OpenAI adds o1 AI Model to its API with new features

Hello, AI Enthusiasts!

Welcome to FavTutor’s AI Recap! We’ve gathered all the latest and important AI developments for the past 24 hours in one place, just for you.

In Today’s Newsletter: 😀 

• OpenAI adds the o1 AI Model to its API with new features

• Google's Whisk AI tool can combine multiple images for generation

• Meta's new LLM architecture tackles fundamental flaws in LMs

OpenAI
⚡️ OpenAI adds the o1 AI Model to its API with new features

OpenAI brings an updated o1 model to its API, adding features such as function calls, JSON-formatted structured output, and image analysis capabilities. The new model, dubbed "o1-2024-12-17," significantly outperforms the previous "o1-preview" version.

Insights for you:

• OpenAI presents the new o1 model "o1-2024-12-17" for the API and ChatGPT, which achieves an accuracy of 96.4% for mathematical tasks and 76.6% for programming tasks and requires an average of 60% fewer tokens for reasoning tasks than its predecessor.

• o1 will begin rolling out to devs in OpenAI’s “tier 5” usage category. o1 replaces the o1-preview model that was already available in the API.

• o1 in the OpenAI API is far more customizable, with new features like function calling, image analysis, and an API parameter, that enables control over how long the model thinks before responding to a query.

Need a personal assistant? We do too, that’s why we use AI.

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI into your own processes, allowing you to optimize your time and resources, while maximizing your output with ease.

Get the free guide here.

Google
📸 Google's Whisk AI tool can combine multiple images for generation

Google Labs has released Whisk in the US, its latest generative AI experiment. Whisk lets you input images for the subject, one for the scene and another for the style. If you are based in the US, you can try it out today at labs.google/whisk,

Insights for you:

• Google Labs has launched a new gen AI experiment called Whisk in the USA, which allows users to create images primarily using visual inputs for subject, scene, and style, rather than relying on lengthy text prompts.

• Whisk utilizes Google's language model to automatically generate a detailed description of the input images, which is then processed by Google's Imagen 3 image generation model.

• Since Whisk extracts only a few key characteristics from your image, it might generate images that differ from your expectations.

Meta
💡 Meta's new LLM architecture tackles fundamental flaws in LMs

Meta has developed a new AI architecture called Byte Latent Transformer (BLT) to solve a fundamental problem with today's language models: they can't reliably work with individual letters. Using just 8 billion parameters, the system outperforms Llama 3.1, despite Llama having trained on 16 times more data.

Insights for you:

• Meta has developed the Byte Latent Transformer (BLT), a new LM architecture that operates directly at the byte level instead of splitting words into predefined tokens, allowing for more precise handling of individual characters.

• By using bytes, BLT has direct access to individual letters, punctuation marks, and special characters, which is beneficial for tasks such as spelling correction, character counting, and processing new character sets.

• To mitigate the increased computational overhead of byte processing, BLT dynamically combines bytes into patches, with the patch size adapting to the complexity of the data, making BLT more efficient than token-based architectures.