New Claude 3.5 Sonnet Model Surpasses ChatGPT-4o
Anthropic has unveiled its latest AI model, Claude 3.5 Sonnet, which is the middle model in Anthropic's lineup. Despite its middle position, Anthropic asserts that 3.5 Sonnet surpasses 3 Opus and boasts a significant speed improvement, reportedly being twice as fast as its predecessor. Benchmarks indicate that Claude 3.5 Sonnet outperforms GPT-4o, Gemini 1.5 Pro, and Meta's Llama 3 in seven out of nine overall benchmarks and four out of five vision benchmarks, though of course, these results should be taken with a grain of salt. Anthropic has also launched a new feature called Artifacts, which enables users to view and interact with the outcomes of their Claude requests, such as editing designs or emails directly within the app.
Reference Links:
After the somewhat hectic week preceding the Apple WWDC event, where they announced their own AI-powered features labelled Apple Intelligence, the company announced that it would not launch some of its features in Europe due to regulatory issues. These features include Apple Intelligence, Phone Mirroring, and SharePlay Screen Sharing enhancements from the latest iOS 18 update. It is also worth noting that neither Apple is paying OpenAI to integrate ChatGPT into iOS 18 nor is OpenAI paying Apple for the exposure. It is simply an exchange of resources.
Reference Links:
Google Deepmind has shown off an audio-to-video AI model on its website, a model that recognizes the scenery of a silent video and creates an audio track based on what it sees using AI. A few demos are available on their page. But more interesting, ElevenLabs has released an open-source video-to-audio model that adds audio to any silent video clip you upload. This application is available at videotosoundeffects.com and works quite well.
Reference Links:
Matt Wolfe (YouTube), ElevenLabs (Twitter)
Ilya Sutskever, co-founder of OpenAI, has announced his new project, Safe Superintelligence Inc. The company aims to develop a powerful and safe artificial intelligence system within a research organization, without any immediate plans to sell AI products or services. This allows Sutskever to focus on his work without commercial pressures and competitive distractions. The initial offering of Safe Superintelligence Inc. will be the safe superintelligence itself, with no other activities planned until then. The applications and capabilities of Superintelligence aren’t yet quite clear.
Reference Links:
OpenAI has acquired Rockset, a leading real-time analytics database, to enhance its retrieval infrastructure. Rockset provides world-class data indexing and querying capabilities, enabling users to better leverage their own data and access real-time information. The acquisition involves integrating Rockset's technology and team into OpenAI's products, enabling businesses to convert data into actionable insights. The financial details of the acquisition were not disclosed, but Rockset has accumulated $105 million in funding thus far. Some members of Rockset's team will be joining OpenAI as part of the acquisition.
Reference Links:
Sam Altman, the CEO of OpenAI, is considering making the company a for-profit business, potentially leading to an initial public offering (IPO). This could involve OpenAI becoming a benefit corporation (B-Corp), which would require balancing profits with social and environmental impact. If successful, this could raise further investment and make the company more financially stable. However, as a public company, OpenAI would be more accountable to shareholders, which might lead to restrictions on free services to encourage users to pay more. OpenAI has stated that it remains committed to its nonprofit mission and will continue to exist alongside any for-profit arm.
Reference Links: