As industry leaders vie for the top spot with the year drawing to a close, here is December’s AI news.
One of the biggest features of this month was the back and forth between Google and OpenAI. Seemingly, trying to one up each other at every turn. OpenAI kicked off the holiday season with their version of ’12 days of Christmas’, releasing new features almost daily, while Google dropped an unruly amount of releases at the same time.
The race for supremacy has always been evident, but this month, it reached a whole new level.
Google’s flood of releases
Google went above and beyond in terms of releases this month, arguably making it the strongest platform for a variety of AI features. Their most disruptive release was their new Gemini 2.0, which emphasises agentic use, meaning the Large Language Model (LLM) can operate without human interference to solve multi-level problems and can act independently. It also now has access to sight, enabling it to interact with the world around it.
One of the most innovative ways google have used it is to create NotebookLM, which helps users engage with their documents by generating summaries, providing explanations, and answering questions based on the uploaded content. The content can be anything from PDF documents to video and audio files.
It also added a “deep-research” feature to their LLM, which allows it to search the web for all sorts of information. In their words: “Deep Research does the hard work for you. After you enter your question, it creates a multi-step research plan for you to either revise or approve. Once you approve, it begins deeply analysing relevant information from across the web on your behalf.” The tool is expected to become incredibly useful in scientific research in particular.
Google did not stop there though, they also released new versions of their image and video generation models with Veo 2 and Imagen 3, the quality of which will have some people in Hollywood licking their lips, while others will be terrified. Here is just one of the many examples they gave upon release:
Moreover, Google has made tremendous leaps in terms of weather forecasting, now able to more accurately predict the weather using AI technology for up to 15 days in the future, called GenCast. This could be a revolutionary new discovery in the age of climate change and the growing number of climate-induced disasters.
They also released a feature which will likely have gamers on the edge of their seats as with their new Genie 2, a new “large-scale foundation world model”, which is able to generate different playable worlds through simple prompts. Currently the model can only generate one minute of play, but you can already imagine the potential this technology will have in future. World Labs, run by Fei Fei Li, also released a similar model which is arguably even better.
Last but not least, something you have probably already seen – as it was everywhere – is Google’s Willow project. Their state-of-the-art microchip completed a benchmark test 10 septillion times faster than today’s supercomputers. It also utilises more qubits than previously possible, significantly reducing quantum errors, which have long been the main challenge in achieving large-scale quantum computing.
’12 days of OpenAI’
A novel approach to tech releases, OpenAI seized on the end-of-year holiday spirit to launch their own version of 12 days of Christmas. Sam Altman, ever the business man, kept the tech world on its toes with this niche marketing ploy, each day releasing an intimate video with employees announcing the new features on Youtube.
The main release happened on the final day which is their newest LLM o3 and o3-mini, it is not yet open for public use but to safety and security researchers. It is set to make great leaps in the reasoning capabilities of such models, even leading to some influential people in the sphere to call it AGI (Artificial General Intelligence).
However, this depends on the fundamental definition of AGI, and OpenAI themselves are yet to officially claim it. The results are impressive though, with the model achieving almost perfect scores in benchmark maths, science, and coding tests.
They also – finally – released their video and image generation model: Sora. The early reviews between Google and OpenAI’s models suggest that the former performs to a better standard. Which means they will be trying to better their product in a more vigorous manner.
More exciting releases include:
For an overview of all their releases head over to their YouTube channel.
Doctor’s 30%, AI 80%
This one needs to be taken with a pinch of salt, as the paper in question is only a preprint, meaning it has not yet been peer reviewed. The research aims to evaluate the performance of medical diagnostic methods and was performed by Harvard Medical School and Stanford researchers, the full paper can be found here.
However, if the findings do come back to be true, new reasoning models will be taking huge leaps in medical detection. These tests were performed using OpenAIs o1-preview, so you can only imagine the results using models such as o3 or even more advanced models.
Genesis project
This project is likely to have great implications for the future, especially in regards to robotics through simulations of real world physics. It is a generative AI model similar to an LLM or video generation model, as it generates a product. This, however, is a generative physics engine capable of producing a 4D physical world. It was developed by a Carnegie Mellon University PhD student and was backed by NVIDIA.
Not only is there a race for supremacy in terms of AI, but there is also a race to develop the most efficient humanoid robot, with companies like Boston Dynamics, Tesla, and multiple Chinese companies battling it out. With technology like this being harnessed in combination with robotics, 2025 is set to be a groundbreaking year!
Jailbreaking prompts
The creation of this new feature by Anthropic (creators of Claude.ai, one of the leading companies in AI), has some people quite excited, but I believe it could have a detrimental impact. Can you remember when people would jailbreak their iPhones or Ipods, and it would make the software do all sorts of new tricks. Well Anthropic’s new ‘Best-of-N Jailbreaking’ does just that but for AI prompting.
When asking ChatGPT or Claude on how to create drugs then it will tell you that it is not able to perform this task, now Anthropic have developed a code that will bypass the security measures of these LLM’s for text, video, and image. Meaning you could literally get answers to any prompt you like. It does so by slightly changing the characters of the prompt, say making an E into a 3, or adding slight spelling mistakes, and then continues to prompt the model in different ways until it finally answers.
MIT energy efficient chip
One of the main topics of discussion when speaking of the success and rise of AI, is that it consumes incredible amounts of energy, one that many skeptics enjoy highlighting. This has seen many tech giants turning back to nuclear energy in a bid to find new solutions, however, the boffins at MIT may have come up with a clever solution bvy using a photonic processor to be more energy efficient.
The way in which AI is currently operated uses huge amounts of hardware and tremendous amounts of energy, but using these chips could minimise the amount necessary with something we have in abundance, namely light. The research was funded by the U.S. National Science Foundation, the U.S. Air Force Office of Scientific Research, and NTT Research, underscoring the high level of importance and seriousness attributed to this work.
OpenAI and Anduril partnership
An aspect of AI that was never going to be avoided was its use as a weapon, hence leading American defence technology company Anduril has now signed an agreement with OpenAI to “develop and deploy advanced artificial intelligence solutions for national security missions”.
They say this is comes in response to the threat that the Americans see China posing and wanting to stay ahead in terms of weapons technology. This will largely be based on creating lethal autonomous weapons, so we might soon be living in a future where robots will be fighting wars for us.
The future
2024 has been an incredible year in terms of AI. Through the developments that have been made this year, they will be able to build on these and make new breakthroughs. What will likely be the most important aspect of 2025, according to many industry leaders, is the large scale use of agents. OpenAI are rumored to release theirs in the coming months and it will surely be the Eureka moment, where the masses start to open their eyes to this agentic future. Each individual having access to an assistant in their pocket (see below a behind closed doors look at OpenAI’s unreleased agents). Another feature that many people are talking about is “infinite memory” which entails your LLM remembering everything that you tell it in real time. So you will have an assistant that never forgets. Finally, there is so much talk about robotics and this will likely be the year that brings them to the forefront of our society.
The world is changing at a rate we have never seen and it is imperative to stay in the know.
Find all previous month’s AI news here.
Find all previous month’s AI news here.
