Artificial Intelligence (AI) is evolving faster than ever—and with it, the investment landscape is rapidly shifting. From the rise of Large Language Models (LLMs) like ChatGPT to the emergence of intelligent AI agents, the world of AI is moving beyond buzzwords into real-world applications that are transforming industries.
In our recent Deep Dive series on the Future of AI with keynote speaker Mr. Chee-We Ng from Oakseed Ventures, we explored the current state of AI, where it’s headed, and where savvy investors should be looking in 2025 and beyond.
Whether you’re a tech investor, startup founder, or simply curious about AI investment trends, here are the key takeaways taken from the session.
1. Where Is AI Today? From Deep Learning to Reasoning Agents
The past decade of AI innovation can be broken into three major waves:
- 2012–2022: Deep Learning Revolution
Breakthroughs in neural networks, such as AlexNet and BERT, laid the foundation for today’s AI capabilities. The use of transformers led to the preliminary creation of Large Language Models (LLMs) like GPT and DALL-E, both of which are widely used today - 2022–2024: Rise of LLMs
Through the use of clever prompts, Large Language Models of this era like Open AI’s GPT-3.5, Meta’s LLaMA, Anthropic’s Claude, and Google’s Gemini showed that machines could generate human-like responses in answering questions, writing code, translating text and creating images. - 2024 and Beyond: Reasoning Models & AI Agents
The newest generation of AI systems goes beyond text generation—they reason, plan, and collaborate with other AI. Think of them as digital teammates rather than just tools.
“The rate at which AI improves has dramatically sped up… that’s one of the results of the reasoning capabilities developed just in the last year.”
2. The Dark Side of LLMs: Current Challenges
As powerful as LLMs are, they come with vulnerabilities that businesses and investors need to understand. One example is Security Risks, AI systems can be “jailbroken” with cleverly crafted prompts, leading to unauthorized or harmful outputs if organisations are not careful. Through the usage of hidden system prompts, hackers have found ways to manipulate LLMs by exploiting the way language is perceived. For example, hackers could get ChatGPT to teach them how to build a bomb and bypass pre-set safety filters when they asked the LLM to write a ‘letter’ to their grandmother on how to make a bomb.
More importantly, despite the amazing advancement of AI, LLMs sometimes ‘hallucinate’ and give users wrong answers. This comes from the algorithm not understanding specific scenarios and just rehashing pre-existing information commonly found on the web.
Not to mention, LLMs also have a high energy consumption rate which requires substantial electricity and costs to function. This is especially more so when AI and LLMs are operated on bigger scales, thus raising sustainability concerns from younger generations.
3. Where Should You Invest in AI in 2025?
Overall, there are four major verticals investors should look out for when conducting venture capital investing in AI.
The first area to look out for is AI Infrastructure, where investors should look into companies that reduce costs and energy usage for operating AI systems. Another key area to invest in is in companies that specialize in the Vertical Applications of AI, such as AI copilots that provide business intelligence ranging from data visualization to insights that help support decisions.
Thirdly, investors should consider companies that use AI to do cybersecurity better, such as those that use AI to determine what data can certain AI models get access to and which personnel can get access to which of these AI models.
Lastly, investors should also consider companies that specialise in the Cybersecurity of AI by offering products or services that address new cybersecurity challenges arising from the use of AI.
4. The Future of AI: Genius or Yes-Man
Looking forward, there are two competing visions for AI’s future.
One possible direction is that AI will soon evolve to consist of AGI (Artificial General Intelligence), which could help humanity break through new discoveries and accelerate scientific progress. This is the “genius in a data center”—an AI that can think and learn like a human across any domain. While progress is being made, experts believe we’re still years or even decades away.
Another possible future direction of AI is that LLM would become a “Yes-Man” in society. Today’s AI more closely resembles a super-powered assistant—able to memorise answers, generate responses but not capable of critical thought. Unfortunately, the current way that AI models are being trained prepares them to excel in repeating information that’s already established as current knowledge. It answers questions but doesn’t ask them nor challenge conventional wisdom, and might even hallucinate information to support certain points desired by the user.
Understanding this distinction is crucial for anyone making long-term bets on AI, especially investors that are keen to dive into AI tech-based companies.
5. Human Roles in an AI World
“We don’t need an A+ student who answers every question. We need a B student who sees and questions what everyone else missed.” – Thomas Wolf, Co-founder (Hugging Face)
As automation grows and AI impacts more jobs, what’s left for people to do? That would be mainly roles which require experience, industry knowledge, strategic thinking, and pattern recognition—particularly around bottlenecks and business nuance. With AI being able to do all the best practices, humans should thus be prepared to challenge, ask questions and form new approaches that have not been used.
Though the topic of what younger generations should now focus on with AI in play hasn’t been discussed in depth, people should focus on gaining value-adding skills that AI can’t achieve.
Final Thoughts
AI Investment Strategy for 2025 and Beyond
The AI market in 2025 presents enormous opportunities—but also new risks and complexities. As LLMs become embedded in customer service, software tools, and even cybersecurity platforms, it’s critical to understand how these models work, where they fall short, and how to invest responsibly. Some investors believe the next generation of winning AI startups will be energy-efficient, security-conscious, enterprise-ready and vertically specialized.
If you’re interested in investing in AI startups, building responsible AI tools, or staying ahead of the curve, now is the time to act.
👉Join us as a member to access exclusive video recordings of these past workshops and follow us on LinkedIn for more updates on future deep dive sessions with insights from industries leaders.
ABOUT ANGELCENTRAL
AngelCentral is the most active and fastest-growing community of angel investors in Southeast Asia. We organize regular pitch sessions, angel education workshops, and provide syndication services. Since inception, AngelCentral has trained more than 1000 angels and our members have invested more than S$35m into the startups.
If you are a startup and would like to raise funds from our members, send in your application here!
If you are an investor and would like to find out more about Angel Investing and what AngelCentral does, we have a coffee session hosted by the AngelCentral team. These informal chat sessions are held twice monthly where we will share about what we do, our membership offers, and/or what Angel Investing is all about. Secure your slots here.
