By the end of 2023, we may be standing at the peak of the generative AI wave. From being named the word of the year to becoming the centerpiece of announcements from major tech companies like Google and Microsoft, it is evident that AI will continue to play a crucial role in 2024.

Whether AI will transform the technology ecosystem or become a fading hype remains to be seen. However, if we are to make predictions about AI in 2024, the optimism of 2023 is likely to continue. New AI superpowers and use cases may emerge. Let's start predicting.

The United States won't be the only AI superpower

This year, the dominance of the United States in the field of AI has been particularly evident, with major tech companies like Meta, Microsoft, OpenAI, and Google releasing their large language models (LLMs) and chatbots, garnering global attention. However, other countries are slowly but surely catching up.

This year, the United Arab Emirates (UAE) emerged as a strong force in the AI race. The government-supported research institution, the UAE Technology Innovation Institute (TII), released their LLM Falcon, an open-source model with 180 billion parameters. With government support and ample capital, the UAE is undoubtedly leading in the LLM competition.

The UAE is also focused on building specific population models. Core 42, a subsidiary of G42 Technologies, released their Arabic language model Jais 30B. Additionally, last week, the Advanced Technology Research Council (ATRC) in Abu Dhabi unveiled a new AI company named AI71.

Meanwhile, China, the formidable rival of the United States, is also seeking a foothold in the LLM competition. Last week, Chinese company Deep Seek released DeepSeek LLM, a model with 67 billion parameters. This open-source model offers both English and Chinese versions and outperforms Llama 2 and Claude-2.

Although the European Union (EU) is not yet fully in place, it is gradually making progress in the LLM competition. In June this year, a Paris-based AI startup raised $113 million in seed funding, increasing its valuation to $260 million, and released an open-source model called Mistral-7B, which integrates with Vertex AI notebooks, thus finding real use cases with tech giants.

In addition, German AI research company Aleph Alpha recently raised $500 million in Series B funding, pushing its valuation to $643 million. These investments may yield results in 2024 and likely lead to a series of AI investments in EU countries.

The open-source movement will gain momentum

Founders and AI enthusiasts envision a future where AI is primarily democratized. Clem Delangue, co-founder and CEO of Hugging Face, made some predictions, with a particular focus on open-source LLMs. He believes that open-source LLMs will match the level of the best closed-source LLMs.

Support for open-source is not only aimed at driving the development of the entire LLM ecosystem but also to avoid the risk of over-reliance on a single or limited number of closed-source models, such as GPT-4 and Anthropic's Claude-2.

When Sam Altman was recently ousted from OpenAI, companies relying on GPT were thrown into chaos as the future of the company was questioned, further prompting experts to advocate for open-source models. Meta's open-source model Llama-2 has been adopted by many companies to build various LLM models.

There is also a movement for openness in AI development, with 70 experts, including Meta's Chief Scientist Yann LeCun, signing an open letter. Additionally, leaders like Elon Musk from Tesla and x.ai have long been advocates of open-source models.

The rise of small language models

Due to the high cost of training large language models, which can reach millions of dollars, and the high GPU utilization, major tech companies are exploring small language models (SLMs) for research. Additionally, prototypes and customization for specific tasks tend to work better on smaller models.

Microsoft's love for small language models was revealed at the recent Ignite event, where the company introduced Phi2 for enterprises. Microsoft previously released a smaller alternative to GPT-4 called Orca, with 13 billion parameters. Meta's Llama 7B, Falcon's 1B and 7B, and Alibaba's recent model Qwen 1.8B all fall under the category of SLMs. As specific use cases for enterprises increase, SLMs will prove beneficial.

The flourishing applications of generative AI in art and science

This year, AI applications in various fields have begun to take off, with two categories being widely mentioned: image/video generation and science, particularly protein folding.

While protein folding has been demonstrated in the past few years, there has been significant development in this field this year. Google's DeepMind recently upgraded the AlphaFold model and continues to make strides. These models are also helping in the conservation of natural habitats.

It can be said that generative AI has found the most use cases in the fields of video and creativity. This year, many startups in the text-to-image/video conversion domain of generative AI have emerged. The recent Pika Labs is a platform for text-to-video and has attracted a group of prominent investors even before the product's actual release.

Other platforms like Midjourney and Runway continue to release upgraded versions of their models. Startups from India have also emerged, signaling the imminent arrival of use cases for generative AI in animation and video production.

Delangue also predicts "major breakthroughs" in time series, biology, and chemistry.

AGI remains elusive

AGI has been widely discussed in 2023, and the discussions will continue in 2024. In the race to achieve AGI, major tech companies are still figuring out how to get there. On one hand, OpenAI is researching Q* and PPO, which are said to contribute to achieving AGI. On the other hand, Yann LeCun not only questions OpenAI's approach but also states that superintelligent AI won't appear in the next five years.

He believes that before reaching human-level AI, we can achieve AI at the level of cats or dogs.

While Google's powerful AI model Gemini is expected to be released next year, the hope of achieving AGI remains distant.

Considering the tumultuous journey of generative AI this year, it is impossible to predict with certainty how AI will continue to revolutionize the world in diverse ways. However, the hype around generative AI is said to fade, and only real use cases will thrive.

Especially in the financial sector, where LLMs have limitations, most companies that integrate ChatGPT and similar models do so for conversations or to improve operational efficiency. Revolutionary use cases are still awaited.