In the middle of an ongoing lawsuit against OpenAI by The New York Times, lawyers from the latter and the Daily Mail are now accusing the startup of deleting data potentially relevant to the case. The setback occurred when OpenAI engineers accidentally erased critical search data on a virtual machine used by the plaintiffs to investigate the AI's training sets. Though most of the data was recovered, the folder structures and file names were lost, forcing the plaintiffs to restart their analysis from scratch.
DeepSeek, a Chinese AI research lab, recently unveiled in an X/Twitter post what could be considered the first reasoning model capable of competing with OpenAI's o1 model, says the firm. Dubbed DeepSeek-R1-Lite-Preview, the lightweight version of the full model, much like o1, is capable of fact-checking itself and takes additional time to "think" things through and avoid common pitfalls. DeepSeek-R1 is also fairly sensitive to anything political. However, many users found this can be circumvented as the model is prone to jailbreaking, including by other LLMs to spit out meth recipes, it seems.
Google's Gemini chatbot was recently upgraded with a memory feature, much like OpenAI's ChatGPT, allowing it to recall user preferences, life details, and the like to tailor interactions more fittingly. Available for subscribers of the $20/month Google One AI Premium plan, memory currently functions only on the web version, despite an iOS app being launched not long ago. Though a few concerns were raised about whether stored memories are used for model training, a Google spokesperson debunked the thought. Users are also free to delete saved memories at any time.
Peter's Chapel in Lucerne, Switzerland, recently experimented with an AI-powered Jesus avatar known as "Deus in Machina," capable of conversing in 100 languages and drawing over 1,000 visitors during a two-month trial starting in August. Albeit emphasized that it was not meant for confessions, the avatar was installed in a confession booth nonetheless and offered responses based on theological texts, providing a mix of profound and superficial answers. Feedback revealed two-thirds of users found the experience spiritual, though some criticized it as impersonal and clichéd.
Data security solutions provider Cyera recently announced that it had raised $300 million in Series D funding at a $3 billion valuation, backed by top investors like Accel and Sapphire Ventures. The company provides an AI-driven data security platform offering real-time visibility, automated policies, and features like data security posture management and advanced data loss prevention. Cyera claims its platform delivers insights in days rather than years, with reputable customers like AT&T, DocuSign, Paramount, and Chipotle backing its product.
Another cybersecurity startup in the news was AI-driven cloud security firm Wiz, which acquired the Israeli cybersecurity company Dazz in what is an approximated $450 million deal, the majority paid in cash and the rest in shares. Founded in 2021, Dazz specializes in automating the remediation of security vulnerabilities, and its platform will be integrated into Wiz Code, a product that links cloud vulnerabilities to their source code to resolve issues faster. Dazz marks Wiz's third acquisition in the past 12 months, adding itself to the pile with Raftt and Gem Security.