On an earnings call on Tuesday, Pinterest CEO Bill Ready highlighted the promise of open source AI models in helping the company keep its costs down as it expands the use cases for visual AI. The image pinboarding site, which often serves as consumers’ first step in their shopping journey, leverages AI technology to power…
How can we reliably test whether large language models actually understand Indian languages and culture in real world contexts? OpenAI has released IndQA, a benchmark that evaluates how well AI models understand and reason about questions that matter in Indian languages across cultural domains. Why IndQA? OpenAI states that about 80 percent of people worldwide…
For years, Australians have been been installing solar panels at a rapid clip. Now, that investment is paying off. The Australian government announced this week that electricity customers in three states will get free electricity for up to three hours per day starting in July 2026. Solar power has boomed in Australia in recent years….
In this tutorial, we explore how an agent can internalize planning, memory, and tool use within a single neural model rather than relying on external orchestration. We design a compact, model-native agent that learns to perform arithmetic reasoning tasks through reinforcement learning. By combining a stage-aware actor-critic network with a curriculum of increasingly complex environments,…
Eugenia Kuyda saw the future of consumer AI before most. She founded Replika, the first major AI companion startup, in 2017 years before ChatGPT launched. Today it has 35 million users. Now Kuyda is back with a new startup called Wabi, which she describes as YouTube for apps – a social platform where anyone can use prompts to instantly create mini apps and share…
The Motion Picture Association (MPA) has sent Meta a cease-and-desist letter demanding that it stop using the term “PG-13,” as first reported by The Wall Street Journal. Last month, Meta announced that teens on Instagram would, by default, only see content that adheres to PG-13 movie ratings. Two weeks later, the MPA sent Meta a…
How can consistency training help language models resist sycophantic prompts and jailbreak style attacks while keeping their capabilities intact? Large language models often answer safely on a plain prompt, then change behavior when the same task is wrapped with flattery or role play. DeepMind researchers propose consistent training in a simple training lens for this…
The University of Pennsylvania confirmed on Tuesday that a hacker stole university data as part of last week’s data breach, during which alumni and other affiliates received suspicious emails from official university email addresses. “We got hacked,” the message from the hackers read. “We love breaking federal laws like FERPA (all your data will be…
If you look at how AI deals are conducted these days, it seems AI companies and their investors are imitating the circle of life. Only, it’s a circle of profit that ensures the money eventually comes back to their own coffers. Case in point: SoftBank, which is investing tens of billions into OpenAI and committing…
Cybersecurity is a massive sector, but startups in the category are more likely to be acquired than go public. Even Wiz, which for a time held the title of the fastest-growing startup, abandoned its IPO ambitions when it agreed to sell to Google earlier this year. In the past few years, there have been scant few significant…
