AI Weekly Digest #33: Big Tech War Towards AGI

Hello, tech enthusiasts! This is Wassim Jouini and Welcome to my AI newsletter, where I bring you the latest advancements in Artificial Intelligence without the unnecessary hype.

You can find me on LinkedIn, Twitter and Medium! Let’s connect!

Now let's dive into this week's news and explore the practical applications of AI across various sectors.

Main Headlines

Here are the three main trends to keep in mind if you are working in AI today.

#1 Meta Invests $18B Worth of GPU to Unlock AGI, and Announces LLaMA 3, Open Source, model rivaling GPT-4

In a significant move towards advancing artificial intelligence, Meta CEO Mark Zuckerberg has unveiled ambitious plans focusing on extensive infrastructure investment, the development of the advanced LLaMA 3 model, and a commitment to open-source AI, signaling a major shift in the race towards Artificial General Intelligence.

Infrastructure Investment: Zuckerberg has announced a significant investment in infrastructure to support the development of AGI. This includes the acquisition of over 340,000 Nvidia H100 GPUs by the end of the year, positioning Meta as potentially having one of the largest GPU capacities of any individual company. This substantial infrastructure, which will grow to nearly 600,000 GPUs by the end of 2024, signifies Meta's commitment to advancing in the field of AI and underscores the importance of computing power in training and running large AI models. Translation? $18B worth of GPUs to accelerate AGI R&D and adoption!

Development of LLaMA 3: Meta is actively working on LLaMA 3, the latest iteration of their large language model. Zuckerberg highlights that unlike LLaMA 2, which was a leading open-source model but not an industry leader, LLaMA 3 aims to be at the forefront of AI development (rivaling GPT-4). It will include code-generating capabilities and focus on advanced reasoning and planning abilities. This indicates Meta's ambition to not only compete in the AI space but to lead, especially in terms of the capabilities and sophistication of their AI models.

Commitment to Open Source: Zuckerberg emphasizes a vision of open-source development for AI models. This approach is seen as a means to democratize access to AI technology, potentially preventing the concentration of power and ensuring broader access to valuable AI advancements. While he acknowledges the challenges and the dynamic nature of the AI field, including safety and responsible deployment, Zuckerberg's inclination towards open-source aligns with a broader vision of shared progress in the development of AGI. This stance contrasts with some competitors who have adopted a more closed approach in their AI development strategies.

#2 Microsoft Adds LLaMA-2 on Azure

This integration offers an alternative to OpenAI models on Azure Platform. Although I do not believe it’s worth investing your time (or money) on LLaMA-2, this integration is worth mentioning for many reasons:

• OpenAI remains (ridiculously) cheap: at a cost of $0.001 per 1,000 tokens, GPT-3.5-turbo offers a more cost-effective solution compared to LLaMA2, which is priced at $0.00154 per 1,000 tokens, despite GPT3.5’s more advanced capabilities.

• Microsoft's diversification strategy: Microsoft's partnership with Meta and its integration of LLaMA2 provide a new open-source alternative to its clients, thereby reducing Microsoft's dependency on OpenAI.

This comes at a moment when there is rising concern in the community about OpenAI's strict safety features, which are causing problems with product compatibility and leading to substantial disruptions for businesses. LLaMA2 can be seen as a safer bet to avoid future disruptions (provided you fine-tune it and accept to pay a higher API cost).

#3 OpenAI has Expanded its Engagement in Sectors related to Politics and Defense.

This week, OpenAI has expanded its engagement in sectors related to politics and defense.

• OpenAI/Pentagon: OpenAI has revised its policy framework to establish a cooperative initiative with the Pentagon. It is reported that this venture will focus predominantly on cybersecurity measures.

• 2024 Elections: OpenAI has initiated a new position titled "Election Program Manager" for the year 2024. This move seems to be in preparation for the anticipated integration of AI technologies in the forthcoming electoral cycle.

#4 DeepMind Did It Again! AlphaGeometry: An Olympiad-level AI system for Geometry

In a fascinating new paper, DeepMind builds an open source neuro-symbolic system to achieve gold medal level Geometry reasoning AI! And that’s NOT even the best part!

The system combines (1) Large Language Model and (2) a Symbolic Deduction Engine! Both work together to build an intuition (LLM) and then work on a full fledge proof (Symbolic Deduction Engine). It’s based on 100% synthetic data (no human demonstrations during the training)

Read my blog post about this new breakthrough!

#5 Realistic AI Generated Images Reach a New Level !

Last October, I wrote a piece entitled “Best Image Generation Model? Dalle3, Midjourney, Stable Diffusion or Adobe Firefly“. The blog post is still relevant ; Yet, since its publication, the release of Magnific's upscaling feature has made realistic portraits pop in an unprecedented way.

Midjourney, remains the best when it comes to realistic AI image generation but it still lacked “something”. AI upscaling enhances photos by restoring realistic details, such as the clarity of eyes, natural skin tones and pores, and intricate hair details, effectively bringing these images to life!

This is it for Today!

Until next time, this is Wassim Jouini, signing off. See you in the next edition!

Have a great Sunday and may AI always be on your side!