In a week bustling with AI news, Anthropic’s AI model Claude Sonnet 3.5 took the spotlight for outperforming OpenAI and Google trails in multiple tests, consequently knocking OpenAI off its top spot on the leaderboards.
Anthropic has been amidst praises for the proficiency of Claude Sonnet 3.5, and an even more robust model, Claude Opus 3.5, is slated for release soon. While anticipation builds up for OpenAI’s counter-move, Apple and Meta have put their launches of respective AI features on hold due to strict enforcement of AI laws in the EU region.
The AI space has recently found itself tangled in legal disputes, with text-to-audio platforms Suno and Udio facing lawsuits by the Recording Industry Association of America for alleged copyright infringement. They are accused of using copyrighted sound profiles to train their AI.
Meanwhile, there has been a surge in concerns regarding misuse of AI, especially in the creation of deep fakes for manipulating public opinion. According to a study by DeepMind, this constitutes a majority of AI misuse cases.
A report published by the University of Oxford revealed that AI models are inclined to generate plausible but entirely inaccurate information, especially when their confidence level or ‘semantic entropy’ is high. Furthermore, even the most advanced large language models (LLMs) have demonstrated challenges in solving simple puzzles, suggesting a deficiency in reasoning capabilities.
Despite their limitations, AI models may understand human behavior better than anticipated. A recent study found that they could predict human anxiety levels by analyzing reactions to different visuals.
However, the definition of ‘open’ models in the AI space remains contentious, with many companies retaining a strong hold on their proprietary models. The Science and Technology Innovation Program at the Woodrow Wilson International Center for Scholars defines “open” AI as a model that is accessible, understandable, and usable by all.
Despite the controversy, the open model space has seen some innovation recently with EvolutionaryScale’s ESM3 – generative model for biology that can create proteins beyond nature’s capacity.
Several AI-centric events are lined up for July 2024, including the MarTech Summit Hong Kong 2024 and the Generative AI Summit Austin 2024.
Additionally, other noteworthy updates include:
– Meta’s erroneous flagging of real photos as AI-generated
– OpenAI’s delay of its GPT-4o’s voice assistant launch due to safety concerns
– OpenAI’s recent acquisition of Multi, a collaborative screen sharing tool
– Chinese AI firms attempting to allure OpenAI users in light of the US company’s plan for API restrictions
– An AI-created ad by Toys “R” Us sparking online backlash.
The AI ecosystem faces many challenges and opportunities, particularly in models’ accuracy, potential misuse and legal implications, but the fast-paced advancements in this field promise exciting prospects.
