Meta seeks to win AI race with Llama 3 after Google takes backfoot

Meta Llama 3: Meta Platforms' latest artificial intelligence large language model (LLM) Llama 3 is poised to be released in July 2024 with a focus on providing nuanced responses to user queries, according to The Information.

Meta's research team has been diligently working to refine the capabilities of the Llama 3 model, slated for release in July. This endeavour seeks to "loosen up" the model, enabling it to offer contextually relevant responses, especially to contentious questions raised by users. This comes amid the criticisms against Alphabet-owned Google's Gemini AI over alleged misrepresentation of historical facts.

What went wrong with Gemini AI?

Google's Gemini AI faced criticism and backlash due to the historical inaccuracies and questionable responses generated by its image-generation feature. Users discovered instances where the AI produced racially diverse, Nazi-era German soldiers, inaccurately portrayed races of historical figures such as US Founding Fathers, and even misrepresented the races of Google’s own co-founders. This led to concerns about biases and inaccuracies in the AI's outputs, prompting Google to pause the image generation feature and work on resolving the issues before relaunching it.

Learning from industry challenges

 Meta is keen on leveraging these industry learnings to ensure that Llama 3 can navigate complex queries without generating inaccurate or inappropriate content. Tests conducted by The Information revealed limitations in Meta's previous iteration, Llama 2, particularly in handling less controversial questions. However, Llama 3 is poised to overcome these obstacles and provide more insightful responses. For instance, it will have the ability to interpret nuanced queries such as 'how to kill a vehicle's engine' within its appropriate context, showcasing a significant advancement in understanding user intent.