Vacation time to recharge batteries is over - 10 key AI developments you might have missed this summer

Time to relax is over. As August draws to a close and the new academic year begins, it's time to summarize significant advancements in AI that occurred during the summer months.

Considering May ’24 was the “The Warmest May on Record Globally”, and also very hot in terms of important AI announcement, I decided to included May as well.

So, what happened in last 4 months? Here's a subjective selection of 10 noteworthy events in the AI industry.

1. IBM Releases Granite Models to open-source and Launches with Red Hat InstructLab

May 6, 2024

IBM open-sourced its Granite models under an Apache 2.0 license, making them available on platforms like Hugging Face. These models, are designed for business applications, came alongside the innovative InstructLab model alignment technique.

[Source: IBM Newsroom]

2. OpenAI Introduces GPT-4o ("o" stands for omni-modal)

May 13, 2024

OpenAI unveiled GPT-4o, a new flagship model enhancing capabilities across text, vision, and audio, all in real time. This faster and more efficient model aims to broaden access to advanced AI tools. Some of the examples presented during launch were simply jaw-dropping. A must see.

[Source: OpenAI]

3. Google Introduces Gemini 1.5 Pro

May 14, 2024

At Google I/O 2024, Google introduced Gemini 1.5 Pro, featuring a remarkable two million token context window. This advancement allows for processing 2 hours of video, 22 hours of audio, or 1.4 million words in a single prompt (sic!). All 7 books of Harry Potter series represent in total 1.1 million words (only).

[Source: Google Blog]

4. Apple Announces Apple Intelligence

June 10, 2024

Apple introduced it;s own version of AI - Apple Intelligence, a privacy-focused personal AI system integrated into iPhone, iPad, and Mac devices. Considering millions of devices that will be able to run a safe and personal "AI in your pocket", it can further accelerate AI adoption. While initially expected to be part of iOS 18 and iPhone 16, it seems now to come only in subsequent v18.1 update in October. More details during Apple Event on September 9th ("It's Glowtime").

[Source: Apple Newsroom]

5. Anthropic Launches Claude 3.5 Sonnet

June 21, 2024

Anthropic released Claude 3.5 Sonnet, setting a new industry standard for AI intelligence. This frontier model outperforms competitors and its top-performing predecessor Claude 3 Opus, while maintaining mid-tier pricing. This is my favorite model for reasoning tasks, although training data cutoff is April 2024, so for example Paris Olympics results are out of scope and for use cases requiring current facts I use Perplexity.

[Source: Anthropic on LinkedIn]

6. EU AI Act Implementation

Published July 12, 2024; Effective August 1, 2024

The EU Artificial Intelligence Act, the world's first standalone AI legislation, was officially published. It adopts a risk-based approach, imposing stricter rules on higher-risk AI applications.

[Source: European Commission]

7. Meta Releases Llama 3.1

July 23, 2024

Meta released Llama 3.1, including 405B, 70B, and 8B models. While referred to as open-source, it uses a custom Open Model License Agreement with some restrictions compared to Apache 2.0. Better term woudl be "open-model" or "open-weights". Major tech companies , including Databricks, Dell, Nvidia, IBM, Snowflake, Scale AI, and more, quickly announced support for running Llama 3.1.

[Source: Meta AI Blog]

8. OpenAI Releases GPT-4o Mini

July 18, 2024

OpenAI launched GPT-4o Mini, a cost-efficient version of GPT-4o. ChatGPT free plan users can experience both models, switching to 4o-mini after exhausting the GPT-4o limit which is 50 messages in five-hours window.

[Source: OpenAI]

9. AI Stocks Experience Significant Downturn

August 2, 2024

AI-related stocks saw a substantial decline, with billions wiped off market caps, and largest AI players like Google, Microsoft, Meta, Amazon losing over 10% of their value between top valuation in July and August 2nd. AI chip manufacturers like Nvidia, AMD, TSMC dropped over 20%, while Intel and ARM almost 40%, raising concerns about a potential AI bubble crash.

[Source: Fortune]

10. Microsoft Releases Phi-3.5 Small Language Models

August 22, 2024

Microsoft unveiled the Phi-3.5 series, including the Phi-3.5-Mini-Instruct, Phi-3.5-MoE-instruct, and Phi-3.5-Vision-Instruct models. These open-source models are provided under MIT license and designed for various tasks such as basic reasoning, complex reasoning, and image/video analysis. Small footprint (3.8B parameters inPhi-3.5-mini) allows to run them on personal devices. In some scenarios Phi-3.5-mini beats Llama 3.1. I wrote some time ago about SLMs punching above their weight

[Source: Microsoft Blog]

The next big event will be Nvidia Q2 earning report tomorrow (August 28th). After the last week of July / first week of August shake-up in AI stock valuation, this will be significant check point for the AI-related industry.

Innovation in the AI field accelerates and we are still on the move from experimenting to full-time inference, and from Chatbots to autonomous AI agents. The industry continues to evolve at a rapid pace, constantly requiring more computing power and more energy to feed the AI beast. Definitely there is more to come.