Cutting Through the AI Hype: What’s Here to Stay

Words by: Team of authors, Photos by: Midjourney, Creative Dock Group Archive

At the beginning of this year, at the Mobile World Congress in February, we witnessed over 100,000 “AI experts” and the corresponding accompanying euphoria. Comparing that with the “transformation business at the grassroots level” over the past months, one thing is clear: the artificial intelligence landscape is in a state of highly hyped expectations. A year ago, AI was still a marginal topic in the same place, but in 2024, you could hardly find an organisation that did not deal with AI. This is a clear sign of a “bubble” in which some of the inflated expectations are bound to be disappointed. Let's examine the substance of artificial intelligence's short- and medium-term impact on business practice.

However, the developments also show that AI is a cross-cutting technology that is blurring industry boundaries. From Xiaomi’s software-defined e-vehicle to Huawei’s own multi-domain AI model, it is clear that AI-driven transformation represents a new stage of development for business models in which the share of data-driven business accounts for a significant - perhaps even decisive - proportion of value creation. The potential applications of AI are diverse, from production to customer service, from internal training to direct-to-consumer marketing.

Companies can save up to 80% thanks to AI already

One obvious trend is the strong investment activity in regions such as the Arab world, China and the US West Coast, while Europe is taking a wait-and-see approach. The question of whether we can still expect a valley of disappointment regarding AI or have perhaps already passed through it depends on one’s perspective. While AI companies, especially in Silicon Valley, are experiencing strong value growth, user companies that have high expectations for AI are primarily able to realise cost savings for the time being, but these savings can be significant, up to 80%, depending on the domain.

Can the use of AI face any challenges? Definitely.

Where the accuracy of AI technologies is not crucial, but where almost good results are sufficient, AI technologies such as Generative AI already work well for researching and generating design documents and plans that are then revised by humans. However, there are still challenges in areas that require precise and excellent results, particularly in the regulated area of banking and insurance, for example. The characteristic of “hallucinating” language models is one of the practical problems that need to be solved pragmatically. CreativeDock’s projects succeed by combining discrete search indices and language models (LLMs).

Pilot projects in the German administration with approaches such as “LLMoin”, an integrated text assistant for internal administrative use based on the Luminous language model developed and hosted in Germany by Aleph Alpha, show that the need for AI is great but that the path to strong automation could be longer than expected. The training of language models is not only expensive but also requires high-quality data, pre-processing of this data and overcoming copyright hurdles.

The euphoria will subside, but transformation through AI “will not go away.” Currently, companies are often in “persevering” mode. However, they need to consistently work on implementation instead. A change of perspective from “fear” to “confidence” is mandatory.

Prof. Dr. Heinrich Arnold , German lead partner at Creative Dock Group

The complexity of using AI increases with the difficulty of tasks

Why is the development of language models such as Mistral AI large so extremely capital-intensive at US$ 500 million to date? Are there challenges to procuring high-quality data through putting together the necessary IT infrastructure and investing in computing power, high-quality IT infrastructure, electricity, and time?

The main cost drivers of AI development are:

1. Computing power: The computing power required is huge, with a demand equivalent to thousands of NVIDIA H100 series GPUs. For example, a model released by Databricks in March 2024 required over 3,000 of these GPUs, with the price per processor being well over 40,000 dollars.
2. The IT infrastructure: supporting such computing power requires high-end IT infrastructures that are only available in the most significant cloud data centres.
3. Electricity: the energy required to train such models is enormous.
4. Time: Several months of training are required to develop the basic model.

The requirements for the training data are diverse and complex:

• High-quality data: publicly available datasets are usually only sufficient for average performance.
• Data processing: labour-intensive pre-processing of data for training.
• Copyright considerations.
• Timeliness of data to achieve relevant and timely results.
• Synthetic data: Generating synthetic data using APIs of powerful models such as GPT-4 is expensive, slow and may violate terms of use.

Advanced use of AI requires top experts. And that won't change anytime soon

There is no standard recipe for effectively blending all of these ingredients together. Only a handful of top engineers at leading vendors have the necessary knowledge and experience to develop such models quickly. Therefore, companies such as Mistral or X have brought their models to market in record time by recruiting employees who were previously involved in developing models for OpenAI and Google.

The best way to maximise the use of AI right now is by utilising platform agnostic AI tools designed by top AI experts for everyday users with limited to no technological expertise.?

The team of authors:

• Prof. Dr. Heinrich Arnold , investor, multi-supervisory board member and German lead partner at Creative Dock Group
• Dr. Ulrich Arnold , CEO of gkv informatik
• Tomas Kovarik , Head of AI R&D, Creative Dock Group

Keep up to date with the latest news in AI optimisation & corporate venture building and subscribe to our monthly newsletter . Do you want to collaborate with us to advocate for the broader use of AI in your company or to increase your business value through new revenue from your existing assets? We are looking forward to hearing from you!