- AI Tangle
- Posts
- ☕️ A New AI Champion in The Ring
☕️ A New AI Champion in The Ring
Anthropic takes the crown from the OpenAI's industry-standard GPT-4, triumphing over it with Claude 3. Competition, however, is also brewing elsewhere as Databricks releases its DBRX MoE model, Intel starts sharing some of the first bits of an AI PC, and Meta looks to make its Ray-Ban's even more futuristic with AI. Join us at AI Tangle as we untangle this week's happenings in AI.
THE BIG AI STORY
Anthropic's groundbreaking AI model, Claude 3 Opus, has recently topped the popular LMSYS Chatbot Arena leaderboard by nature of human blind-ranking human votes, surpassing OpenAI's GPT-4, marking the first time the poster child of AI models has been dethroned since last year. Better yet, all three of Claude's variations made it to the top 10, solidifying Anthropic as a frighteningly efficient competitor in the scene.
What are the results?
The Chatbot Arena, run by the Large Model Systems Organization (LMSYS), is a platform that compares large language models in randomized and anonymous "battles" using the Elo rating system, and the whole Claude 3 family claimed spots in the top 10, with Opus at #1, Sonnet at #4, and Haiku at a comfortable #6. Haiku, the smallest of the Claude 3 family, was an especially impressive outlier, capable of offering equal performance to many top models despite its lightweight and cheap nature as Anthropic raises the bar for future competitors.
6 QUICK HITS
A trio of notable names in tech has banded together to push the idea of an "AI PC," those being Intel, Microsoft, and Qualcomm. Microsoft has been silent about the finer details, but Intel is stating to share some of Microsoft's requirements for OEMs to build an AI PC, namely that it must have a key for Microsoft Copilot. OEM partners would need to provide a combination of hardware and software, which adds a Neural Processing Unit (NPU) to the list of requirements.
The US government has mandated that all federal agencies must have a senior leader overseeing their AI systems to ensure safe and responsible use. These new regulations, recently announced by Vice President Kamala Harris from the Office of Management and Budget (OMB), state that federal agencies must also establish AI governance boards, submit annual reports on their AI systems, and ensure their AI deployments meet safeguards against algorithmic discrimination.
Databricks recently introduced DBRX to the world, the company's generative AI model with a $10 million investment, presenting itself as the best performing open source model on standard benchmarks, such as Meta's LLaMa 2. However, its accessibility is questionable at best, requiring a substantial hardware investment of 4 expensive Nvidia H100s and posing many licensing issues. Moreover, the model fails to outperform the industry-standard GPT-4 despite DBRX's introduction being so much later into the life cycle of current-generation LLMs.
Tech giant Meta is looking to give its Ray-Ban's the AI treatment next, boosting it with a multimodal AI model, according to a report by The New York Times. The project has been in early access since December of last year and can perform translation in five languages, along with object, animal, and monument identification with a simple "Hey Meta" to get it going. Though not 100% accurate as of writing, the company will most likely continue refining the features currently available through an early access waitlist for users in the US.
Showcasing confidence in Anthropic's future potential, Amazon has invested an additional $2.75 billion in AI company Anthropic, bringing its total investment to the planned $4 billion after its initial $1.25 billion investment in September of last year. Anthropic's AI models compete at levels, and with the wildly successful launch of Claude 3 and its performance at the LMSYS Chatbot Arena leaderboard, it's no surprise that Amazon is doubling down on its investment.
Cybersecurity startup Cyera sets out to tackle data protection after a $300 million round of funding
Cyera, a cybersecurity startup that builds AI-enhanced tools for enterprise data protection, is finalizing a funding round of nearly $300 million, which would triple its valuation and put it at $1.5 billion, with the deal expected to close in April. Though not what Cyera aimed to do right out the gate, the company addresses the challenges posed by AI, including potential breaches of internal intellectual property and data protection policies, including scenarios where companies may cause self-inflicted breaches.
4 AI TOOLS
PageGenie - Create automated high-quality SEO content for your brand at scale with PageGenie, an all-in-one AI-enhanced service that handles everything SEO.
GEES - Covering everything from brainstorming to project delivery, GEES' AI-powered platform simplifies the design workflow, enabling users to consolidate all their work into a single cohesive file.
DryMerge - Let DryFlow's AI turn your repetitive tasks into an easy-to-understand automatic workflow - sales lead notifications, meeting summaries, and personalized reports, all from a simple description.
Syncly - Syncly helps businesses understand their customer's pain points by analyzing everyday communication data using AI feedback analysis techniques, resulting in categorized feedback and revealing hidden negative signals.
AI READ & WATCH
The U.S. Electricity Problem with AI (3-min read)
As AI becomes more prevalent in everyday life, more and more companies are in a rush to build their own, but could it be too big of a strain on the grid? Join Morning Brew for a quick and insightful look into the electricity glutton.
Making AI Accessible (37-min watch)
Andrej Karpathy, a founding member of OpenAI and former Senior Director of AI at Tesla, recently spoke with Stephanie Zhan at Sequoia Capital's AI Ascent, discussing the significance of creating a more open and dynamic AI ecosystem and exploring ways to make building things with AI more accessible.