Conventional marketing wisdom suggests a lull in activity during the holiday season. Evidently, this memo was lost on the innovators in the generative AI arena. The latter half of December witnessed a flurry of groundbreaking announcements, underscoring an industry that cannot afford a pause given the rapid developments and high costs for those who fall behind.

An aside: That conventional marketing folks are not running the show in the GenAI fields was evident in many different forms in OpenAI’s series of announcements since our last note, but none more than their leap from their O1 model straight to O3. They seemingly skipped O2 altogether, perhaps the most vital element in the 'O' series from a marketing viewpoint! This highlights how the fast-paced engineering culture of generative AI firms outpaces conventional marketing wisdom, emphasizing substance over style.

It isn't easy to list all the announcements we came across while on the pen down. Here is a filtered ten that stayed with us:

10. Google Raises the Bar with Gemini 2.0 (December 11th)

In our view, the tenth most important announcement this December came from Google with the introduction of Gemini 2.0. This isn't just another incremental update; it's a full-fledged base model upgrade from the current 1.5 series. For the tech-savvy, this signifies much more than meets the eye. Base model upgrades, even those seemingly small half-step increments, represent substantial architectural advancements and performance gains compared to minor updates within the same model series.

While the rest of this post will delve into specific features and innovations, it's crucial first to acknowledge Google's dominant position in the AI landscape. Their models consistently rank at the top of most Western model performance benchmarks, and Gemini 2.0 is poised to solidify their leadership further.

9. Gemini 2.0 Flash: Your Phone's 2025 AI Brain (December 11th)

It was more than the base model change as Google unleashed Gemini 2.0 Flash, a specialized model built for speed and efficiency. Imagine this: your phone seamlessly integrates AI into everything you do. Flash makes this possible. Need to identify a landmark in a photo quickly? Done. Want real-time translation during a conversation with someone speaking a different language? No problem. Flash's speed and multimodal capabilities mean AI can understand and respond to your needs instantly, whether it's through text, images, or audio. This is a game-changer, paving the way for a future where our phones become knowledgeable assistants, capable of understanding and responding to the world around us in ways we've only dreamed of.

8. Instant Copiability: The AI Arms Race Heats Up (December 12th)

A day after Google unveiled Gemini 2.0 Flash and its potential to transform our phones into AI-powered assistants, OpenAI showcased almost identical capabilities. OpenAI's model can process live camera data, provide real-time feedback through voice, and even share screens, essentially turning your phone into an AI-powered "eye" that can understand and interact with the world around you. The simultaneous announcements of similar features might be accidental, but they highlight the intense competition and the speed at which AI technology is evolving in multiple environments globally. Companies must now prioritize rapid iteration and continuous improvement to stay ahead in this ever-accelerating AI arms race.

7. Google Whisk: The "Point-and-Shoot" Revolution Comes to Image Editing (December 16th)

Google's Whisk, a preliminary lab product, hints at the future of image editing. It allows users to edit images using other images as prompts, essentially "remixing" visuals innovatively. Text-to-image now includes working with, merging, and altering your own images for new visual presentations that previously required sophisticated software and expertise. While AI tools won't replace expert tools, they're akin to the arrival of point-and-shoot cameras in the early 2000s, democratizing features previously exclusive to high-end SLRs. This signals a broader trend: AI rapidly changes how we interact with digital content, posing risks for all established software players, processes, and expertise.

6. Quantum Leaps: Exciting Progress, But Patience is Key (December 5th - 30th)

December saw a flurry of quantum-related announcements, from Google's demonstration of "time crystals" on December 5th to IBM's ambitious roadmap for quantum computing on December 30th, which includes the release of their largest quantum computer yet in 2025 and a quantum-centric supercomputer. We also witnessed breakthroughs in quantum energy teleportation and near-instantaneous data transfer across entangled quantum nodes. While these advancements significantly accelerate the field, their practical applications may not be fully realized by 2025. The market's enthusiasm for quantum technology may ebb and flow, but for long-term investors, these developments underscore the transformative potential of quantum computing and communication. Patience will be key, as the true impact of these breakthroughs may unfold over the coming decades, not months.

5. Collaborative AI: OpenAI Projects and Google Canvas Reimagine Teamwork (December 13th & 14th)

OpenAI and Google both launched platforms for collaborative AI projects in mid-December. OpenAI's "Projects" feature allows users to create shared workspaces with persistent chat history and dedicated AI models for specific tasks. Google's Canvas provides an AI-powered collaborative brainstorming, writing, and editing canvas. These tools signal a shift towards AI-augmented teamwork, where humans and AI collaborate seamlessly on complex projects. Imagine brainstorming sessions with AI generating ideas or writing projects with AI providing real-time feedback and suggestions. This could revolutionize software development, design, and research, enabling teams to work more efficiently and creatively. Microsoft also jumped into the fray with Azure Communication Services enhancements (December 20th), enabling developers to integrate AI capabilities like real-time transcription and translation into their collaborative applications seamlessly.

4. O3: The Code Whisperer - AI Transforms Programming, Again (December 22nd)

OpenAI's O3 represents a significant leap forward in AI-assisted programming. Built on "chain of thought" reasoning, O3 can break down complex coding tasks into logical steps, enabling it to generate entire code blocks from simple prompts, debug intricate codebases, and even optimize algorithms for better performance. For example, a user could ask O3 to "write a Python function to process large datasets efficiently," it would generate the code and explain its logic and approach. This translates to significant gains in programming efficiency, potentially reducing development time and costs. O3 can also refactor existing code for clarity and scalability, automating tasks that previously consumed hours of developer time. This has profound implications for the software development landscape, empowering junior developers and freeing up senior programmers to focus on higher-level design and innovation. While still in its early stages, O3 demonstrates the transformative potential of AI in coding, paving the way for a future where software development is faster, more efficient, and more accessible.

3. O3: Redefining the Limits of AI and AGI (December 22nd)

OpenAI's O3 made waves by exceeding human performance on the ARC-AGI benchmark, a test designed to assess general intelligence. While not true AGI, this demonstrates AI's remarkable progress in tackling complex tasks once considered exclusive to humans. O3's success stems from its "chain of thought" reasoning, allowing it to break down problems and synthesize new approaches. This shows that AI can adapt and learn in previously unimaginable ways. As models like O3 continue to push boundaries, we can expect them to tackle increasingly complex challenges, from scientific research and creative problem-solving to tasks requiring nuanced understanding of human language and behavior. While AGI may remain a distant goal, AI models are constantly knocking down the challenges we pose, paving the way for a future where AI plays an increasingly integral role in our lives.

2. Robotics LLMs: The Dawn of Truly Intelligent Robots (December 18th)

A team of researchers unveiled a groundbreaking development: Large Language Models (LLMs) designed specifically for robots. These models allow robots to understand and respond to complex instructions in natural language, bridging the gap between human commands and robotic actions. Imagine telling a robot to "tidy up the living room," and it understands the individual tasks involved and the nuances of your preferences. This breakthrough can potentially revolutionize various industries, from manufacturing and logistics to healthcare and home assistance.While this specific research is leading the charge, it's important to note that this is a trend that other major players in the AI space are also pursuing. Companies like NVIDIA, focusing on robotics simulations and AI-powered robot control, and OpenAI, with their ongoing research in reinforcement learning and robot manipulation, are likely to incorporate LLMs into their robotics development. This signifies a significant shift in the field, where robots are no longer just programmed to perform specific tasks but can understand and respond to human language, making them more versatile, adaptable, and collaborative partners in various environments.

1. Usage Costs: A Seismic Shift in the AI Landscape (December 25th)

Throughout 2024, we witnessed a dramatic decline in LLM inference or AI usage costs. As AI expert Simon Willson highlighted, these costs plummeted 27x in 2024. He demonstrated how he could generate detailed text descriptions for 68,000 photos in his library for under $2! But just as we were absorbing this trend, A Chinese developer unveiled a further game-changer: DeepSeek-V3, a model that rivals the performance of leading LLMs, like those from Meta or Anthropic, but was trained for a mere $6 million. This cost is less than a tenth of the training cost of the rivals.

As a result, DeepSeek-V3 can offer the token or usage pricing at a fraction of the rivals’, too. After the already crashed usage prices in 2024 by around 97%, the year starts with another 90% fall on top. The pessimists will surely be more concerned for LLM makers, particularly those who haven't prioritized cost efficiency. However, this dramatic cost reduction is a powerful catalyst for mass adoption.

The AI Tsunami: Hold On Tight!

It's tempting to declare "mission accomplished" after compiling this top 10 list, but the truth is, we've barely scratched the surface of December's AI explosion. Our list doesn't even account for OpenAI's audacious new $200-per-month product (a potential game-changer in accessibility), the mind-blowing video generation capabilities of ChatGPT’s Sora or Google’s Veo 2 (both released around December 15), or the staggering advancements in "agentic" AI, including Devin AI’s official launch on December 11. There were substantial releases from Apple and Microsoft, and Nvidia, too, was doing its bit!

The sheer volume and velocity of breakthroughs in December paint a clear picture: AI is thriving with an intensity rarely witnessed in human history. It's a multi-front revolution, with progress erupting in every direction. Whether it's the democratization of image editing, the rise of collaborative AI, robots gaining the ability to understand us, or the relentless drive to push the boundaries of what AI can achieve, the message is clear: buckle up because the AI tsunami is gathering momentum. The future is hurtling towards us at an exhilarating pace, and those who embrace this wave of innovation will be the ones who ride it to success.