OpenAI's new o1 model: A leap in AI reasoning

Just days ago, the AI world was buzzing with the announcement of OpenAI's secretive "Strawberry" project. Now known as the o1 model, this AI powerhouse has shattered benchmarks. But does it live up to the hype? We take a look at o1 below.

 By the CogX R&I team

September 18, 2024

OpenAI has just unveiled o1, a groundbreaking AI model that's making waves in the world of AI. Codenamed "Strawberry" during development, this latest offering is set to redefine what we expect from AI when it comes to tackling complex tasks in STEM fields.

The o1 model isn't just another incremental update: Early tests show it performing on par with PhD students in physics, chemistry, and biology. But what really sets o1 apart is its ability to recognise mistakes and improve its responses, giving it a significant edge in analytical tasks.

OpenAI's o1 model has demonstrated impressive intelligence, particularly in its performance on the Norway Mensa IQ test. Source: Maxim Lott

Scaling compute not only at training time but also at inference time: OpenAI's approach with o1 focuses on "reasoning" rather than just generating language. This means the AI can now tackle multi-step problems with a more reflective, deliberative process. It's like giving an AI the ability to think before it speaks, the longer it thinks, the more accurate and insightful its outputs get.

However, o1 isn't without its limitations. Compared to its predecessor GPT-4o, it lacks some key features like web browsing, file uploads, and image processing. It also doesn't support API functionality for tool usage, function calling, and streaming – features that many developers and enterprises have come to rely on.

ChatGPT Plus and Team users can preview the o1 model by selecting 'o1-preview' in the model selector. For those with a STEM focus, there's also the "OpenAI o1 mini" variant, designed for faster responses in maths and science.

Now read the rest of the CogX Newsletter

Will AI take over? A conversation with Jaan Tallinn

AI, as we are repeatedly told, is poised to revolutionise our world. However, alongside promises of innovation and technological development, concerns about its potential risks are also mounting.

Lawmakers worldwide caution about potential misuse of AI systems by malicious actors, while AI lab employees voice safety concerns publicly. Even some of the world’s leading AI experts believe that preventing AI-related risks should be a global imperative.

The rapid progress in AI has ignited concerns that extend far beyond the specter of job losses. We are now grappling with the potential for AI to become a tool for spreading misinformation at an unprecedented scale, influencing our democratic processes, and even outpacing human intelligence in ways we can scarcely imagine. At its most extreme, this anxiety extends to the possibility of advanced superintelligent AI systems posing a fundamental threat to human existence itself.

To shed light on these pressing issues, we engaged in a conversation with Jaan Tallinn about the current state of AI development, its trajectory, and the steps needed to ensure it becomes a force for good rather than an existential threat.

 … want to keep reading? Check out the full OpEd here on the CogX Blog

Google's AI transforms documents into personalised podcasts

Source: NotebookLM

The Audio Overview tool leverages Google's advanced Gemini AI models to not just summarise content but to create a dynamic, conversational experience around it. Unveiled at Google I/O and integrated into the NotebookLM note-taking app, this AI-powered system aims to cater to auditory learners — and those who prefer listening over reading.

Key features of the Audio Overview tool include:

• AI-generated synthetic personalities discussing document content, complete with banter and jokes.

• Personalised learning experiences tailored to individual preferences.

• Ability to convert various document types, including slides and text files, into audio format.

While currently focused on academic and research applications, the technology's potential extends to personalised news reports and social media updates, hinting at a future where AI-driven audio content becomes increasingly prevalent.

Also in the news

Mistral unveils new multimodal AI model: French AI startup Mistral has released Pixtral 12B, a multimodal model capable of processing both text and images. This new model offers a range of potential applications, including image captioning and object counting.

Google launched Ask photos: A new AI-powered search feature that allows users to ask more complex questions about their photos. This feature, powered by Google's Gemini AI model, can help users find specific photos, identify themes, and even recall past events.

Adobe's new AI video tools: Adobe has showcased its upcoming generative AI video tools, including text-to-video and image-to-video capabilities. These tools offer users more control over video creation, allowing for adjustments to angles, motion, and shooting distance.

Tencent unveils GameGen-O: A leap towards AI-powered open-world game generation.

Source: Tencent 

Chinese tech powerhouse Tencent is pushing the boundaries of AI-generated gaming with GameGen-O, a revolutionary system that could redefine how we create virtual worlds. This cutting-edge diffusion transformer model can bring entire open-world games to life, complete with dynamic characters, environments, and complex events.

Here are some fascinating aspects of the GameGen-O project:

• The AI trained on data from over 100 next-generation open-world games

• GameGen-O can generate subsequent frames based on complex instructions, hinting at the potential for interactive storytelling

  
  

Though current outputs are more akin to short video clips than fully playable games, Tencent researchers view this as the first step on an exciting journey. While Google recently made waves by using AI to recreate Doom, Tencent's leap forward suggests we're accelerating towards a future where AI becomes the ultimate game designer.

In case you missed it

Deem Mind unveils a new robot powered by diffusion technology.