Claude continues to evolve!

Along with- Sam Altman wants to use AI for better Healthcare

Hey there, 

Last week - I came across a podcast episode of Joe Rogan hosting Sam Altman on his popular show. Surprisingly, it had less than 50,000 views after 7 days of it going live! When I started hearing the episode, I realized that it was an older episode reposted by them.

That made me think about the world we are heading into! AI is getting used to create and re-circulate content and that is getting fed into future AI models as new content! 

Thankfully, new updates to Claude highlight AI-created content by marking it - hopefully more models do this going forward!

On that note, as we navigate the complexities and advancements of AI, let's shift our focus to an upcoming opportunity to deepen our understanding: the DataHack Summit.

The agenda is now live! Explore the complete schedule online to discover all the sessions, workshops, and activities planned across the four-day event. This is your chance to plan your participation and make the most of this comprehensive GenAI conference.

Now coming back to the 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

Anthropic has recently updated its Claude AI with a new feature that allows users to publish and share 'Artifacts.' 

Artifacts are basically standalone pieces of content, such as code snippets, documents, or interactive components created by AI. This not only promotes greater visibility of AI-generated content by enabling users to make their Artifacts public with shareable links, but it also introduces a remixing capability. 

Thus, users can share, access, and modify Artifacts made by others to foster a community-based approach to developing and refining AI-generated materials. (source)

Quora's Poe introduces 'Previews' for creating interactive apps with AI chatbots

Similar to Anthropic’s Artifacts, Poe’s Previews provide dedicated workspaces for editing and enhancing AI-generated content like code and documents.

This feature allows users to create interactive apps directly within chatbot conversations like data visualizations, games, and tools.

These apps can utilize multiple chatbots, such as Meta’s Llama 3 and GPT-4o, and can integrate information from uploaded files, including videos. These apps can be shared via a link. (source)

Thrive AI Health has been launched to revolutionize personal health with AI coaching

Thrive Global and OpenAI's startup fund have partnered to launch Thrive AI Health, an initiative to develop an AI health coach that offers personalized health advice based on biometrics and lifestyle habits.

This initiative, led by Sam Altman and Arianna Huffington, aims to harness AI to address health challenges, especially chronic conditions. The model leverages advanced AI to identify patterns and suggest behavior changes, aspiring to democratize high-quality health advice and potentially equalize health disparities. (source)

Recently, Elon Musk tweeted about the upcoming release of the Grok-2 AI model, a more advanced version of xAI's Grok series, set to launch in August. This new large language model (LLM) promises enhanced features and improved performance compared to its previous model. 

Musk also highlighted that Grok-2 will reduce its reliance on internet-sourced training data, aiming to refine the quality of its responses and avoid commonalities seen in chatbot language patterns. (source)

Google DeepMind has introduced a new AI training methodology called JEST, short for Joint Example Selection Training, which significantly reduces the environmental impact and cost of training AI models. JEST is 13 times faster and 10 times more energy-efficient compared to traditional methods.

The process-

It involves using a smaller model to evaluate and rank batches of data, which are then used to effectively train a larger model. This batch-focused approach, enhanced by multimodal contrastive learning, allows for quicker and more efficient learning processes by selecting the most learnable data subsets. However, JEST still requires well-curated datasets and faces challenges in optimizing reference distributions. (source)

At the World AI Conference in Shanghai, SenseTime, a leading Chinese AI company, unveiled its latest model, SenseNova 5.5 - a real-time multimodal AI model.

SenseTime claims SenseNova 5.5 rivals GPT-4o, the flagship model of Microsoft-backed OpenAI.

This launch comes at a time when OpenAI is halting its services for users in China, blocking access to its tools and services from July 9. (source)

AI is playing an important role in transforming next-generation advertisements. Recently Volvo launched a commercial that was completely made from AI. This test was performed with Runway ML gen-3 using text to video, though some small details were retouched in AE after the edit was done. The best part was this complete commercial took less than 24 hours to make. (source)

Meta has introduced a multi-token prediction method for training large language models (LLMs). This innovative approach, which forecasts several future words simultaneously rather than sequentially, significantly accelerates the training process and enhances overall performance. 

Initially focusing on code completion, this method promises to lower computational demands and potentially reduce both the costs and environmental impacts associated with AI development. (source)

Kyutai has launched Moshi, a groundbreaking AI assistant capable of real-time conversation, developed in just six months. Unlike traditional models, Moshi uses an innovative "Audio Language Model" that processes audio as pseudo-words, enabling quicker and more natural interactions. 

With a latency of about 200-240 milliseconds, Moshi's efficiency marks a significant advancement in AI communication technologies. Although it's built on a relatively small model, its performance is promising. Kyutai plans to make Moshi open-source, expanding its accessibility to developers and researchers globally. (source)

California robotics company Figure has launched a humanoid robot at a BMW assembly plant in South Carolina, revolutionizing manufacturing with its autonomous capabilities. 

Equipped with neural networks and leveraging OpenAI's technology, this robot can autonomously perform tasks with impressive precision and adapt its actions in real time. 

Valued at over $2.6 billion and backed by tech giants, Figure's humanoid robots are designed to address labor shortages in demanding roles by automating repetitive tasks across industries, from automotive to coffee making, enhancing efficiency and safety in the workplace. (source)

Tool: Outfit.fm

If you're running a clothing brand and struggling with how to present your clothes effectively, Outfit.fm offers a powerful solution. This AI-powered platform allows you to create professional, high-resolution images of your garments without the need for models or expensive photoshoots, simply by uploading a photo of your item. 

It's an affordable, efficient way to enhance your online product presentations and attract more customers.

Problem statement: Use the tool for the DataHack Summit. Suppose the organizing team plans to give personalized T-shirts to each participant, reflecting the diverse tastes and preferences of a varied audience. The challenge is to efficiently display multiple T-shirt designs on the event's registration page without the high costs and logistical challenges of traditional photoshoots involving multiple models and settings.

Solution:

  • Sign up

  • Upload your designs

  • Select a design and click 'Generate'

  • View the outfit on various models

By uploading basic images of the T-shirts, Outfit.fm will use AI to place them on varied models in different poses and settings, thus creating appealing visual content.

  • In this video AI playground, the presenter explores generative AI by discussing its functionality, usage, risks, and limitations, alongside topics like autonomous agents, human roles, prompt engineering, AI in product development, the origins of ChatGPT, various model types, and mindset strategies related to AI and that too in a simpler way.

  • In the recent episode of Leading with Data, I had a chat with Ines Montani about her experience in developing spaCy and Prodigy, discussing the broader impacts and applications of generative AI in various industries, addressing common challenges in the field, and exploring the future directions of this transformative technology.

  • If you are interested in enhancing their skills in managing and deploying AI models effectively. This course "Prompt Compression and Query Optimization" by DeepLearning.AI is designed to teach techniques for optimizing the use of prompts in Large Language Models. 

  • This is a great read into Amazon's practice of conducting a Weekly Business Review (WBR), a cornerstone in its operational strategy.

I have been trying to debate with myself whether we are in an AI bubble. The investments flowing in AI are unparalleled and at this stage the only company which has been able to show some business returns is OpenAI. To put some context by numbers - OpenAI annualized revenue is ~$3.4 Bn and the capital flow in AI is running in Trillions of dollars.

On the other hand, I feel that this is just the start of a revolution we have never seen so benchmarking it against any other technology is unfair. If AI can displace many job families, it will create value and would benefit the companies who are supplying the intelligence. The investments, adaption, and returns would all be at a scale we have not seen before - so we may think we are in a bubble - but we are not!

What do you guys think?

Are we in an AI Bubble?

Login or Subscribe to participate in polls.

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.