GPT2-chatbot takes the world by storm!

Amazon’s new enterprise assistance

Hey there, 

A mysterious new LLM model called “GPT-2 chatbot surfaced quietly on the LMSYS leaderboard (without any documentation!). Early users were impressed as the model seemed to have better reasoning and math capabilities along with the generation capabilities we are used to.

You could almost predict what would have happened next. Loads of tweets, speculations, and rumors with tweets from Sam Altman adding to the fire! The chatbot went down due to “unexpectedly high traffic” and “capacity limit.”

All of it is under 48 hours! Welcome to the new world order :) 

If you are in India - witness the new world order at DataHack Summit 2024. It is going to be the place to meet leaders and practitioners and to experience Generative AI in person. The super early bird offer will run out soon! Conference dates - 7th to 10th August 2024.

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 

Please note: This is an abbreviated version of our newsletter due to email length restrictions. For the complete experience, visit our website and enjoy the full, unabridged edition.

Table of Contents

A mysterious new LLM model “GPT-2 chatbot” took the internet by storm. Why? According to many experts, its results may be better than GPT-4. Here are some tweets from the early users sharing their experiences.

Surfacing quietly on the LMSYS Chatbot Arena website without official documentation, its impressive performance includes solving an International Math Olympiad problem in one shot!

Who are the creators? Many researchers believe it is from OpenAI itself. There’s a possibility that it is from a lesser-known company or research group.

As of now, the GPT-2 chatbot has been temporarily removed due to “unexpectedly high traffic” and “capacity limit.” - which could potentially be a signal that it was not OpenAI, but who knows! (source)

Amazon has just revealed its new GenAI-powered Assistant - Amazon Q, aimed at revolutionizing software development and business intelligence across industries.

It has 3 variations Q developer for enhancing software development, Q business or data-driven decision making, and Q Apps for codeless AI app creation, cater to developers, businesses, and app builders.

These AI-powered tools aim to boost productivity, analytics, and innovation across enterprises by streamlining coding, leveraging natural language data queries, and democratizing AI app development without coding expertise. (source)

Eight prominent U.S. newspapers sued OpenAI and Microsoft for copyright infringement over the alleged unauthorized use of millions of their copyrighted articles to train AI models like ChatGPT and Copilot.

The lawsuit, filed in the same court as the New York Times' similar case, accuses the AI firms of removing copyright information and claims reputational damage from AI "hallucinations" fabricating false information attributed to the newspapers.

The outcome could reshape how news companies are compensated in the AI era, though many publishers have opted for paid deals with AI companies instead of litigation. (source)

Apple has renewed discussions with OpenAI to potentially integrate OpenAI's technology into upcoming features for the iPhone.

These talks aim to incorporate OpenAI's capabilities into Apple's iOS 18 operating system. While Apple is also negotiating with Google for its Gemini chatbot, the tech giant has not yet decided on its AI partnerships and may choose multiple providers. (source)

New DHS Advisory Board for safe AI integration in national infrastructure

The U.S. Department of Homeland Security (DHS) announced the formation of a high-profile advisory board, including CEOs from leading tech companies like OpenAI, Microsoft, Google, Nvidia, and others, to guide the integration of artificial intelligence (AI) into national critical infrastructure.

This 22-member board will focus on practical solutions to ensure AI's safe deployment in essential sectors such as energy, utilities, transportation, and more, addressing potential AI-related disruptions that could affect national security, public health, or safety. (source)

China-headquartered Stardust Intelligence released Astribot S1, a fully autonomous humanoid robot.

Astribot S1 is known for its exceptional manual skills and human-like agility, capable of performing household tasks such as folding clothes, shaving cucumbers, and even cooking with remarkable speed and precision.

This robot has the capability of handling up to 22 pounds per arm setting a high standard for future humanoid robots. (source)

Tsinghua University and the Chinese AI startup ShengShu Technology announced the launch of "Vidu," a new text-to-video AI model.

This model is comparable to OpenAI's Sora and represents a significant advancement in AI technology, enabling the conversion of text into video content.

Interestingly, the model shares its name with an existing product called VIDU, a service established in 2021 that creates personalized Loom-style videos at scale. (source)

Synthesia has launched a new generation of AI-powered avatars, called Expressive Avatars, which are driven by the EXPRESS-1 model to deliver ultra-realistic performances in video communications.

These avatars can mimic human speech and expressions with remarkable accuracy, adjusting their tone, body language, and facial expressions based on the context of the conversation.

This capability makes them not just digital renders but digital actors, capable of generating unique and nuanced performances that far surpass pre-recorded dynamics offered by other solutions.  (source)

Elon Musk’s xAI seeks $6 billion in funding

Elon Musk's AI startup, xAI, is currently in the process of raising $6 billion at a pre-money valuation of $18 billion, positioning itself as a formidable rival to OpenAI, which Musk co-founded.

This ambitious fundraising effort underscores Musk's strategic use of his diverse business portfolio, which includes Tesla, SpaceX, and other ventures, as a reservoir for crucial data that could potentially enhance xAI's capabilities. (source)

Adobe introduces VideoGigaGAN, a novel generative video super-resolution (VSR) model that can produce high-quality, temporally consistent upsampled videos with fine-grained details.

While existing VSR approaches have shown impressive temporal consistency, they often generate blurrier results compared to image super-resolution methods due to limitations in their generative capability.

VideoGigaGAN addresses this by extending the success of the large-scale image upsampler GigaGAN to the video domain while introducing techniques to significantly improve the temporal consistency of the upsampled videos by 8X. (source)

NVIDIA announced it is acquiring Run.ai, to help customers more efficiently utilize their AI computing resources across complex deployments spanning cloud, edge, and on-premises infrastructure.

It enables enterprises to centrally manage and optimize shared GPU compute infrastructure, easily access resources for complex AI workloads like large language models, efficiently utilize GPU clusters and gain flexibility from the open architecture.

The acquisition will allow NVIDIA to offer Run.ai's capabilities to its HGX, DGX, and DGX Cloud customers (source)

Snowflake, the cloud computing company, has unveiled Arctic LLM, an "enterprise-grade" generative AI model optimized for tasks like generating database code and SQL queries.

Snowflake's flagship, Arctic LLM was trained on public web data for around $2 million using an efficient mixture-of-experts architecture with 480 billion total parameters.

Snowflake claims Arctic LLM outperforms some rival models on coding and language understanding benchmarks.

Arctic LLM will be primarily pushed through Snowflake's own Cortex AI platform as a way to unlock more value for the company's customers through AI-powered data applications and services. (source) 

Tool: Prompt Perfect

Creating high-quality AI prompts can be a challenging task, requiring a deep understanding of prompt engineering and the specific use case. Prompt Perfect is a tool that automatically generates optimized prompts for various AI tasks.

Problem Statement: Suppose you're launching a startup. Use Prompt Perfect to Create amusing product description prompts that inject personality and humor into your brand messaging. 

  1. Sign up or log in

  2. Select AI Task

  3. Provide Input

  4. Utilize AI-assisted prompt generation

  5. Review and refine

  6. Use the prompt

Here is my Output from the problem statement given.

Access the tool here: Try PromptPerfect AI - Experience AI-assisted prompt engineering.

For more information about the tool, check out our guide on using PromptPerfect AI on the Analytics Vidhya blog.

  • In the recent episode of Leading with Data, I had a conversation with Vijay Karunamurthy about his experience working as the field Chief Technology Officer at Scale AI, where he focuses on democratizing the capabilities of Generative AI and Large Language Models by partnering with major model providers, governments, and Fortune 500 companies.

  • Deeplearning.ai launched a new course called Prompt Engineering for Vision Models along with Comet. The course teaches how to prompt Vision models including SAM and talks about different techniques for prompting Vision models.

  • LLMs currently predict one word at a time. A recent paper published an approach to predict several tokens at a time. They were able to get better performance, faster generation, and no increase in cost!  I think this is similar to a child evolving its thinking from thinking in words to thinking in sentences and the benefits look good. It could become the way forward for most LLMS!

Come to think of it - we were at a capability of generating a few coherent lines a couple of years ago. Today, an anonymous bot surprises the world with its performance and logical reasoning.

All of this, when we know “a big release” is around the corner from OpenAI this Summer! What are your thoughts on what OpenAI would bring to the world this Summer?

What are your thoughts on what OpenAI would bring to the world this Summer?

Login or Subscribe to participate in polls.

Would love 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.