Adaptive ML

What is GSPO? Here's the simplest, fastest explanation you'll find.

❄️ Slush 2025 wrapped: cold outside, but absolutely warm (if a bit smoky!) inside, from stages to the media lounge and dinner invites! On the Startup Stage, our CEO Julien Launay shared why reinforcement learning is key to unlocking agentic AI for enterprise and powering large-scale deployments that continuously improve, adapt to customer feedback, and deliver real efficiency and productivity gains. Fantastic meetings, unexpected encounters, and of course, soaking up that Finnish energy (and cardamom buns)! See you next year! 🚀

Excited to share that our CEO, Julien Launay, joined Anand Nimkar of Deloitte on the latest episode of #TheRealWorldofAI podcast from Battery Ventures. Together, they unpack what’s really happening as enterprises push past POCs and move GenAI into production. The conversation covers: 🔸 Why many organizations are still stuck in pilot mode 🔸 How reinforcement learning is enabling safer, more reliable enterprise agents 🔸 What it actually takes to integrate AI into legacy systems 🔸 Real-world customer examples in aviation, insurance, and education 🔸 Why specialized SLMs outperform massive LLMs for enterprise workloads If you’re thinking about where enterprise AI is headed next, this episode is a must-listen. 🎧 Listen here: https://lnkd.in/gmD2G766

🚀 Accel’s annual report of the top AI & Cloud companies is out, and we’re very proud to see Adaptive ML featured among them! The report offers key insights into how the global AI and cloud ecosystem is evolving, and which companies are defining the new stack. A huge thank you to everyone contributing to making this possible. 👉 Full report here: https://lnkd.in/efFVFxzN

We’re heading soon to Helsinki for Slush! 🇫🇮✨ Our CEO Julien Launay will take the Startup Stage to share why we believe that Reinforcement Learning is the key to delivering on the promise of agentic AI for enterprise, bringing efficiency and productivity gains to large-scale deployments. We’ll be around all week, and very keen to connect, exchange ideas, and meet fellow builders shaping the next wave of AI innovation. See you at #Slush2025 🚀

Matt Turck's 2025 MAD Landscape just dropped! 🎤 ⬇️ ….with Adaptive ML leading the pack for reinforcement fine-tuning 🎉 We’re honored to be included in Machine Learning, AI & Data (MAD) Landscape 2025 for a second year — as the only AI Developer Platform focused on reinforcement learning. As Matt notes, “𝑹𝒆𝒂𝒔𝒐𝒏𝒊𝒏𝒈 + 𝑹𝑳 𝒊𝒔 𝒕𝒉𝒆 𝒇𝒓𝒐𝒏𝒕𝒊𝒆𝒓.” -  In 2025 RL’d models from o1 to Deepseek gave us general agents that can use common tools. However, to unlock meaningful business value we still need to develop enterprise agents that understand business context and can interact reliably with business systems. Enter Adaptive ML, we provide the tools and ML engineering support needed to define RL environments, generate training data and tune custom enterprise agents that continuously improve, learning from business feedback. 👉 Read the full landscape here: https://lnkd.in/gJGSM3tC

Matt Turck's 2025 MAD Landscape just dropped! 🎤 ⬇️ ….with Adaptive ML leading the pack for reinforcement fine-tuning 🎉 We’re honored to be included in Machine Learning, AI & Data (MAD) Landscape 2025 for a second year — as the only AI Developer Platform focused on reinforcement learning. As Matt notes, “𝑹𝒆𝒂𝒔𝒐𝒏𝒊𝒏𝒈 + 𝑹𝑳 𝒊𝒔 𝒕𝒉𝒆 𝒇𝒓𝒐𝒏𝒕𝒊𝒆𝒓.” -  In 2025 RL’d models from o1 to Deepseek gave us general agents that can use common tools. However, to unlock meaningful business value we still need to develop enterprise agents that understand business context and can interact reliably with business systems. Enter Adaptive ML, we provide the tools and ML engineering support needed to define RL environments, generate training data and tune custom enterprise agents that continuously improve, learning from business feedback. 👉 Read the full landscape here: https://lnkd.in/gJGSM3tC

AI agents aren't magic.🪄🙅♂️ They're language models that learn to predict very specific text. When ChatGPT says 🔍"searching..." it's generating text that looks like: <tool_call>search terms</tool_call> The system detects these tags, pauses the model, runs the actual search, and feeds results back in. That's it. The model is still just predicting text. So why are agents blowing up now? How does this work under the hood? We built a full visual breakdown 👇

We’re excited to be featured in Sifted’s AI 100 supported by N47, highlighting the most promising AI startups for 2025, alongside so many incredible innovators. 💫 A huge thank-you to our team and community for making this possible. 📘 Read the report: https://lnkd.in/gZnEB97

🚀 Upcoming Webinar: Reinforcement Fine-Tuning with SK Telecom Join us on October 28th, (12:30pm ET / 9:30am PT) for a live conversation with Eric Davis (VP of AI, SK Telecom) and Alessandro Cappelli (Co-founder & ML Researcher, Adaptive ML). They’ll walk through how SK Telecom used Adaptive Engine to reinforcement fine-tune small, specialized LLMs for multilingual customer support—delivering low-latency, cost-efficient, and privately deployable GenAI solutions. 👉 Key takeaways: > Why PPO outperforms supervised fine-tuning > When to use adapters vs. full-model fine-tuning > Practical dataset + training setup tips > How to operationalize specialized LLMs in safety-critical workflows Register now 👇