Welcome to the latest edition of FindBiometrics’ AI update. Here’s the latest big news on the shifting landscape of AI and identity technology:
Lambda Labs is in talks to raise $300 million in funding from investors including SoftBank and Legendary Entertainment founder Thomas Tull. The company specializes in offering relatively low-cost GPU-based cloud compute infrastructure, and sells GPU desktops. The funding would bring its valuation to $1.5 billion.
Kleiner Perkins is leading a funding round that would value the startup Together at over $300 million, before a new $50 million investment. Together runs a cloud service that rents out AI chips, and the company also sells software designed to help other companies use open-source Large Language Models.
OpenAI is planning to launch a memory storage feature for its developer tools, which could reduce costs for app makers by up to 20 times, according to a Reuters report. It’s also planning to give third party developers access to its AI platform’s “vision” capabilities, which allow the GPT-4 chatbot to read and interpret images.
OpenAI CEO Sam Altman recently told staff that the AI giant is generating revenues at a pace of $1.3 billion a year, implying that the company is making over $100 million a month. That would mark a 30 percent increase from this past summer.
The company has also changed the “Core values” listed on its careers page. The previous ones were “Audacious”, “Thoughtful”, “Unpretentious”, “Impact-driven”, “Collaborative”, and “Growth-oriented”. The new ones are “Intense and scrappy”, “Scale”, “Make something people love”, “Team spirit”, and, above all, “AGI focus”, referring to the idea of ‘Artificial General Intelligence’. “Anything that doesn’t help with that is out of scope,” the page says.
Adobe has announced three new generative AI models, including the upgraded version of its flagship image generator, Firefly Image 2, and other Firefly models geared at generating vector images and design templates, respectively. A new “Generative Match” feature is meant to get a generated image to match the style of another image, and automatically attaches metadata to identify it as AI-made.
China’s National Information Security Standardization Committee has published security regulations for companies involved in artificial intelligence. They include a blacklist of sources that must not be used in the training of AI models, with periodic security assessments to verify compliance. They also require organizations to get consent from individuals before using their personal information or biometric data in AI training. And no information that pertains to “overthrowing the socialist system” or “damaging the country’s image” can be used.
Cap_able, an Italy-based company selling clothing that it says is designed to confuse AI systems, is bringing its knitwear to the US market through a partnership with the Philadelphia boutique Trunc. The company says that extensive testing has found that individuals wearing its clothes were not recognized by YOLO, an object detection system. But it also insists that its main objective is to raise awareness about privacy, AI, and biometric data. Prices for its Manifesto Collection garments start at close to $400.
A group of sixteen lawmakers have sent an open letter to the White House asking the Biden administration to incorporate its “AI Bill of Rights” into an anticipated Executive Order concerning AI. The AI Bill of Rights was announced last year as a white paper credited to the Office of Science and Technology Policy (OSTP), and broadly called for the conscientious development and use of AI technologies such as facial recognition. The lawmakers would like to give it some teeth.
MSU Denver Marketing Professor Mick Jackowski believes “we may only be five to 10 years away from AI completely transforming the virtual selling experience,” after visiting the Haaga-Helia University of Applied Sciences in Finland. The latter school has a laboratory dedicated to research using AI and biometrics in marketing applications. Prof. Jackowski was impressed with the work, but said that applying tools like face scanning to virtual meetings doesn’t deliver reliable results — yet.
A new app called Lipdub can convert up to one minute of video-recorded speech into any of 28 languages. Created by New York-based startup Captions, the app uses face biometrics and deepfake technology to convincingly render the subject’s speaking in a given alternate language, while maintaining the same facial expressions as in the original uploaded video. CEO Gaurav Misra says the system is based on “zero-shot” AI models that don’t need to be trained on a particular speaker in order to generate successful output.
The chatbot’s take: This week we asked ChatGPT to read the tea leaves in OpenAI’s Careers page.
October 13, 2023 – by Alex Perala