10 Hidden Generative AI Predictions for 2024

The year 2023 marked a significant milestone in the evolution of Generative AI. A multitude of Large Language Models (LLM) achieved performances akin to human intelligence. More crucially, interfaces such as ChatGPT democratized this technology for non-technical individuals. But later in the year, we witnessed a classic innovation chasm: Within two months of its release, ChatGPT amassed an impressive 100 million users, encapsulating nearly all global innovators in an unprecedented adoption rate. Yet, one year on, the user base had only just doubled, unveiling a struggle to penetrate the broader market.

Despite this apparent struggle and obvious trends e.g. video generation, AI-powered healthcare, and regulatory challenges, I have identified 10 less obvious trends that I believe will dominate 2024. Additionally, I present four opportunities that, in my opinion, hold immense long-term potential, but may require more time to fully materialize.

Generative AI Trend #10: AI as default in established software

Currently, engaging with AI services like ChatGPT requires separate sign-ups and a learning curve of new techniques to get good results. This remains a barrier for the majority of people, who haven't recognized AI's potential to enhance daily life.

However, we're seeing an increasing number of tools and devices incorporating AI seamlessly for their users, without separate signups, making it accessible to millions. For example, Microsoft has integrated Copilot across its entire Office suite, and Google has introduced Duet AI for Google Workspace. Adobe's Firefly is deeply embedded in all Adobe products. Recently, Google announced Gemini Nano, a Generative AI that operates directly on phones without needing an internet connection, so that sensitive data never needs to leave the device. I predict that almost all major SaaS providers will incorporate AI features (similar to Canva), and we will see a surge in general-purpose tools like Spark, an AI-powered email client. Providers that don't offer AI integration will risk losing relevance, triggering widespread FOMO (Fear Of Missing Out).

Generative AI Trend #9: Agents

Agents represent a novel layer over AI chat interfaces: They are programs designed to connect multiple conversations for enhanced outcomes. In December, Mixtral 8x7B, an open-source model, demonstrated that routing smaller models could achieve equal or superior performance compared to larger, monolithic models. This approach (mixture of expert architecture) also addresses individual model biases and context length limitations. We can anticipate the emergence of niche models, each specializing in tasks like SEO, research, fact-checking, content writing, translation, graphic design, and editing. These models will be interconnected and driven by an overarching Agent AI, potentially enabling the production of thousands of articles in a single day.

Generative AI Trend #8: Google won't catch up => Blackhat AI techniques

In December Google released its ChatGPT competitor Gemini in three versions: Nano, Pro, and Ultra. Many experts criticized Google for its fake portrayal of real-time multi-modal AI interactions. Moreover, the first benchmarks didn't support Google's claim of superiority, suggesting that Gemini Pro was on par with GPT-3.5 turbo rather than being a significant advancement. Therefore I don't expect that Gemini Ultra will outperform GPT-4 when it is released in the next weeks. To cap it all, there are rumors, that GPT-4.5 is just around the corner and will be released around the same time, restoring the old distance.

This is a fundamental strategic problem for Google: Google's biggest income stream is search engine ads. More people may switch to better AI tools for navigational requests, resulting in market shifts. On the ethical front, I fear the SEO agencies the most: A peculiar species who base their business model on tricking search algorithms. If the search engine market tightens, maybe they will orient themselves towards AI, applying their learned mindset to bias AI intentionally.

Generative AI Trend #7: Content Overkill

The ease of content generation, already evident in 2023, will continue to grow. Multiple large YouTube channels auto-generate thousands of shorts, generating substantial revenue. However, this ease also means that content creation is accessible to a broader range of less skilled (video editing) individuals, potentially leading to a massive influx of content. I predict, that in 2024 the number of released content, including apps will be 5-10x compared to 2022.

This phenomenon raises both technical and ethical questions. Remember the early days of Spotify, Netflix, and Amazon Prime? The overwhelming amount of available content disrupted established media consumption patterns. For most of us, it took a while until we recalibrated our quality requirements while wasting eminent time (and maybe mental health as well).

Also, there will be huge pressure on music and video editors as well as graphic designers. To grab attention and stand out in the ocean of content, the required quality bar will increase significantly, creating a survival risk for mediocre professionals. While there are regulatory efforts to label AI-generated content, I don't believe that this will have an impact: If consumers cannot discern a difference, they are unlikely to care whether AI aided the creation process.

Generative AI Trend #6: Scientific Breakthroughs by AI

Last year's studies highlighted AI's potential in scientific fields. We can expect specialized AI models for purposes like protein folding, revolutionizing the connection of scientific data worldwide. Currently, the speed of scientific progress is limited by the learning rate of individual researchers. AI could remove this barrier, though it raises ethical concerns and necessitates safety guardrails. Undoubtedly, the genie is out of the bottle, much like with atomic technology. A new race between political systems may emerge, hopefully, more akin to the space race than the arms race.

Generative AI Trend #5: Regulatory Tooling

The finalization of the EU AI Act in December brought regulatory considerations to the forefront. The EU's own impact assessment estimates the cost for companies developing AI systems to be around 300k/year. Many experts fear therefore negative impact on product pricing and competitiveness. However, AI could also be part of the solution. Regulations have been an issue for centuries, throwing big hurdles, especially to startups, e.g. in the medical sector. Bootstrapping in highly regulated industries is almost impossible. While most regulations have a good reason behind them, all documentation processes are highly ineffective by design:

Documentation rarely matches reality: First, documenting every change slows deployment time and creates therefore an intrinsic motivation to reduce it to the absolute minimum. Second, all documentation needs to be updated conscientiously early on, otherwise the documentation corrupts with time. Third, qualified personnel to verify the congruence of documentation is missing. As a result, notified bodies often certify the documentation rather than the actual solution, leading to costly bureaucracy that benefits those companies who understand this game, without necessarily achieving the intended security goals.

Generative AI could bridge the gap between documentation and reality, creating self-documenting systems. For instance, AI could generate diagrams from code as needed, ensuring no discrepancy between documentation and actual implementation. New team members could use tools like GitHub Copilot for code explanations, eliminating the need for quickly outdated inline documentation. AI tools like Otter could passively document meeting decisions and other AI tools compare company practices with legal requirements, generating actionable to-do lists. In this way, Generative AI could streamline compliance in regulated markets, enhancing market access.

Generative AI Trend #4: Turbo Models

In May, Nvidia's market capitalization exceeded USD 1 trillion, mirroring its dominance in the AI chip market. I expect an impending chip shortage, driven by numerous AI projects, demanding more efficient AI models.

OpenAI has demonstrated that turbo models can deliver comparable performance at lower costs. Google introduced Gemini Nano, a model that runs directly on mobile devices, creating no server costs. Initially, the focus was on developing more powerful models, but it's becoming clear that the LLM layer will eventually be a commodity. Continuous evaluations show that discerning the quality of outputs from different models is increasingly challenging for human evaluators. There are now numerous models providing adequate quality, and Mixtral has shown that a combination of average models could be more effective in the long term. Additionally, new models like SDXL turbo, allow real-time image editing, facilitating unprecedented collaboration between humans and AI. Ultimately, the model with the lowest resource demands may prevail.

Generative AI Trend #3: Multimodal AI models

ChatGPT's mobile app now supports voice-based interaction. However, it does not inherently understand voice; it converts it to text, as the base model isn't trained on voice input. This conversion process can potentially lead to a loss of nuanced elements like irony.

True multimodal models can process various media types as input and output. Google Gemini's presentation, though perhaps overstated by marketing efforts, illustrated this concept. The practical implications of such models may not be substantial in the immediate future, but it's anticipated that many companies will focus on developing these capabilities this year.

Generative AI Trend #2: Workplace AI

When I grew up, teachers often emphasized that robots would eventually replace all blue-collar jobs. This shift never fully materialized, as the cost of developing and producing robots often outweighs the expense of human labor. However, significant productivity improvements have transformed blue-collar work. For example, in Rüsselsheim, the birthplace of Opel (and myself), the blue-collar workforce has drastically reduced. There were once 60.000 workers there, and now only a small fraction remain, putting a lot of pressure on these families.

AI, while not replacing humans, will enhance productivity, particularly in office roles previously unaffected by automation. Experts predict that by 2030, AI will surpass human capabilities in most office tasks. This transition, affecting even high-skilled professions like law and medicine, will recalibrate the job market. Despite potential social challenges due to the rapid pace of change, there is optimism for an improved quality of life, even relieving skilled labor shortages and demographic shifts.

I don't expect job cuts for the next year, but I expect an integration of AI in most office jobs. Shortly, job descriptions in traditional office roles are likely to emphasize AI proficiency, reflecting this evolving landscape, and pushing more and more people into evaluating AI for their careers.

While the last year mostly focused on B2C AI products, I expect a huge wave of B2B and internal AI tooling this year.

Generative AI Trend #1: FOMO will waste billions

Almost every company will realize this year, that they could improve performance and gain an advantage by leveraging AI technology.

Currently, a pervasive sense of FOMO (fear of missing out) drives companies of all sizes to heavily invest in AI. This often leads to redundant and unnecessary projects with inflated budgets. This misdirected expenditure will likely cause frustration among company leaders, who may mistakenly blame AI technology instead of their own strategic decisions. For instance, it's questionable why a mid-sized industrial firm has the skillset to develop its own AI model. Most companies lack the internal structure to release AI solutions quickly, resulting in obsolete artifacts before their investments hit the market, doomed to failure from the start. My opinion is, that many companies are currently on the wrong side of the make-or-buy decision, especially for low-level tooling. The growing need for prudent consultancy in this field is evident. But everyone with 6 months experience in that field can claim themself a senior. Unfortunately, I don't think this waste can be stopped.

Bonus: Opportunities, but not a trend for 2024 in my opinion, probably will arrive later:

• Ethics and Safety Guardrails: The US election will probably raise new concerns, but it will be too late for immediate measures. Both the US and EU are cautious about overly restrictive measures, wary of losing technological ground to competitors like China. Historically, significant safety measures are often implemented only after major incidents.
• AI in Education: Systemic changes in education are slow. Past generations have been wary of new technologies, from calculators to internet resources like Wikipedia. Similarly, today's apprehensions about AI in education persist, delaying its meaningful integration.
• Market Shakeout: AI Tool repositories, list up to 10,000 different AI Tools, and every month hundreds of new tools arrive. There will be a time when the self-regulatory power of the market will consolidate, but I don't expect that to happen already within the next year. Instead, there will be probably 100+ base models by the end of next year. I don't think, that this is a good idea, especially for open source, because potential brain power is split among many projects.
• Augmented Reality: Many people forget, that something like Second Life existed since 2003, but failed to reach the mass market. Mostly, because integration with reality falls short, and building 3d assets where too expensive. AI technology could mitigate these issues. Apple's upcoming Vision Pro might propel AR into the mainstream, overcoming Meta's struggle with it, but the high cost of AR glasses remains a barrier for now. Nevertheless, numerous AR projects are expected to be started later this year.

Do you agree with these insights? Are there any important trends I've overlooked? Share your thoughts in the comments.