Claude 3 Opus beats GPT 4

Along with: Is Mind-Controlled Gaming the Future?

Hey there, 

LLM competition is getting fierce! Claude 3 Opus recently surpassed GPT 4 on the LMSYS Chatbot Arena. However, OpenAI's rumored GPT-4 successor, launching this summer, could reverse the game once again and take the lead. 

How long would OpenAI take the lead in the AI race? With that thought, let’s dive in!

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!

Please note: This may be 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

Anthropic's Claude 3 Opus defeated GPT-4 in the human-judged LMSYS Chatbot Arena.

Unlike other AI benchmarks, the LMSYS Chatbot Arena has people directly rank the output of two AI models in a blind comparison.

OpenAI has held the #1 rank for a very long time only to become second to Claude 3 Opus. But with the advent of GPT5, will it be able to sustain its rank? (source)

Apple has confirmed its annual Worldwide Developers Conference (WWDC) for June 10-14, 2024 and this year could be big for Generative AI.

Recent Google collaborations and AI startup acquisitions suggest Apple may integrate powerful GenAI features into iOS 18, macOS, iPad OS, and VisionOS

Even Samsung introduced some of the powerful features in its Galaxy launch event, let’s see what Apple has in store for us. (source)

Musk announced that Grok will now be available to all X Premium subscribers, expanding access beyond the previous Premium+ restriction. X Premium costs $8/month (web) or $84/year. 

Grok's key advantage is its ability to access real-time X data, a feature unavailable to competitors. (source)

Emad Mostaque is stepping down from his role at Stability AI to pursue “Decentralised AI” and “to pursue decentralized AI.”

Despite raising $101 million in late 2022 and the viral success of Stable Diffusion, Stability AI struggled with shrinking cash reserves, high computing costs, and a copyright lawsuit from Getty Images.

Mostaque's departure marks a pivotal moment, with Shan Shan Wong (COO), and Christian Laforte (CTO) interim co-CEOs appointed to navigate the company through its challenges. (source)

Nvidia's market surge threatened by new tech alliance

As Nvidia garners dominance in the AI hardware market with its GPUs and CUDA software, essential for AI applications, it has outpaced tech giants like Amazon and Google in the stock market, driven by partnerships with major firms such as Microsoft and OpenAI.

This success prompted the formation of a coalition including Qualcomm, Google, and Intel, focused on developing Intel's OneAPI to diversify AI development across different accelerator chips and reduce dependency on Nvidia's ecosystem, offering greater hardware flexibility and choice in AI advancements. (source)

OpenAI is set to engage with studios, talent agencies, and media executives in Los Angeles to discuss partnerships and introduce Sora, its upcoming text-to-video generator, to more filmmakers.

Sora, expected to publicly launch later this year, has already granted early access to select A-list directors and actors, showcasing a strategic effort to embed this technology within the mainstream media and entertainment industries. (source)

Enter SceneScript by Meta’s Reality Labs Research, augmented reality smart glasses designed to enable real-time understanding and interaction with the physical world.

These glasses aim to empower visually impaired individuals with step-by-step navigation and offer sophisticated digital assistants that understand spatial queries.

SceneScript stands out by using end-to-end machine learning to deduce room geometry from visual data, forgoing the necessity of hard-coded rules. (source)

According to reports by Business Insider, OpenAI is preparing to launch GPT-5, the latest version of its AI model, with a potential release in the summer.

As per the report, GPT-5 is currently in the training phase, with plans for extensive safety testing, including an internal "red teaming" process to identify and address potential issues before its release.  (source)

Elon Musk's Neuralink made a significant advancement by live streaming a paralyzed patient, Noland Arbaugh, playing online chess through a brain chip implant.

After becoming paralyzed from a diving accident, Arbaugh received Neuralink's implant in January, which enabled him to control a computer mouse using his thoughts. 

While experts did not consider the demonstration a breakthrough, it represents a significant improvement in computer interaction for the patient. (source)

Sakana AI, a Tokyo-based AI startup founded by former Google researchers David Ha and Llion Jones, introduces a pioneering approach to AI development with its Evolutionary Model Merge.

This novel method automates the creation of specialized foundation models by efficiently amalgamating diverse open-source models, yielding advanced capabilities in the Japanese language, vision language, and image generation.

These advancements, achieved without conventional training techniques, mark a potential paradigm shift towards more innovative and efficient AI model development. (source)

Code QL by Github is an AI tool that can automatically fix code vulnerabilities, fix over two-thirds of the vulnerabilities it finds, and cover more than 90% of alert types in the languages it supports. 

This means developers will spend less time fixing bugs and security teams will have fewer vulnerabilities to deal with. (source)

The H1 humanoid robot by Unitree recently set a new speed record for humanoid robots at 3.3 meters per second (about 7.4 mph), with potential speeds up to 11 mph, still behind Usain Bolt's peak of 27.33 mph.

Equipped with 3D LiDAR for advanced navigation, the H1 signifies a leap toward emulating human motion, though it does not yet match human speed. (source)

Tool - Suno.ai

Music shapes our world. It relaxes us, energizes workouts, and fuels the best parties. But for those with truly unique tastes, finding the perfect music can be tough. What if you could design your soundtrack?

Problem Statement - Let's create a simple tune for your newsletter reading sessions.

Solution:

  1. Visit Suno.ai.

  2. Sign up for the platform.

  3. Provide a brief description of your desired song.

  4. Choose between instrumental or non-instrumental.

  5. Here are the results:

For more information about the tool, check out this blog

How many AI songs do you think are going to get published on Spotify, YouTube music, and other such apps?

  • In the recent episode of Leading with Data, I had a chat with Dr. Anand Rao about his journey, focusing on AI evolution, finding joy in his pursuits, and sharing career wisdom.

  • This new JavaScript course, by Andrew Y. Ng, and taught by Laurie Voss, focuses on building a full-stack web application using RAG in JavaScript, with a special emphasis on RAG Web Apps with LlamaIndex.

  • This article by Analytics Vidhya discusses MIT's development of a technique called "distribution matching distillation" (DMD), which accelerates AI image generation by up to 30 times, making tools like DALL-E 3 and Stable Diffusion significantly faster.

  • This article by Analytics Vidhya discusses the details of NVIDIA's launch of five free Generative AI courses, aimed at revolutionizing AI education with beginner-friendly content on topics like RAG applications and neural networks.

Over the weekend I watched Netflix’s “3 Body Problem”. While the storytelling is fascinating and captivating - some parts throw questions about the future we are headed towards.

The biggest of them was how far are we from creating a headset like the one in 3 body problem?

We are seeing better LLMs, computing, storage, brain interfaces, and better AR Spatial Computing devices on one hand. We are also seeing more efficient batteries and displays - makes me think how far are we?

San-Ti (in the series) also said we will be ahead of this technology in less than 400 years. What do you think?

When do you think we will be wearing VR headsets like the one in 3 body problem?

Login or Subscribe to participate in polls.

Back-of-the-envelope calculations say that if we apply Moore’s law to each of these areas - compute, storage, AI (we are seeing faster progress), brain interface - this should just be achievable around the 30-year mark.

Would love to hear your thoughts!

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.