Koios AI

Proudly resharing PsySafe AI, an open-source initiative by our founder focused on psychological safety in AI. At Koios AI, we're committed to personalised AI that truly supports users! 🚀

What does AI get wrong about human communication? Most AI today assumes a one-size-fits-all approach to communication. The dominant fine-tuning method, reinforcement learning with human feedback (RLHF), optimises for the most common user preferences. 👉 But there's a problem. People don’t all communicate the same way. In reality, communication isn't one-size-fits-all. Some of us love deep, analytical discussions. Others prefer quick, high-level takeaways. Some enjoy a bit of humour; others want straight-to-the-point answers. 🦄 Personality shapes how we engage with the world—so why doesn't AI adapt to it? It has to do with how LLMs are trained. So far, the assumption of preference uniformity has been a reasonable one for the things labs have optimised for—e.g. factuality, coherence, and safety. When AI enters education, coaching, and mental health, where communication styles vary widely, the uniformity assumption isn’t just limiting—it’s actively harmful. AI that prioritises "the most common" communication style risks excluding diverse voices and making interactions feel unnatural or even frustrating. 📊 Studies show that when messages are tailored to personality, engagement and effectiveness skyrocket. 🗣️ AI that gets you could transform everything from tutoring to mental health support. 💡 But without reliable personality detection, personalisation risks being superficial—or even misleading. The best minds at OpenAI, Anthropic, and Google DeepMind are starting to realise this. I recently saw a Research Scientist position focused on Personality & Model Behavior—a sign that this shift is coming. The future of AI isn't just about making models more powerful—it's about making them more human-aware ♥️ This is something I am deeply interested in and actively working on at Koios AI. If we want AI to feel more natural and useful, we need to move beyond the uniformity assumption and train models to recognise and adapt to diverse communication styles. ✨ The best AI isn’t just smart—it’s personal. #AI #personality #psychology #business #AIAgents #LLM

Can AI truly understand you? 🧠 🔬 In my latest research, I explore how LLMs adjust their responses based on user personality—both when traits are explicitly provided and when inferred from conversation. The results? AI can adapt, but it's far from perfect. The challenge isn't just tweaking responses—it's accurately detecting personality from limited input and ensuring personalisation is both reliable and meaningful. Personalisation can have huge effects. Two major studies led by Sandra Matz reinforce this: 📚 Study in Nature Scientific Reports found that AI-generated messages tailored to personality significantly boost engagement and effectiveness. 📊 A large-scale PNAS study tested psychological targeting on 3.5 million individuals and found that matching messages to personality traits increased clicks by 40% and purchases by 50%. 🧐 This isn't just theory—companies, political campaigns, and digital platforms already leverage this effect to drive real-world outcomes. But here's the challenge: 1️⃣ LLMs can adapt, but they need reliable signals. Without accurate personality detection, personalisation risks being shallow or misaligned. 2️⃣ Personality-aware AI could transform communication. Imagine AI tutors, mental health chatbots, or customer service reps that adjust their tone and content to match you. 3️⃣ Personalisation isn't just a gimmick—it drives results. Better user engagement, higher retention, and a tangible lift in ROI are all tied to AI that gets you. The challenge is making it reliable at scale. ✨ The best AI isn’t just smart—it’s personal. And that’s exactly what we’re working on at Koios AI.

We are extremely excited to be partnering with Opinium in 2025! Keep an eye out as there’s a lot of incredible work to come out of this relationship.

We’re excited to announce our new partnership with Arcadia Consulting a global leader in mindset transformation and true architects of change. Arcadia is known for creating practical, people-first strategies that help organisations unlock their full potential. By focusing on mindset—the foundation of individual and team success—they’ve transformed workplaces worldwide. At Koios AI, we’re passionate about innovation that makes a real difference. This collaboration combines Arcadia’s decades of experience in mindset coaching and cultural transformation with our hyper-personalised conversational AI. Together, we’re developing a virtual coaching tool that helps people and teams perform at their best by focusing on mindset, growth, and resilience. This partnership will redefine how organisations drive individual and team performance. Watch this space as we share more about what’s next with Arcadia Consulting!

🤖 Recently, I experimented with "o1 pro", a cutting-edge language model, by applying it to gaps in my own research—an area I know inside out. The result? It failed ❌ 🔄 The model mostly rehashed the prompts I gave it. Even when I pushed for new solutions, it struggled to deliver anything genuinely novel or original. 💡This highlights a crucial point: expertise isn't something you can automate away—at least not yet. LLMs excel at tasks with clear rules and measurable outcomes—think mathematical proofs or coding challenges. But when it comes to creative leaps or reasoning through ambiguity, such as in novel research, they stumble. 🧠 Here's the twist: the most valuable skill in the age of AI might just be philosophy! Framing open-ended problems, anticipating edge cases, and communicating effectively with the LLM are now essential skills for any AI engineer. Interestingly, LLMs can assist with this meta-thinking. They may not do all the heavy lifting, but they can serve as powerful discussion partners to help refine our thought processes. 🚀 Pairing deep domain expertise with a philosophically minded approach to problem-framing unlocks the real magic. This is where innovation happens. So, what's the takeaway? 🔑 The value of experts who truly understand their domain is higher than ever. The better you know your field—and the sharper your ability to ask the right questions—the more you can leverage AI to spark insights and explore new frontiers. #AI #LLMs #innovation #research

As we wrap up 2024 here at Koios, we didn't want to reflect on the past but focus on the future. 2025 is the year Artificial General Intelligence (AGI) transforms from aspiration to reality. And we are proud to play a critical role in this mission. Our focus is clear: bridging the gap between human understanding and AI responsiveness. By integrating emotional intelligence, real-time psychological insights, and hyper-personalisation, we’re helping to ensure that AI can truly connect with people—not just respond to them. We’re eager to collaborate with innovators like OpenAI, Google, and Anthropic to enhance the adaptability and personality-awareness of LLMs. Similarly, we’re excited to work with leaders like ElevenLabs, WaveForms AI, and Hume AI to elevate voice agents with emotional intelligence and lifelike interactions. 2025 is the pivotal year, and through partnerships like these, we’ll help lay the foundation for a future where AGI isn’t just intelligent—it’s deeply human. We really can't wait to start the new year!

On 23rd June 2021, my daughter was born—a date I’ll never forget. Not only because it’s her birthday (she’d kill me if I forgot), but also because it’s the birthday of Alan Turing. I’ve always known Turing as the father of AI, the genius who cracked the Enigma code, and the face proudly displayed on the £50 note. But it wasn’t until I started building Koios AI that I truly realised the depth of his influence on my life. Today, the race to pass the Turing Speech Test is in full swing. Companies such as WaveForms AI, Hume AI and ElevenLabs are perfecting voice AI that sounds indistinguishably human. But here’s the thing: without psychological awareness, even the most realistic voice is just surface-level mimicry. At Koios AI we’re doing something fundamentally different. We’ve cracked the code to validated personality prediction from voice, combining acoustic and linguistic analysis to not just hear words, but understand the person behind them—their personality, behaviour, and emotional state—all in real time. This isn’t just about making AI sound human. It’s about making it feel human, by enabling large language models (LLMs) and voice agents to adapt and connect with real emotional intelligence (EQ). Without this layer of hyperpersonalisation, true human-like communication will remain out of reach. Alan Turing was one of the greatest minds in history, a British great, and at Koios we’re working to live up to his legacy. #TuringTest #VoiceAI #KoiosAI #PsychologicalAI #HumanConnection #Innovation

Using true personalisation through psychological awareness is proven to increase trust, sales and improve relationships. Ask any successful commercial leader and they’ll tell you, their success is largely down to their ability to understand the human they are dealing with and influence them by adjusting their style and delivery of communication to match their needs. That is how Koios can help you increase your top line. The The Telegraph published their article on our technology last night, adding to the growing interest and validation of our work. Read more here; https://lnkd.in/ezxeJ5iS