OpenAI gets new memory feature

Along with: When Politics met AI

Hey there, 

After a week of Google Gemini buzz (which we hinted about in our previous edition), Sam Altman drops details about GPT-5. ChatGPT evolved to get a memory feature and Nvidia launched its Llama-based chatbot, which runs locally. 

Let's check the action this week.

(Note: Do read till the end to be awed by AI)

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 

  • Question to ponder before we meet next!

Table of Contents

OpenAI is trying out a new “memory ”feature that lets ChatGPT remember and use information from old conversations into new ones. Until now, ChatGPT's memory was in silos, starting each chat with no recollection of previous ones - but this is no longer the case. The feature is being rolled out as you read this, so if you don’t have it yet - just wait for some time.

Another notable feature is the temporary chat option. It allows you to have conversations without using the memory feature, ensuring they won't be stored in history or used by OpenAI for model training. (source)

In our newsletters, we've discussed Politics and AI extensively, but this piece of news stands out. Honestly - I didn’t expect the first AI-based election win to come from Pakistan! But it did - Imran Khan, the former Prime Minister of Pakistan, recently posted a prerecorded video on X, declaring his win in the Pakistan Election. However, there's a noticeable discrepancy - his voice doesn't align with his lip movements. This is because it was created using AI.

With 2 of the world's largest nations - India and the U.S. going for elections this year - AI is all set to play a crucial role. (source)

Nvidia's H100 and A100 chips are generalized, all-purpose AI processors used by all major AI builders including the likes of Google, Amazon, Meta, and OpenAI. (source)

However, several big players are now creating their chips tailored to their specific requirements. This approach can lower energy use and might also reduce costs and design time.

To compete in the approximately $30 billion custom chip market, Nvidia has set up a special team focused on making custom AI chips for these companies.

Microsoft has announced that it is redesigning the look of its Copilot AI search and the chat experience of its chatbot, BingChat. They are not just refreshing the design but also adding some fascinating functions for AI image creation and editing. Additionally, they're rolling out Deucalion, a shiny new AI model that's going to power Copilot like never before.

Furthermore, Microsoft broadcasted an ad during the Super Bowl match on Sunday. The ad aimed to show the world that AI is not here to take your job but to help you make your dreams come true. It featured a diverse parade of everyday people listing their thwarted ambitions, with Microsoft's launch of Copilot emphasizing that it is here to help. (source)

Recently, Google has renamed Bard to Gemini and launched Gemini Ultra 1.0, which is said to be the most powerful AI model from Google with its advanced features and creative capabilities.

How is Gemini different from Gemini Ultra 1.0?

Here's the scoop: Gemini has three versions: There’s Nano for mobile devices, Then Pro- perfect for complex tasks like language translation, and then comes Ultra, which is said to be the most powerful, advanced, and premium version. (source)

Nvidia has just unveiled 'Chat with RTX,' a custom GPT Chatbot trained on your data, including documents, videos, and more.

The best part is- It operates offline and is optimized for Windows PCs with RTX acceleration. It offers some amazing game-changing benefits: it's trainable, supports all file extensions; and provides relevant assistance in searching, summarizing your material, and creating new content, all while ensuring your data remains secure and offline. (source)

Meet Frame, the latest genius creation from Brilliant Labs, packed with AI smarts. It’s a pair of glasses that not only serves your basic eyewear needs but also comes with an advanced multimodal AI assistant called Noa. It's like having a super-smart buddy right on your nose, blending AI seamlessly into your daily wear. From summing up texts to translating languages on the fly, Frame is all set to herald a new era of human interaction. (source)

  • Agent 4 is a tool that lets you create custom voice experiences for callers to your business or mobile phone. It allows you to set up a smart agent that can either answer incoming calls or make outgoing calls using your voice. This intelligent agent can interact using the information and content you've provided, and it can also integrate with your existing systems, useful for handling customer inquiries, making appointments, or any other task where direct voice communication is valuable.

  • Elevenlabs, a free AI voice generator tool, transforms text into speech effortlessly. It enables users to instantly create natural-sounding AI voices in any language, making it ideal for video creators, developers, and businesses.

  • B12, a tool designed to help you build a website for your business using AI, offers industry-specific drafts, along with support from their design and copywriting experts, to assist you in building the website.

  • LinkedIn introduces new AI features to improve networking. They've revamped the Network Tab, splitting it into two sections, 'Grow' and 'Catch Up.' aimed at helping users expand their network and stay updated on existing connections.

  • Keyframer, an Apple-developed tool, uses LLM to animate still images based on natural language prompts and generates animation codes from user prompts.

  • Meet Aya, a multilingual LLM who's fluent in 101 languages. It was trained on a massive multilingual dataset and outperformed other open-source models in tasks like understanding, summarizing, and translating different languages. This makes Aya a fantastic resource for multilingual research and paves the way for more nuanced cross-cultural conversations. (source)

  • In the recent episode of 'Leading with Data’, I spoke with Sebastian Raschka, who shared his strong passion for education and his desire to make AI and Deep Learning accessible through his recently published book on AI.

  • Keeping the Valentine’s Day week in mind - here is a crazy application trick of AI to find your love: Click here

As promised at the start of the newsletter - here is a small experiment I did. You see 2 audios below - one recorded by me in person and one generated by an AI trained on my voice. 

 Options:

Which of the 2 is generated and which is recorded?

Login or Subscribe to participate in polls.

On a separate note, I have a feeling that GPT 5 is just around the corner and it would not only raise the bar of what AI can do but also set the momentum for the action in the next 12 months.

What feature would you want to have/see in the GPT 5?

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.