Hey there, 

This week's newsletter is a bit shorter than usual, but don't worry – we've still got you covered with all the latest and most interesting AI developments. The reason for the brevity? Our team is away at our annual offsite meetup. We're taking some quality time to connect in person, reflect on our journey through 2024, and map out where we want to go in 2025.

And with the holiday season kicking into full swing, we'll take a short break and return with our usual in-depth coverage in the New Year’s.

What would be the format? Every week, we will break the newsletter into the following sections:

• The Input - All about recent developments in AI

• The Tools - Interesting finds and launches

• The Algorithm - Resources for learning

• The Output - Our reflection 

Table of Contents

• The Input

  • Meet NVIDIA's latest: A supercomputer that won't b …

  • Veo 2, Imagen 3, and Whisk: The future of visual s …

  • Key highlights of OpenAI till day 10

  • Pika Labs pushes boundaries with a powerful V2 vid …

  • Microsoft's new Phi-4: redefining small language m …

• The Tools

• The Algorithm

• The Output

The Input

Meet NVIDIA's latest: A supercomputer that won't break the bank

NVIDIA has introduced the Jetson Orin Nano Super Developer Kit, a compact and powerful generative AI supercomputer priced at just $249- half the cost of its predecessor. The device offers a 1.7x performance boost, 67 INT8 TOPS, and increased memory bandwidth, making it an ideal tool for hobbyists, students, and commercial developers exploring AI, robotics, or computer vision.

Key highlights:

• Enhanced Performance: Designed to handle popular generative AI models, it supports multiple AI pipelines thanks to its NVIDIA Ampere GPU and 6-core Arm CPU.

• Developer-Friendly: Fits in the palm of your hand and supports prototyping for edge AI applications.

• Software Ecosystem: Access NVIDIA AI software like Isaac, Metropolis, and Holoscan, plus tools for synthetic data generation and pre-trained model fine-tuning. (source)

Watch the video by Jensen Huang to get an idea of just how small it is. :)

Veo 2, Imagen 3, and Whisk: The future of visual storytelling

Google has unveiled exciting updates to its AI creativity tools: Veo 2, Imagen 3, and a new experimental feature called Whisk.

Veo 2 sets a new standard in video generation with enhanced realism and cinematic precision. The model produces 4K-quality videos, expertly interpreting prompts like “shallow depth of field” or “18mm lens” to deliver professional results. Whether it’s capturing dynamic human expressions or surreal abstract scenes, Veo 2 creates stunning visuals while embedding SynthID watermarks for safety and authenticity.

On the image front, Imagen 3 enhances Google’s image generation capabilities. Known for its vibrant colors and intricate detail, the model faithfully interprets prompts to produce a range of styles, from photorealistic to artistic. Now available globally through ImageFX, Imagen 3 empowers users to create with even greater control and creativity.

The newest addition, Whisk, takes creativity a step further by allowing users to remix and transform images. By combining visual inputs with Imagen 3’s advanced rendering and Gemini AI’s understanding, Whisk enables unique outputs, from digital art to custom collectibles. (source)

Key highlights of OpenAI till day 10

As December comes to a close, OpenAI's Shipmas event has been a thrilling ride, packed with game-changing innovations in AI. A quick roundup of the exciting updates for you:

• Day 6 introduced ChatGPT's Advanced Voice Mode with video capabilities, allowing users to engage in dynamic interactions, like receiving step-by-step guidance on tasks such as making pour-over coffee. This feature builds on OpenAI’s vision of blending text, audio, and video for a seamless experience.

• On Day 7, OpenAI launched Projects, a tool to help users organize conversations, upload files, and store chats. It's a major boost for productivity, empowering users to tailor their ChatGPT experience for both personal and professional use.

• Day 8 marked the global rollout of SearchGPT, making search a more integrated, conversational experience for all logged-in users. The tool’s ability to provide search results in real-time while chatting is a game-changer, and the feature is now even faster and more mobile-friendly.

• Day 9 was all about developers, with OpenAI unveiling o1, a series of AI models designed to tackle complex tasks. This launch, along with enhanced SDKs and new API tools, opens up new possibilities for developers to create agentic applications.

• Finally, Day 10 brought ChatGPT to your phone via calls and WhatsApp. Whether on a smartphone or even a rotary phone, users in the US can now connect with ChatGPT via voice for seamless access without an internet connection. (source)

Pika Labs pushes boundaries with a powerful V2 video model

Pika Labs has launched Pika 2, the next iteration of its AI video model, promising unmatched customization and control. Known for fun, social content like melting objects or exploding cakes, Pika Labs now targets OpenAI's Sora, the current leader in AI video creation, with technical capabilities that rival its competition.

Key Highlights of Pika 2:

• Scene Ingredients: Users can upload images to control characters, objects, and settings in their videos, seamlessly integrating them into customized shots.

• Advanced Prompt Adherence: The model follows detailed prompts to generate videos without skipping important elements, even for complex scenarios.

• Improved Motion Handling: Enhanced physics understanding enables more realistic and fantastical motion, such as flying humans or alien worlds.

Brands like Balenciaga and Fenty, along with influencers, have already embraced Pika's creative AI. With v2’s innovations, it’s set to redefine AI video creation for everyone from casual users to industry professionals.(source)

Microsoft's new Phi-4: redefining small language models

Microsoft has introduced Phi-4, a 14-billion-parameter small language model (SLM) designed to excel in complex reasoning and mathematical tasks. Despite its relatively compact size, Phi-4 demonstrates remarkable capabilities, surpassing even larger models like Gemini Pro 1.5 in math problem-solving, thanks to its high-quality synthetic and organic training data and advanced post-training techniques.

Key features of Phi-4:

1. Exceptional Math Performance: Phi-4 outperforms larger models on math competition problems, showcasing its ability to tackle complex reasoning with precision.

2. Responsible AI Integration: The model incorporates features from Azure AI Content Safety, such as prompt shields and groundedness detection, to ensure safe, responsible, and reliable use.

3. Accessible Innovation: Phi-4 supports developers with tools for AI risk management, model evaluation, and real-time quality monitoring, fostering responsible deployment and real-time safety checks.

Phi-4 is now available on Azure AI Foundry under an MSRLA and will be launched on Hugging Face next week, marking a significant step forward in AI model development and deployment. This compact model demonstrates the potential for small language models to deliver high-quality performance in advanced AI applications. (source)

The Tools

Tool: Poe

Unlike chat platforms focusing on a single AI model, Poe aggregates multiple models, giving users the flexibility to switch between different capabilities. This makes it a go-to platform for users who want to experiment with or compare AI responses.

How to Access

• Log in to Poe.

• Enter your prompt.

• Generate responses.

• Compare answers from multiple open-source models.

• Choose the best response.

The Algorithm

• In the recent episode of Leading with Data, I had an engaging conversation with Anand Ranganathan, Chief AI Officer and Co-founder of OneByZero. With over a decade of experience, he shared insights on his journey in AI, transitioning from IBM to entrepreneurship, founding Unscramble, adapting to startup dynamics, key lessons, future AI trends, and advice for aspiring professionals.

The Output

As we wrap up this eventful December, it's clear that the pace of AI advancements has been nothing short of exhilarating. We've seen incredible strides, and it's only the beginning. Now, let's take a moment to absorb all that's happened, reflecting on how far we've come. With the future unfolding before us, we eagerly await what exciting developments AI has in store for us next. The journey is just beginning, and there's so much more to explore.

We also want to take this opportunity to wish all our readers a very happy New Year! We look forward to meeting you again next year with more exciting news, in-depth coverage, and insightful discussions that we hope will continue to inspire and inform. Here's to a bright and innovative year ahead!