AI in 2024

31 Dec 2024 / 17 minutes to read

Elena Daehnhardt

E-mail Twitter GitHub Pinterest LinkedIn

Midjourney 6.1: Fireworks show the number 2025 in the sky

I am still working on this post, which is mostly complete. Thanks for your visit!

Table of Contents

Looking forward to 2025, I could not resist thinking about what happened in AI in 2024. The year 2024 was a very exciting year to observe AI advancements technology-wise, the rise of specialised and multimodal AI models, significant progress in AI creativity, increased focus on responsible AI development, and wider adoption across diverse industries. The AI Act is published, and AI laws will further evolve. I will highlight subjectively the most interesting happenings in AI in 2024!

Key moments

Let’s look back at the key moments of 2024. I will mention arguagbly the most exciting things happening about AI.

Key players

Many businesses, organisations, education institutions and governments shape the landscape of AI in 2024. AI-based startups are blooming, and new companies and technologies emerging daily.

I think that the most well-known gigantic AI companies that have focused their efforts on AI development to date are:

Without their contributions, 2024 will be very boring in AI. Disagree? Write me :)

Generative AI

Generative AI is one of the hottest AI trends. It encompasses various AI models that generate different types of content, from text to images, music, and video content.

If you are interested in knowing how Generative AI differ from Large Language Models (LLMs) - read my post Generative AI vs. Large Language Models

According to McKinsey’s State of AI Report 2024, organisations are rapidly integrating generative AI tools into their workflows, with a significant increase in adoption compared to 2023. Businesses are finding measurable benefits and realising genuine value from these technologies - read in McKinsey’s State of AI Report 2024.

Companies use a mix of off-the-shelf generative AI tools and customised models McKinsey’s State of AI Report 2024. Generative AI is most commonly used in Marketing and Sales, Product and Service Development, and IT Departments. Organisations report cost decreases in human resources due to generative AI use, and about 65% of organizations are regularly using generative AI McKinsey’s State of AI Report 2024

McKinsey’s State of AI Report 2024 informs that inaccuracy and intellectual property infringement are increasingly considered relevant risks while 44% of organisations have experienced at least one negative consequence from generative AI use. Inaccuracy is the most commonly experienced negative consequence McKinsey’s State of AI Report 2024.

Chatbots

In short, advanced chatbots provide us with access to generative AI with the help of a user interface.

Advanced Conversational AI platforms moved beyond simple chatbots to become sophisticated virtual assistants capable of handling complex multi-turn conversations, understanding nuanced language, and integrating with various services.

Chatbots have become more intelligent in their interaction with humans and in their customisation preferences. For instance, Anthropic’s Claude AI introduced customizable writing styles, allowing users to tailor their interactions with different response styles.

Multimodal AI

AI systems are becoming more versatile by integrating different modalities, such as text, images, and potentially audio and video. This paves the way for more natural interactions and richer user experiences.

OpenAI invested heavily into Multimodal AI with the most exciting developments including:

Sora: OpenAI stunned the world with Sora, a text-to-video model capable of generating highly realistic and imaginative video clips from simple text prompts. While not publicly available, it represents a significant leap in generative AI.
GPT-4o: OpenAI launched GPT-4o, a faster, cheaper, and more capable flagship model that excels across text, vision, and audio. It enables real-time, emotionally aware conversations and can “see” and reason about the world through a device’s camera.

In my post Multimodal AI I have explained what Multimodal AI is, its real-life applications and covered the leading techniques and research directions of Multimodal AI
Voice Engine: OpenAI previewed Voice Engine - read in Navigating the Challenges and Opportunities of Synthetic Voices , a text-to-speech model that can clone voices with remarkable accuracy from just a 15-second sample. Its release is being approached cautiously due to potential misuse.

In brief, multimodal AI moved beyond simply combining text and images. Models began integrating other modalities like audio, video, and even sensor data, enabling applications like sophisticated video understanding, real-time language translation with lip-syncing, and more accurate environmental perception for robotics.

Explainable AI

Explainable AI (XAI) refers to a set of techniques and methods that make the decisions and predictions of artificial intelligence models understandable to humans. Instead of AI being a “black box,” XAI aims to provide insights into why a model made a specific prediction or recommendation, increasing transparency and trust.

The need for transparency and explainability grew as AI systems became more complex. Research and development focused on techniques to make AI decision-making more understandable to humans, fostering trust and accountability.

In my post Explainable AI is possible, I have highlighted some key points of AI explainability and how it can be achieved

You can check the Python’s library SHAP in SHapley Additive exPlanations documentation that leverages game theory’s Shapley values to provide a unified framework for feature importance and prediction explanations, making it valuable for debugging models and identifying key factors driving predictions across various domains.

Increased Context-windows

The context window in a large language model (LLM) refers to the amount of text the model can “remember” and consider when processing information or generating a response. Think of it as the model’s “short-term memory” or the “input window” LLM can access.

A larger context window makes the LLM smarter and more capable of handling complex, nuanced, and lengthy text inputs and outputs. Conversely, a smaller context window limits the model’s ability to “remember” and can lead to inconsistencies or a lack of understanding in longer interactions.

Recently, Google significantly expanded the context window of Gemini 1.5 Pro, allowing it to process vast amounts of information (up to 1 million tokens initially, later 2 million in preview). This unlocks new possibilities for analyzing complex documents and codebases.

However, other LLMs also take part in the context-window competition, for instance:

LLM	Developer	Latest Version	Max Context Window (Tokens)
GPT-4o	OpenAI	GPT-4o (May 2024)	128,000
Gemini 1.5 Pro	Google AI	1.5 Pro (May 2024)	2,000,000
Claude 3	Anthropic	Claude 3 Opus (Mar 2024)	200,000
Llama 3	Meta	Llama 3.3	128,000

AI Regulations and Ethics

Governments and international organisations began establishing more concrete regulatory frameworks for AI, focusing on issues like bias, fairness, transparency, AI ethics and accountability. This marked a significant step towards responsible AI development and deployment.

The European Union officially adopted the AI Act, a landmark law that regulates AI systems based on their risk level. The EU’s AI Act marks a significant step towards regulating AI, with a focus on risk management and ethical considerations, read in Shaping Europe’s digital future.

Interested to read about AI act? Read Cláudia Lima's post Regulation on artificial intelligence has already been published.

International collaborations on AI governance will increase, potentially leading to frameworks or agreements to harmonise AI regulations and ethical principles. OECD AI Principles overview is a starting point for this discussion.

Open Source AI

The Open-source models such as LLAMA make AI development more accessible and lead to new innovations. We may see the emergence of more personalized and proactive AI assistants and specialised agents.

Open-source AI models and tools empower businesses to develop customised AI solutions without substantial infrastructure investments. This fosters greater flexibility and allows for tailored applications, according to The most important AI trends in 2024 .

GPT Store

In 2024, OpenAI introduced the GPT Store, enabling users to customize ChatGPT for various applications, including academic access and creative tasks. I liked creating my GPT agents and used other agents for productivity.

AI in Healthcare

AI applications in healthcare have become more robust, particularly in diagnostics and treatment predictions.

AI-powered scientific research and innovations might grow with the help of tools such as AlphaFold Server. AlphaFold Server helps predict how proteins will work with other molecules in cells. You can use the AlphaFold Server for free if you’re doing non-commercial research, so anyone can make predictions, no matter their resources. With AlphaFold it is possible to see how different structures like proteins, DNA, RNA, ligands, ions, and chemicals interact.

AI in Robotics

Integrating advanced language and vision models will enhance robots’ ability to understand instructions, navigate environments, and interact with humans.

AI-Enhanced Creativity and Music

AI tools such as Midjourney evolve daily. AI art is part of our life and is also shown on this blog :) I wish AI-generated art would be produced with well-drawn human hands and good details.

Generation AI in Music is rapidly advancing. AI music platforms like Udio and Suno gained traction, prompting discussions around copyright violations and intellectual property rights when using such AI tools - read in The RIAA versus AI, explained .

Self-driving cars

Thanks to advancements in AI, self-driving cars and other autonomous systems improved their perception and decision-making capabilities. This included better handling of complex real-world scenarios and improved safety.

In 2024, Tesla released an update to its Full Self-Driving system, increasing vehicle autonomy but raising safety concerns in complex environments.

AI hardware

Alongside software advancements, significant improvements were made in AI-specific hardware, including new generations of GPUs, TPUs, and specialized AI chips designed for efficient neural network processing. This hardware progress enabled the training and deployment of even larger and more complex models.

Read about top AI hardware companies and producers of specialized AI chips in Top 20 AI Chip Makers: NVIDIA’s Upcoming Competitors

Future predictions?

I cannot predict the future with 100% accuracy yet :)

However, I suggest the following AI developments in 2025 and later.

Specialised LLM models

At least one major AI company will likely announce new, even more powerful or specialised models in the future. These models and AI platforms will allow us to do scientific research, have a more productive life, and explore limitless creativity without writer’s block :)

The possibilities of practical applications are endless. For instance, specialised LLMs for Legal Document Review can quickly analyse vast amounts of legal documents, identify relevant clauses, potential risks, and other key information, and save lawyers significant time and effort.

Multimodal AI advancements

Multimodal AI could change our television, entertainment, and education sectors. The fusion of different media can enhance our learning and create educational and entertainment environments with speedy access to any information and fantastic immersion when using virtual reality.

Moreeover, Multimodal AI can further enhance medical diagnosis. Imagine a system that analyzes medical images (X-rays, MRIs), patient records (text), and even heart sound audio to provide a more comprehensive and accurate diagnosis.

Jobs safety

There is much worry about job safety and employment when AI takes on the easy task of automating human jobs. The impact of AI on employment needs careful consideration, as well as strategies for workforce adaptation.

While AI is expected to create new job opportunities, concerns remain about its potential displacement of existing roles. The need for reskilling and upskilling initiatives is becoming increasingly critical, as we read in Stanford’s report in Measuring trends in AI.

However, I believe that humanity will adapt to AI advancements and embrace a more fulfilling and productive life on a personal level, as happened during the Industrial Revolution.

Safety and security in AI

The rapid advancement of AI also brings significant challenges. The potential for AI-generated misinformation and deepfakes remains a serious concern. Ensuring that AI systems are fair and unbiased is an ongoing challenge. The security and safety of AI systems will become increasingly important.

Privacy-Preserving AI

Concerns about data privacy led to the increased adoption of federated learning, in which models are trained on decentralized datasets without directly sharing sensitive information. This enabled collaborative model training while preserving user privacy. I think that privacy-preserving AI and technologies such as federated Learning will further evolve to provide us with a safe AI environment.

The further read

Indeed, I could not cover all the exciting moments of AI in 2024 in just one post. You can explore more at the following resources:

1. Moving Past Gen AI’s Honeymoon Phase: Seven Hard Truths for CIOs - discussed the main challenges of deploying Gen AI in organisations;

2. Implementing Generative AI with Speed and Safety - is about effective risk management in Generative AI integrations;

3. McKinsey’s State of AI Report - provides insights into AI adoption, value realization, and emerging trends;

4. Stanford’s AI Index Report-offers a comprehensive overview of AI advancements, including technical progress, economic impact, and ethical considerations;

5. Time’s 100 Most Influential People in AI - highlights the individuals shaping the future of AI;

6. IBM’s Top AI Trends - explores key trends like open-source models, multimodal AI, and AI agents.

Conclusion

2024 was a great year for AI, featuring advancements in LLM specialisation, multimodal capabilities, and a focus on explainability. Open-source models gained popularity, and initial regulatory frameworks started to evolve, addressing AI’s ethical and societal implications.

References

About Elena

Elena, a PhD in Computer Science, simplifies AI concepts and helps you use machine learning.

E-mail Twitter GitHub Pinterest LinkedIn

Thanks I have used these fantastic AI tools to create this post:
Gemini (previously Bard)
Grammarly
ChatGPT
Midjourney

Citation

Elena Daehnhardt. (2024) 'AI in 2024', daehnhardt.com, 31 December 2024. Available at: https://daehnhardt.com/blog/2024/12/31/ai-in-2024/

« Previous: Multimodal AI All Posts