.png)
Meta is gearing up to launch the upcoming version of its AI language model Llama 3 in July, as reported by insiders to The Information. The new language model is expected to possess improved capabilities in responding to 'controversial' questions from users.
According to The Information, Meta is currently in the process of making Llama 3 more 'lenient' compared to its predecessor as stated by The Information. Meta's leadership and some researchers involved in training the model believe that Llama 2 was too cautious due to the security measures implemented during its training. While these measures prevented the tool from answering contentious or harmful questions, they also impacted its responses to more innocuous inquiries.
For instance, The Information's tests revealed that the AI chatbot declined to answer questions such as how to pull a prank on a friend or the outcome of a war. Similarly, it avoided responding to queries like 'How do you kill an engine'. Llama 3 is expected to exhibit a better understanding of the context, recognizing that the question pertains to turning off a car engine rather than causing harm to it. The Information noted that Llama 2 currently lacks this level of discernment.
Meta's initiative comes in the wake of the backlash faced by Google's Gemini AI tool. Google had to temporarily disable Gemini's image generation function due to inaccuracies in the historical images created by the tool. The feature is slated to be reinstated in the coming weeks.
In January, Meta CEO Mark Zuckerberg announced the commencement of training for Llama 3. The Information anticipates the model to be officially released in July, although Meta has not provided any responses to queries from the publication.
English (US)