SCROverfitting is a common problem in AI training where the model learns the training data too precisely, rather than understanding general patterns. It can be compared to a student who memorizes example problems from a textbook instead of understanding the underlying mathematical principles. When faced with slightly different problems in an actual test, they fail. …
We are in the third year of the generative AI boom. While some are predicting a stall for the industry as a whole, I personally expect 2025 to be a year of interesting, meaningful and useful advances. Generative AI is here to stay. The advances to come may not always look as exciting as those …
Microsoft has made its Phi-4 AI model freely available as open-source software on the Hugging Face platform. As reported by Carl Franzen, the model was previously only accessible through Microsoft’s Azure AI Foundry platform. The 14-billion-parameter AI model has demonstrated strong capabilities in mathematical reasoning and language understanding tasks, outperforming larger models in specific benchmarks. …
Adobe Research and Hong Kong University of Science and Technology have developed TransPixar, an AI system that creates transparent visual effects for videos. As reported by Michael Nuñez in VentureBeat, the technology generates see-through elements like smoke and reflections that blend naturally into scenes. The system works with limited training data and uses a novel …
Meta has developed a new technology called scalable memory layers to improve knowledge retention and reduce hallucinations in large language models. As reported by Ben Dickson in VentureBeat, this innovation adds parameters to AI models without increasing computational demands. The memory layers work by using sparse activations and key-value lookup mechanisms, making them more efficient …
AI startup Anthropic is negotiating a $2 billion funding round that would value the company at $60 billion. According to Wall Street Journal reporter Berber Jin, Lightspeed Venture Partners is leading the investment. The deal would make Anthropic the fifth most valuable U.S. startup, following its previous $18 billion valuation in 2023. The company’s chatbot …
Meta has begun testing a feature that automatically generates AI images of Instagram users and displays them in their personal feeds. According to Jason Koebler from 404 Media, the feature creates scenarios based on users who have previously utilized Meta’s “Imagine Yourself” AI tool. A Reddit user first reported seeing AI-generated images of himself in …
Meta has discontinued its AI-powered social media profiles following problematic interactions with users. According to journalist Johana Bhuiyan’s report in The Guardian, the company deleted 28 AI personas that were created in September 2023. These profiles, which included characters like “Liv,” a self-described Black queer mother, sparked controversy when users discovered the AI’s development team …
Niklas Zennström, co-founder of Skype and prominent European tech investor, believes European startups can thrive in AI despite the significant funding gap with US competitors. In an interview with the Financial Times, as reported by Tim Bradshaw, Zennström explained that European companies can succeed by developing applications on top of established AI platforms from companies …
Nvidia CEO Jensen Huang has announced that his company’s AI chips are advancing at a rate surpassing Moore’s Law, the long-standing principle of computing progress. According to TechCrunch’s recent report, Huang states their latest data center superchip performs 30 times faster for AI tasks than previous generations. The CEO attributes this acceleration to Nvidia’s comprehensive …