👑 Gemini 1.5 Pro dethrones GPT-4o

Also: Meta plans 10x more compute for Llama 4

Author

Evolving AI
August 02, 2024

Welcome, AI enthusiasts

Gemini 1.5 Pro has emerged as the new leader in AI benchmark rankings, outperforming prior records set by other leading AI models such as OpenAI’s ChatGPT-4o. Mark Zuckerberg said on Meta's second-quarter earnings call on Tuesday that to train Llama 4, the company will need 10x more compute than what was needed to train Llama 3. Furthermore, OpenAI CEO Sam Altman says that OpenAI is working with the U.S. AI Safety Institute on an agreement to provide early access to its next major generative AI model for safety testing. Let’s dive in!

In today’s insights:

  • Gemini 1.5 Pro dethrones GPT-4o

  • Meta plans to use 10x more compute to train its Llama 4 AI model

  • OpenAI partners with U.S. AI Safety Institute

Read time: 4 minutes

🗞️ LATEST DEVELOPMENTS

Evolving AI: Google’s Gemini 1.5 Pro model has surpassed OpenAI's GPT-4o for the first time on the LMSYS Chatbot Arena leaderboard.

Key Points:

  • Google’s Gemini 1.5 Pro surpasses GPT-4o and Claude-3.5 Sonnet.

  • Early testing shows robust performance in multi-lingual, technical, and multimodal tasks.

  • Ethical considerations emerge as the model’s capabilities expand rapidly.

Details:

Google's release of Gemini 1.5 Pro marks a significant advancement in AI, quickly capturing the top spot on the LMSYS Chatbot Arena leaderboard with an ELO score of 1300. This positions it ahead of competitors like OpenAI’s GPT-4o and Anthropic’s Claude-3.5 Sonnet. The model excels in a variety of tasks, particularly multi-lingual processing and technical fields like mathematics and coding. Its expansive two-million-token context window enhances its ability to handle complex, multimodal inputs, making it a transformative tool for enterprises in data analysis and software development. However, as AI models grow more sophisticated, ethical considerations and responsible use become critical topics.

Why It Matters:

Gemini 1.5 Pro’s ascent is a pretty significant milestone in the rapidly evolving LLM landscape. As competitors like Anthropic, Google, and, since recently, Meta continue to gain ground with increasingly capable models — the pressure is on for OpenAI to respond with the next groundbreaking release.

Source: Meta

Evolving AI: Meta is preparing to dramatically increase computing power for its upcoming Llama 4 AI model.

Key Points:

  • Meta plans to use 10x more computing power to train Llama 4.

  • The company's strategic partnerships with major cloud providers aim to maximize the adoption and accessibility of Llama 4.

  • CEO Mark Zuckerberg believes AI agents will become common on websites soon.

Details:

Meta is gearing up to launch its next-generation AI model Llama 4 with a focus on increased compute power and wide accessibility. CEO Mark Zuckerberg emphasized the potential of open-source AI, likening its future impact to that of Linux in the software industry. The Llama 4 model will be multimodal, meaning it will have further enhanced capabilities. The company is investing heavily in AI infrastructure with plans to expand compute clusters and data centers to support the model's development. Meta's infrastructure spending is projected to reach $37-40 billion annually, which reflects its continued commitment to open-source AI advancement. Despite some investor skepticism regarding high AI and Metaverse expenditures, Meta's financial performance remains strong, with Q2 revenue and profit showing significant growth.

Why It Matters:

Zuckerberg's vision of AI agents becoming standard for online businesses could transform how companies interact with customers. As AI agents become as common as websites and social media profiles, businesses can offer personalized and instant support. This will improve user experience and satisfaction across all business sizes. However, the widespread adoption of AI agents also raises important concerns about data privacy, security, and the potential impact on human jobs.

Source: Shutterstock

Evolving AI: OpenAI partners with the U.S. AI Safety Institute to share early access to new AI models for safety evaluations.

Key Points:

  • OpenAI has announced a collaboration with the U.S. AI Safety Institute to grant early access to its upcoming generative AI model for safety testing.

  • This partnership is part of OpenAI's efforts to address safety concerns.

  • The collaboration aligns with OpenAI's increased federal lobbying and involvement in AI regulatory initiatives.

Details:

OpenAI has announced a partnership with the U.S. AI Safety Institute to provide early access to its next major generative AI model, emphasizing its commitment to AI safety. This collaboration follows a similar agreement with the U.K.'s AI safety body and comes amid criticism of OpenAI's safety practices. The company had previously disbanded a team dedicated to developing safety controls for "superintelligent" AI, leading to criticism and resignations. In response, OpenAI committed 20% of its compute resources to safety research and formed an internal safety commission. OpenAI's recent actions, including its endorsement of the Future of Innovation Act, could indicate a strategic move to influence AI policy at the federal level.

Why It Matters:

This move follows the complaints filed with the Securities and Exchange Commission (SEC) from anonymous whistleblowers requesting an investigation into whether or not OpenAI illegally restricted workers from communicating with regulators. In May, the company received backlash for its restrictive offboarding policy forbidding ex-employees from criticizing OpenAI. Even acknowledging that such an NDA exists is a violation of the agreement.

💡 Tip of the Day

Over 25 industry veterans and experts have released a course on LLMs. Whether you want to learn more about prompt engineering or how to fine-tune the best OpenAI models, there's something for everyone. The best part is that it's free and open to everyone.

🎯SNAPSHOTS

Direct links to relevant AI articles.

⚖️ Lawsuit: AI music startup Suno claims training model on copyrighted music is ‘fair use’.

🤖 Taco Bell: Taco Bell’s drive-thru AI might take your next order.

🔎 Google: Google makes it easier to remove explicit AI-generated deepfakes from search.

🔥 Google Deepmind: Google DeepMind's new Gemma 2 model outperforms larger LLMs with fewer parameters.

📈Trending AI Tools

  • 🎥 AnyClip - AI-powered tool for video content management (link)

  • 🎨 Yep - AI-powered no-code landing page builder (link)

  • ✍️ AITorke - Tool to create unique content for blogs, videos, and social media, faster (link)

  • 🌍 EasyTripAI - AI-powered travel companion! Everything you need for the perfect trip in one place (link)

  • 🤖 Ribbo.ai - AI-powered customer service agent for your business (link)

  • 💬 Mano - AI Chrome Extension that provides flawless support, answers all your questions, & creates helper Agents directly into any site (link)

Reply

or to participate.