Meta AI Chief’s Surprising Prediction

Meta’s Chief AI Scientist, Yann LeCun, quips that giant language models (LLMs) like ChatGPT, Google’s Gemini, and Meta’s Llama won’t achieve artificial general intelligence (AGI). In keeping with LeCun, LLMs have the intelligence of a very brilliant house cat, lacking logic, physical understanding, and protracted memory. While OpenAI has given ChatGPT a semblance of working memory, LeCun is not convinced this elevates it much. He wryly dismissed the push to regulate super-intelligent AI, suggesting we first need AI smarter than feline companions. Meanwhile, LeCun’s 500-person team at Meta is ambitiously working on developing AI with real common sense and world comprehension. He expects this enterprise, aiming beyond today’s chatbots that produce surreal images and peculiar illustrations, to take a few decade. Read more.

Amazon is reportedly developing a paid subscription version of its Alexa virtual assistant, powered by its Titan AI model. Expected to launch later this yr, this upgraded Alexa will feature latest generative AI capabilities derived from Amazon’s Titan series of huge language models (LLMs). The Titan Text Premier model, able to processing prompts with as much as 32,000 tokens, incorporates a retrieval-augmented generation (RAG) feature for enhanced responses. Conversely, the entry-level Titan Text Lite model supports simpler text processing tasks with prompts containing as much as 4,000 tokens. The pricing for this upgraded version of Alexa is anticipated to offset the price of the underlying LLM, with subscription fees purported to range from $20 per 30 days to a “single-digit dollar amount.” Read more.

Elon Musk’s discourse at VivaTech 2024 forecasts a future where AI and robots might render employment optional, pivoting towards a society reliant on automated services. He suggests a paradigm shift necessitating “universal high income” to sustain livelihoods on this scenario, distinguishing it from universal basic income. Musk’s concerns concerning the societal implications of AI’s evolution echo through his references to Ian Banks’ “Culture Book Series,” portraying a utopian AI-led society. He ponders the existential query of human purpose in a jobless world, proposing that human involvement in AI development may imbue it with purpose. Despite acknowledging AI’s risks, he advocates for proactive measures, acknowledging its potential societal advantages amidst escalating technological advancement. Read more.

Google Meet is getting a much-needed boost within the audio department with the introduction of “adaptive audio.” This AI-powered feature tackles the age-old problem of echoey conference calls with multiple laptops in a room. By synchronizing microphones and speakers, Google Meet goals to create a smoother, more frustration-free experience.

This feature is a boon for smaller teams and budget-conscious organizations lacking dedicated conferencing equipment. It offers greater flexibility in meeting locations and ensures clear audio whatever the setting. Google plans a measured rollout to specific Google Workspace subscriptions, with the feature conveniently enabled by default but offering the choice for manual control. Read more.

In keeping with The Washington Post, OpenAI didn’t intentionally seek an actor resembling Scarlett Johansson for its Sky voice assistant. The corporate chosen actors based on criteria similar to warmth and engagement, not Johansson’s likeness.

The chosen actor’s natural voice, resembling Sky’s, was confirmed by initial voice test clips. OpenAI emphasized that its Chief Technology Officer, Mira Murati, made decisions regarding AI voices, with limited involvement from Sam Altman. Johansson, approached by Altman in September 2023, declined to voice ChatGPT but later raised concerns about Sky’s similarity to her voice. Read more.