BBC mulls in house AI model from text archives; MIT develops AI tool for 30x faster high quality image generation; AI could automate 84% of government jobs, says study; Yotta data services targets global AI market with cheapest compute rates- this and more in our daily roundup. Let us take a look.
BBC plans to create an in-house AI model using text archives, aiming to enhance production processes. The model, potentially a Large Language Model, would be solely accessible to BBC. Talks with tech firms on archive access for AI training reported, but BBC denies commercial agreements. Focus also on addressing bias concerns within AI models, Reuters reported.
Also read: Samsung Galaxy Ring may become your nutritionist! Know what's likely coming
AI now produces high-quality images 30 times faster in a single step. Traditional diffusion models, requiring complex iterations, have been revolutionized by MIT CSAIL researchers. Their framework, Distribution Matching Distillation (DMD), simplifies the process into a teacher-student model, retaining image quality while significantly speeding up generation. This advancement merges principles of GANs and diffusion models, potentially becoming a new standard in generative modelling, according to an MIT News report.
New research from The Alan Turing Institute suggests AI can automate 84 percent of UK government services, including passport processing and voter registration. Dr. Jonathan Bright highlights the potential for significant time savings. The study examines 143 million repetitive transactions, indicating AI's potential to reshape modern governance, NDTV reported.
Yotta Data Services CEO claims to offer the world's cheapest AI compute rates, citing prices as low as $1.8 per GPU hour for multi-year contracts. With a strategic partnership with NVIDIA, Yotta aims to address global GPU shortages. Operating in special economic zones and leveraging energy supply from group companies, Yotta plans to achieve breakeven within 2-3 years,
Read more on tech.hindustantimes.com