Monthly Tech Bites #37 | The Future of AI: Scaling Limits, Adoption Trends, and GenAI’s Role in Education

Hey there, it’s Jerzy – Chief AI Officer at Miquido, and you’re reading your monthly dose of AI and business news! ??

For a couple of years now, the mantra of artificial intelligence has been "scale, scale, scale." More data. Bigger models. Greater compute. But, as Ilya Sutskever (OpenAI cofounder) points out, we’ve hit a limit. Scaling LLMs is no longer yielding the returns it once did. The future of AI isn’t just about making LLMs bigger; it’s about making them better, smarter, and more purposeful.

The cost of training these massive models has skyrocketed into the tens (even hundreds) of millions. As models like GPT-4 and Gemini Ultra push the $200M boundary, the question becomes: Is this sustainable? And, more importantly, where do we go next?

I believe the answer lies in smarter use cases, not brute force.

In today’s newsletter, we’ll explore:

?? AI adoption from three perspectives – insights from Goldman Sachs, Anthropic, and Ethan Mollick

?? Not sure how to harness GenAI? Browse ideas from our curated catalogue!

?? Generative AI in the beauty industry – standout case studies

?? AI in education – a threat, an opportunity, or simply an inevitable future?

?? Are we at a GenAI plateau? Thoughts from OpenAI’s Ilya Sutskever

Let’s dive in!

The reality of AI adoption: Insights from Goldman Sachs, Anthropic, and Ethan Mollick

Despite the relentless buzz around AI, the reality is much less glamorous: adoption is slow, and many businesses are still struggling to make GenAI truly impactful in their operations.

What supports this uncomfortable truth, challenging the rosy “success stories” often found in our LinkedIn echo chamber?

Let’s start with findings from a recent Goldman Sachs report:

• · Only 6.1% of companies use AI to produce goods or services.
• · Just 13% of CFOs report a “very positive” ROI from AI investments – a steep drop from 27% in March.
• · 65% of CFOs cite limited ROI as a major challenge, underscoring the ongoing struggle to realize AI’s full impact.

While some industries, particularly finance and insurance, are beginning to reap benefits, adoption remains stagnant in other sectors, such as education (more on that in today’s news) and manufacturing.

Anthropic’s Clio analysis

Results from Anthropic’s Clio offer further clarity. Clio analyzed over one million user conversations to understand how AI models like Claude are used in practice.

Here’s what they found:

• · Claude's most popular use case is?web and mobile app development, accounting for?10.4%?of analyzed conversations.
• · When including AI/ML and DevOps tasks, programming-related activities comprise 20.3% of all use cases.
• · 31.7%?of use cases focus on text-based tasks, such as content creation, academic writing, career development, marketing, and translations.
• · More complex tasks like strategy, data analysis, or data visualization only account for a small fraction (3.5-5.7%) of Claude.ai’s use cases.

Limited diversity in GenAI use cases

All these findings highlight an important point: current GenAI applications remain one-dimensional, dominated by content creation and programming tasks. As Ethan Mollick notes in his article 15 Times to Use AI, and 5 Not To, broadly available GenAI tools (Claude, ChatGPT, Gemini, etc.) excel “when speed, volume, and expertise are the goals.” However, they fall short “when learning, accuracy, or deep thinking are essential.”

The key phrase here? Broadly available.

Generative AI can drive complex, precision-oriented tasks – even in high-stakes industries like banking or medicine. For instance, we recently hosted a webinar with AIDIFY on AI implementation in the pharmaceutical industry, showcasing real-world success.

However, such breakthroughs require far more advanced approaches than off-the-shelf tools or quick no/low-code solutions. Success hinges on technologies like:

• · Autonomous agent architectures
• · Model fine-tuning techniques
• · RAG (Retrieval-Augmented Generation)

These need to be paired with rigorous security measures and robust evaluation frameworks. Achieving this demands significant investment, effort, and expertise – not shortcuts involving low-quality libraries.

At Miquido, we actively address this challenge with our AI Kickstarter framework.

The AI Solutions Catalog

I frequently talk to people who want to tap into GenAI's potential but don’t know where to start. They often ask: Where can I find inspiration, examples of successful implementations, and business cases?

This is understandable. As a relatively young technology, generative AI is not yet widely or commercially utilized. Finding solid benchmarks in a given industry often means sifting through layers of marketing hype.

To address this, we recently prepared a catalog of ideas for relatively low-cost GenAI implementations.

Interactive knowledge bases, meeting planning tools, product recommendations, product descriptions, and much more – check out our collection of inspirations!

Generative AI in the beauty industry

Speaking of GenAI use cases, the beauty industry is quietly showing just how transformative this technology can be. From virtual makeup try-ons to personalized skincare solutions, brands use GenAI to engage customers, streamline operations, and drive remarkable sales growth.

And the numbers back it up: the AI market in beauty reached $3.2 billion in 2023 and is on track to more than double to $7.8 billion by 2028, growing at a strong 19.6% annually.

Leading brands are setting new standards:

• · Sephora’s Pocket Contour uses AI to recommend the perfect shades based on facial analysis, solving a challenge faced by 70% of makeup buyers.
• · Olay’s Skin Advisor doubled global conversion rates by providing tailored skincare recommendations.

The takeaway? The beauty industry proves that generative AI isn’t just hype – it’s delivering real, measurable impact.

Want to learn more? Check out the full article on our website!

AI in education: A threat, an opportunity, or simply – the inevitable future?

I believe education is a sector where the potential of GenAI is simply immense. We've experienced this firsthand while working with our client, NOLEJ – an innovative EdTech company whose GenAI-powered engine generates high-quality training materials based on verified sources in various formats.

Throughout months of collaboration, I witnessed NOLEJ receiving multiple industry awards: Winner of Best EdTech Startup out of 7,000 global competitors and selection for TechCrunch Battlefield Top 20. These achievements stem from a brilliant business idea: combining GenAI with deep domain expertise.

The numbers speak for themselves

The potential of AI in education is further reinforced by data shared in a recent AI Made Simple article:

• 97% of EdTech leaders view AI as beneficial to education.
• 35% of schools in the U.S. are already implementing generative AI initiatives.
• Key impact areas include increasing productivity (43%) and enabling personalized learning (30%).

What comes next?

Alarmingly, 49% of teachers in the U.S. report being unprepared to integrate AI into classrooms, while 54% of schools lack clear AI usage policies. And this is in the United States, where the pace of GenAI adoption and technology availability – unlike in Europe (e.g., limited access to tools like Sora in the EU) – is significantly higher!

That’s why the work of EdTech companies like NOLEJ is so essential – it bridges the gap between AI’s potential and its practical implementation. By combining generative AI with deep educational expertise, they’re not just creating tools but reshaping how learning happens.

Is the AI scaling era ending?

According to Ilya Sutskever, co-founder of OpenAI, we’re entering a new chapter in AI – one where bigger doesn’t always mean better. Speaking at the NeurIPS conference in Vancouver, Sutskever pointed out that while computational power continues to grow, high-quality data has become a critical bottleneck. Synthetic data, though promising, hasn’t yet matched the impact of real-world data.

The focus is shifting

Sutskever outlined several exciting directions for AI’s evolution. One key area is intelligent agents – AI systems capable of acting autonomously, making decisions, and interacting with the world much like humans, but with the computational strengths of large language models.

Another promising frontier is sequence-to-sequence learning, which allows AI to process complex tasks, workflows, and data with greater accuracy and flexibility.

Perhaps most compelling is the idea of self-aware systems – AI that understands its role, environment, and context. This level of awareness could redefine how humans and AI collaborate, transforming AI from a passive tool into an active problem-solving partner.

Sutskever’s vision sees AI evolving into a collaborative force, not just a static tool. Much like the mobile revolution, the next era of AI will centre around utility, ecosystems, and purpose.

So – while models themselves may stop growing in size, innovation will come from intelligent tools, agents, and applications that harness AI’s full potential.

This truly is an exciting time for AI – and for all of us witnessing its rapid evolution.

Thank you for being part of this journey throughout 2024.

Wishing you a joyful Christmas and a fantastic New Year!

See you next month,

Jerzy