (Yandex had to sell all of its assets at fire sale prices in the wake of the global backlash against Russia’s invasion of Ukraine.) Avride markets autonomous vehicles for taxis, but also makes cute little delivery robots for deliveries in an urban environment. Customers of Uber Eats can use the app to order their food, summon the delivery bot, and open the secure hatch to retrieve their food, after which the bot trundles back for another order. February in New Jersey does not sound like an ideal time to trust a robot to bring you warm food, but Avride insists that their bot is up to the rigors of normal ice, snow, and sleet. We guess that, compared to the Siberian tundra, winter in New Jersey is a breeze.

Uber Eats rolls out food delivery robots in Jersey City

Philips and Mass General partner for real-time AI clinical insights

The various devices that support the care of hospitalized patients generate a tremendous amount of real-time data, but most of it is isolated. The welter of different sensors for everything from heart rate, to blood oxygen, to rate of delivery of IV medications often come from different manufacturers, and they display their data in different ways. Notes from the patient’s chart need to be accessed in a separate system altogether. Now Global health technology giant Philips has partnered with world class hospital system Massachusetts General Brigham to create a “clinical information ecosystem” that will weave all the disparate sources of clinical data into an integrated AI-enabled data resource that can give real time insights into the patient’s condition. The initial goals of the collaboration include data integration, real-time “smart alerts” (rather than all the incessant, infernal beeping), and longer term research and process improvement.

Philips and Mass General partner for real-time AI clinical insights

Head-mounted robotic systems for retinal injections

Sub-retinal injections of medications are an important procedure for treating a variety of age-related, congenital, and other sight-threatening conditions. The most common is age-related macular degeneration, which affects approximately 1 in 10 of US adults over 50. However, accurately depositing medication just beneath the retina without causing damage is a highly delicate undertaking, at the limit of human skill. Now a team of researchers at the University of Utah and elsewhere have designed a head-mounted robotic injection system that permits precise injection of the sight-saving medication.

Head-mounted robotic systems for retinal injections