In partnership with

AI Evolution: The Must-Know Updates from Q2

Google's Major AI Upgrades

Overview: Google has recently unveiled Gemma 2, the next evolution in its lightweight AI model series, along with several upgrades to its Gemini 1.5 Pro model, including a 2 million token context window and improved coding capabilities.

Details:

• Gemma 2 is available in two sizes: a 9 billion parameter model and a larger 27 billion parameter model, with a smaller 2.6 billion parameter version hinted for future release.

• The 27 billion parameter model delivers performance comparable to models over twice its size.

• The 9 billion parameter model outperforms similar models, such as Llama 3 8B.

• In addition, Google has enabled access to Gemini 1.5 Pro's 2 million token context window, allowing for the processing of much longer inputs.

• The new coding capabilities for Gemini Pro and Flash models enhance their ability to generate and execute Python code, improving accuracy in mathematical and data reasoning tasks.

Why it matters: These advancements position Google’s Gemma series as a powerful contender in the open-source AI landscape, offering top-tier performance while remaining efficient and cost-effective on a single GPU. Additionally, the introduction of Gemini’s 2 million token context window significantly expands the potential applications for users.

Apple and Meta discuss major AI partnership

Details:

• Apple is considering partnerships beyond OpenAI, exploring integrations with Meta, Perplexity, and Anthropic to join ChatGPT in the new Apple Intelligence system.

• These discussions are surprising given the longstanding rivalry between Apple and Meta, particularly over privacy concerns and App Store policies.

• The proposed deal would allow Meta and other AI partners to offer premium subscriptions through Apple devices, with Apple receiving a share of the revenue.

• Apple's software chief, Craig Federighi, hinted at plans to offer multiple AI options during WWDC.

Why it matters: This potential partnership marks a significant shift in the relationship between these long-time rivals. The AI revolution may be fostering collaboration, with Apple aiming to provide users with the best AI options for various tasks and Meta gaining a substantial new distribution channel for its AI initiatives.

Nvidia’s new open-source AI models

Overview: Nvidia has introduced Nemotron-4 340B, a family of open-source language models designed to generate high-quality synthetic training data and develop powerful AI applications across various industries.

Details:

• The three models (Base, Instruct, Reward) form a ‘pipeline’ for creating synthetic data to train new, powerful LLMs.

• Instruct creates high-quality synthetic training data (and was trained on 98% synthetic data), while Reward filters the data for the highest-quality examples.

• The Nemotron-4 models match or exceed open-source competitors like Llama-3, Mixtral, and Qwen-2 across a variety of benchmarks.

• NVIDIA also released Mamba-2 Hybrid, a selective state-space model (SSM) that surpassed similar transformer-based LLMs in accuracy.

Why it matters: Nvidia's release of these free, open-source models not only rivals the capabilities of leading competitors but also excels in generating the synthetic data necessary for developing new LLMs. This solidifies Nvidia's position as a multifaceted AI powerhouse.

Learn from investing legends.

Warren Buffett reads for 8 hours a day. What if you only have 5 minutes a day? Then, read Value Investor Daily. We scour the portfolios of top value investors and bring you all their best ideas.

Subscribe with one click.