At the end of 2023, we are probably standing at the high crest of the generative AI wave. From being termed the word of the year to being the crux of every major big tech company announcements this year, including Google, Microsoft, and more, it is a no-brainer that AI will continue to be crucial in 2024 too.

Will AI transform the tech ecosystem or be the hype that fizzles out, is something that remains to be seen. However, if we were to make AI predictions for 2024, the optimism from 2023 AI prediction persists. It’s likely that new AI superpowers and use-cases will emerge. Let’s get predicting.

US will not be the only AI superpower

While this year saw the dominance of the US in the AI space with big tech companies such as Meta, Microsoft, OpenAI, and Google releasing their LLMs and chatbots that got the world talking, other countries are slowly (but surely) catching up. 

This year saw the emergence of UAE as a promising force in the AI race. UAE’s Technology Innovation Institute (TII), a research institute supported by the government, released their LLM, Falcon, a 180 billion parameter open-source model this year. The government support combined with abundance of capital comfortably places UAE ahead in the LLM race. 

UAE is also focusing on building its demographic specific-models. Core 42, a subsidiary of tech company G42, released its Arabic-language model, Jais 30B. Furthermore, last week, the Advanced Technology Research Council (ATRC) in Abu Dhabi unveiled a new AI company, named A171.

Launch of A171 in Abu Dhabi. Source: Multiplatform AI

Meanwhile, US’s arch nemesis China too is finding its feet in the LLM race. Last week, Deep Seek, a Chinese company working on AGI, released DeepSeek LLM, a 67 billion parameter model. The open-source model is available in both English and Chinese, and outperforms Llama 2 and Claude-2. 

Though not fully there yet, the European Union too is slowly progressing in the LLM race. Paris-based AI startup that raised $113M in seed round, taking its valuation to $260M in June this year, released Mistral-7B, an open-source model which will be integrated with Vertex AI Notebooks, thereby finding an actual use-case with the tech giant. 

Furthermore, Germany-based AI research and development company Aleph Alpha recently raised $500M in Series B, pushing its valuation to $643M. These investments will probably reap the benefits in 2024, likely resulting in a string of AI investments in EU countries. 

Rallying for open source will gain steam

Founders and AI enthusiasts foresee a future where AI will mostly be democratised. The co-founder and CEO of Hugging Face, Clem Delangue, has predicted a number of things, out of which open-source LLMs is something he has heavily bet on. He believes that open-source LLMs will match the levels of best closed-source LLMs. 

The support for open-source is not just for promoting the whole LLM ecosystem, but from evading the dangers of over-reliance on a single or limited number of closed source models such as GPT-4, Anthropic’s Claude-2 and others.

When Sam Altman was recently ousted from OpenAI, companies that relied on GPT went into a frenzy as the future of the company was questioned, further compelling experts to advocate for open-source models. Meta’s open-source model Llama-2, has been adopted by companies for building a number of LLM models. 

There was also a movement for openness in AI development where 70 experts, including Meta’s chief scientist Yann LeCun, signed the letter. Furthermore, Tesla and x.ai leader Elon Musk has always been a promoter of open source models.

Rise of small language models

With immense costs that runs into millions of dollars, and high GPU utilisation associated with training large language models, big tech companies are looking to work on small language models (SLM). Furthermore, prototyping and customisation for specific tasks will work better on smaller models. 

Microsoft’s love for small language models was unveiled at the recent Ignite event where the company launched Phi2 for enterprises. Microsoft had earlier launched Orca, a 13 billion parameter, considered to be a smaller alternative to GPT-4. Meta’s Llama 7B, Falcon’s 1B, 7B, and Alibaba’s recent model Qwen 1.8B fall under the bucket of SLMs. With the rise of specific use cases in enterprises, SLMs will prove to be beneficial.

Generative AI in arts and science will flourish

The year set the wheel in motion with AI finding applicability across domains with two clear categories being rampantly spoken about — image/video generation and science, especially protein-folding. 

While protein folding applications had been making its way in the last few years, this year witnessed huge developments in that area. Google DeepMind released upgrades to AlphaFold models just a couple of months ago and are continuing its momentum. The models are also finding a way to help nature preserve its habitat. 

It won’t be wrong to say that generative AI has found maximum use cases in video and creative fields. This year saw a number of startups emerge in the space of generative AI text-to-image/video conversion. The recent Pika Labs, a text-to-video platform saw an influx of notable investors before the actual release of the product.

Other platforms such as Midjourney and Runway are  continuously releasing upgraded versions of the models. Indian startups have also emerged, thereby ringing the oncoming of generative AI use-cases in animation and video production. 

Delangue also predicts that there will be ‘big breakthroughs in time-series, biology and chemistry’. 

AGI still remains hazy

A topic so vastly debated in 2023, AGI discussions will continue in 2024 as well. In a race to achieve AGI, big tech companies are still wading their way to understand how to get there. On the one hand, OpenAI is working on Q* and PPO that will supposedly help reach AGI, Yann Le Cun, on the other hand, has not only been dissing OpenAI’s approach but has also stated that AI superintelligence will not happen in the next five years.

He believes we can get to cat- or dog-level AI before reaching the human-levels. 

Though Google Gemini, a powerful AI model, is slated to release next year, the hopes of AGI from it is still a distant dream. 

Going by the whirlwind of a year it has been for generative AI this year, it is unfathomable to exactly predict the diverse nature of which AI will continue to revolutionise the world. However, the generative AI hype is said to fade, and only actual use-cases will thrive.

With limitations in LLMs, especially in the finance sector, most companies integrating ChatGPT and other similar models are for either conversation or to improve their operational efficiencies. Revolutionary use cases are still awaited.