In this week’s AI news, Adobe demonstrated its VideoGigaGAN model which can upscale blurry videos up to eight times sharper. Technology like this could enable HD versions of historical archives and old TV shows. However, this generative AI surge is also leading to a rise in deep fake porn and political propaganda. Following this, Facebook’s Meta has its Oversight Board assessing its handling of explicit deep fakes as it formulates a strategy to manage content moderation.
In other development, Meta released two versions of the language model Llama-3: a smaller 8-billion-parameter and a larger 70-billion-parameter model. Both have scored impressive results on benchmark tests, with the Llama-3-70B model rivaling Google’s Gemini 1.5 Pro. Reports suggest that Meta is training a massive 400B and multimodal versions of Llama 3. In smaller models, Microsoft unveiled the Phi-3 Mini. Despite its size, this model proves powerful, outperforming models twice its size, making it a strong candidate for future mobile applications.
Meanwhile, NVIDIA’s stocks faced a 10% fall but this does not indicate a decrease in investor AI enthusiasm, and normal service is expected to resume soon. Additionally, the Defense Advanced Research Projects Agency (DARPA) reported positive outcomes in simulations with its AI fighter jet pilot and recently tested an AI-piloted F-16 jet against human pilots. This has raised concerns that AI’s autonomous capabilities could escalate cybercrime, as it can be programmed to exploit cybersecurity vulnerabilities.
In job automation, Wendy’s fast-food chain is adopting AI for its drive-thru services. Also reported this week, Microsoft’s VASA-1 model can convert a single image and audio into lifelike talking face videos, and AI can accurately predict an individual’s political beliefs through expressionless faces. Other noteworthy discussions involved AI’s ethical dilemmas and advancements in AI assistant technology.
Finally, in an exclusive interview series exploring the intersection between AI, robotics, and blockchain, Cogniteam CEO Dr. Yehuda Elmaliah provided insight on their software enhancing robotics development. Radovan Kavicky, an AI & Data Science evangelist at AIslovakIA, highlighted the importance of safely developing Artificial General Intelligence while criticising OpenAI’s approach.
