Oops: Did OpenAI’s o1 model get leaked?

Along with: Can ChatGPT’s new search integration challenge Google’s dominance in search engines?

Hey there, 

I spent a week in South Africa and was blown away by the Energy and willingness of people to learn AI! I met a few start-ups working on African language models and learned about the exciting work they were doing.

I also spoke at the AI Expo 2024 about creating world-class AI Mastery programs and released a book on the same. You can access the book here!

In the meanwhile, Trump is back as the President of the US. It would be interesting to see how he tackles the age of AI in his coming tenure.

Now moving forward with this week’s newsletter:

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Tools - Interesting finds and launches

  • The Algorithm - Resources for learning

  • The Output - Our reflection 

Table of Contents

As per the report published by Tom’s Guide, OpenAI’s latest AI model, called “o1,” briefly leaked online when users discovered a URL tweak that provided access to the full version. 

Initially released in preview form to paying users, the model has impressed early users with enhanced capabilities, such as solving complex math problems, analyzing images, and processing large data files, surpassing the limitations of the preview. 

OpenAI confirmed that the leak occurred during preparations for limited external access but has since been resolved. The model’s official full release date remains uncertain, though experts are optimistic about its advances in reasoning and problem-solving capabilities. (source)

ChatGPT had previously launched SearchGPT and now has integrated the search feature into ChatGPT positioning it as a competitor to Google, Perplexity, and other search engines. 

With enhanced memory, real-time web browsing, and multimodal support, it provides accurate, conversational responses with links to trusted sources. 

This update lets users find quick answers, stock updates, and more within the chat. Unlike Google’s list of search results, ChatGPT offers synthesized, in-depth answers, while its multimodal abilities and Custom GPTs offer a more tailored, interactive experience than Perplexity. These features make it highly effective for both search and complex tasks. (source)

OpenAI has introduced SimpleQA, a new open-source benchmark for evaluating the factual accuracy of language models. It focuses on clear, short questions with a single answer, challenging even advanced models like GPT-4 and Claude. 

With over 4,300 questions, SimpleQA tests models across various domains and measures precision and calibration. The results show that models still struggle with factual accuracy, even as they improve at recognizing when they are unsure. SimpleQA aims to encourage more reliable and truthful AI responses. (source)

Apple recently launched Apple Intelligence, its highly anticipated generative AI software, with the first set of features available as of October 28, exclusively on iPhone 15 Pro and newer models. Apple plans to expand this AI software through updates in December and April, aligning with iOS 18.2 and iOS 18.3. Here’s what to expect:

  • Enhanced Writing Tools: Improved to handle more specific user requests.

  • ChatGPT Integration: Enabling richer, conversational AI capabilities.

  • Visual Intelligence: A new Camera Control feature on iPhone 16, offering object recognition and language translation.

  • Customizable “Genmojis”: Expected in December, allowing users to personalize emojis with AI.

Apple’s approach contrasts with the rapid launches of its rivals, as it continues to roll out its AI features gradually for refined user experience and added functionalities.(source)

Oasis, the first fully AI-generated, real-time open-world game, has launched, redefining possibilities in gaming technology. Built without a traditional game engine, Oasis generates gameplay, visuals, and physics entirely through AI in real-time based on user input. 

The model is designed with a transformer-based architecture, allowing users to experience diverse environments and complex interactions. This launch marks a major step towards fully AI-generated interactive worlds with potential applications far beyond gaming.(source)

Google introduced its new "Learn About" feature which uses AI to help users explore various topics, from koi fish to cooking, with both simple and detailed answers. 

While the tool provides organized, digestible information and interactive prompts, its simplified mode sometimes redirects users to e-commerce sites, which could disrupt the learning experience. The in-depth mode, however, offers detailed information in a clean, ad-free format. Though promising for both casual and young learners, improvements are needed to ensure a more seamless, ad-free experience. (source)

Runway has launched an Advanced Camera Control feature for its Gen-3 Alpha Turbo video editing platform, allowing users to precisely pan, zoom, and track around AI-generated subjects to create intentional, immersive shots. 

The tool, which costs $12 per month, keeps AI subjects consistent as the camera moves, enhancing storytelling and context within scenes. (source)

Osmo has introduced "scent teleportation" technology, allowing users to send and receive scents digitally. 

By analyzing scents and creating a digital “scent code” using AI, the technology reproduces aromas through a "scent synthesizer" device that blends stored molecules to recreate smells. 

These scent codes are sent over the internet, enhancing VR, therapy, education, and personal messaging with sensory experiences. Challenges include improving scent storage and synthesis, but Osmo envisions this as a new layer of immersive digital communication. (source)

Meta’s FAIR team has released new AI research artifacts to advance robotic touch perception and human-robot interaction. Its key innovations include:

  • Meta Sparsh, a general-purpose touch model for diverse tasks and sensors.

  • Meta Digit 360, a tactile fingertip sensor with human-level precision.

  • Meta Digit Plexus, a platform for integrating various tactile sensors onto a single robotic hand.

These technologies aim to improve machine dexterity, aiding applications from healthcare to VR. (source)

Imagine sending a memory-not just in words or images, but in scent. Osmo’s scent teleportation lets you transmit aromas digitally, adding a hypnotic, sensory layer to how we connect and immerse in the virtual world.

It digitizes smells by “reading” the molecular composition of scents, “mapping” them using AI, and “writing” or recreating them through synthesized molecules.

To access the scent generation tool:

  1. Log in or sign up.

  2. Paste the URL of an image representing the scent you want.

  3. Click "Generate."

  4. View the predicted scent formula.

Here’s the scent formula generated from my picture of rain. Link

  1. If you are interested in building AI-driven teams and want actionable insights on designing and implementing effective AI Mastery programs in your organization, the AI Mastery Playbook by Kunal Jain will give you a deeper understanding of how to develop and integrate these strategies.

  2. To everyone curious about AI's impact on the future, this video from future investment initiatives offers invaluable insights from leaders at the forefront of AI innovation, sharing their perspectives on industry disruptions, responsible AI use, and the rapid advancements shaping tomorrow.

What do you think is coming from Trump on AI? 

He hasn’t said much about it. I think he will go “nuclear” to solve the Energy challenges and will likely keep a light regulatory environment.

Is he going to appoint Elon as an AI advisor? What do you think are the chances?

Will wait to hear from you!

How do you rate this issue of AI Emergence?

Would love to hear your thoughts

Login or Subscribe to participate in polls.

Reply

or to participate.