OpenAI’s o1 launch, Amazon Nova foundation models, NotebookLM leaders exit Google, Meta’s Llama 3.3 release — the top 4 AI news stories of the week

#1 — OpenAI launches the o1 model and announces ChatGPT Pro

OpenAI has officially moved its o1 reasoning model out of the preview phase and made it a core part of ChatGPT. LRMs differ from LLMs in that they “think” more and correct their answers - which means they can “solve complex reasoning problems”. One particularly cool new feature is the ability to analyze images, so if you’re a visual thinker or need help solving complex problems, this model could be incredibly handy. I’ve tried it on some tough mathematical modeling tasks, and it performed surprisingly well. o1 will first be available to users with Plus and Team subscriptions, and then enterprise and education customers will have access next week.

In addition, OpenAI’s o1 has introduced a new approach called reinforcement fine-tuning, which allows businesses to create specialized “expert” models. Instead of a more generic AI approach, RFT focuses on building a highly trained AI assistant just for your particular field or unique tasks. I predict this will be a huge trend next year, as companies seek more personalized AI solutions.

Finally, OpenAI announced its new ChatGPT Pro subscription, priced at $200 per month, which provides users with access to all its latest “models and tools” – a collection that now includes o1.

#2 — Amazon reveals Nova foundation models

Amazon has stepped into the spotlight with its new Amazon Nova family of multimodal, generative AI foundation models. These models aren’t just about text — they can handle images and videos as well. While Amazon’s benchmark claims might sound almost too good to be true, it’s clear they’re aiming to become a major player in the AI space. Integrated with Amazon Bedrock, these models come in flavors like: Micro, Lite, Pro, and Premier, with different sizes and capabilities for text generation; the image-generating Canvas; and the video-generating Reel. The promise? Faster performance, lower costs, and powerful customization options like fine-tuning. This means developers can more easily build apps that understand and create multimedia content, from stunning visuals to rich video-based experiences.

#3 — NotebookLM leaders leave Google for their own AI startup

NotebookLM, a Google project that gained a lot of buzz, just lost three of its key figures — former team lead Raiza Martin, designer Jason Spielman, and engineer Stephen Hughes. They’ve jumped ship to start their own, completely stealthy, AI venture. We don’t know much about it yet — not even its name — but they’ve suggested that it will be all about making advanced AI models more directly useful to average people. If NotebookLM’s success at turning notes and podcasts into smarter digital tools impressed you, keep an eye on what this team does next. It might just push consumer AI even further into our daily lives.

#4 — Meta’s Llama 3.3 multilingual LLM arrives

Meta has introduced Llama 3.3, a text-only multilingual large language model that supports eight languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai. With a massive 70B-parameter setup and training on 15 trillion tokens, this model is all about cost-effective scalability, improved reasoning, and broader language support. For anyone working in global markets or dealing with diverse user bases, Llama 3.3 might help create more inclusive and intelligent AI experiences across multiple languages.