UnlikelyAI

When AI headlines claim "reasoning has got real" or models have "absolute capability," it's like reducing all human activity to "becoming more intelligent." Technically true, but it's a trivialisation of human life. State of AI Report 2025 offers a comprehensive look at this past year's AI developments but what's beneath the surface? There is a recurring pattern that comes up in AI reports: the grand narrative of steady, optimistic progress. However, building human-like intelligence can be unpredictable. Engineers tend to be over-confident and deployment tends to underwhelm with timelines stretching into the indefinite future. Human intelligence itself is not structured because evolution didn’t create a neatly contained reasoning system in our minds. It created a complex web of mutually dependent abilities that all have to pull together. What does this mean? It means that reasoning is not one thing. The thinking our team goes through when coding AI is fundamentally different from the thinking they do when planning a product. So when we ask for AI that 'reasons,' we're really asking for many different capabilities, not one. If we want a better guide to technical development and investment strategy, we have to keep our feet on the ground and recognise that AI tools need to be tailored to different use cases in accordance with different needs. Keep an eye on our newsletter, for more grounded perspectives on AI development: https://lnkd.in/eHkGsY8p https://lnkd.in/egdRCSuy

Lloyds Banking Group is exploring neurosymbolic AI in their Innovation Sandbox with UnlikelyAI. Check out QA Financial latest announcement which highlights how our explainable, transparent technology is supporting responsible innovation in financial services. We're proud to be partnering with one of the UK’s most trusted banks on the future of compliant AI. Read the full article: https://lnkd.in/d5TPAG8V

By using an LLM for accounting automation you may be submitting work that isn't compliant. That's negligent at best and in some cases, potentially illegal. When LLMs generate disclosure checklists or filings, legal responsibility stays with the signing accountant, not the software vendor. Regulators assess the output, not the process. Before you automate compliance workflows, ask this: Who owns the output when the algorithm is wrong? Because speed without accountability isn't innovation. It's risk. Watch our Engineering Manager Dillon Montag explain the framework we use to balance automation with accountability.

Looking forward to our founder and CEO William Tunstall-Pedoe FREng joining this incredible conversation at Oxford Generative AI Summit 2025 on the 16-17 October!

Approximately 49% of ChatGPT’s 700M weekly users ask for advice, not action, based on recent Harvard/NBER research: https://lnkd.in/gXEZ5q6F. “Advice” includes seeking guidance and decision support, and “action” means asking the model to generate specific outputs or complete tasks. It is clear that people still trust AI more as an advisor than a doer. When it comes to letting it do the work, especially in regulated environments, hesitation kicks in. So what’s holding back the shift from asking to doing? And more importantly, what would it take for you to trust AI to act? 👉 Tap through the slides to explore #AI #DigitalTransformation #EnterpriseAI #TrustInAI

To achieve real ROI with AI, it’s critical to understand the difference between LLMs and SLMs. Gartner predicts that by 2025, 30% of enterprise generative AI projects will underdeliver due to poor model-fit strategies. So what does this mean? Not every task needs an LLM and by assuming a one-size-fits-all approach, you risk wasting resources and getting poor results. Both models have their strengths and the real ROI starts with knowing when to use which. 👉 Check out the slides to learn more. #AI #LLMs #SLMs #EnterpriseAI #ROI

If you are an accountant, you know the feeling of stress during busy period. AI promises to help, but here's the problem: 43% of accountants say their biggest concern with AI is accuracy, according to Wolters Kluwer, 2025. And when AI gets disclosure checklists wrong? You're back to square one, double-checking every line and creating more work for yourself. AI that forces you to manually verify everything it produces isn't automation, it's just extra steps. Your team needs to choose tools purpose-built for the precision accounting demands. Learn more about the AI tools that you can trust: https://www.unlikely.ai/

We want to give a huge shoutout to our interns - you’re all winners! 🏆 We welcomed this amazing group at the start of summer, and over the past three months, they’ve each made an incredible impact on the team. From left to right, they are Jeffrey Wong, who developed a novel method for tackling ambiguity and hallucination in Large Language Models. Sindi Gjonaj, who helped design a powerful workflow to turn complex insurance policies into structured data, paving the way for smarter automation in claims handling. André Simmen, who’s been one of our smart marketing minds, planned and executed successful marketing campaigns for insurance solution. Daniel Jones, who progressed and evaluated our hybrid NeurosymbolicRAG approach for a more precise and efficient EnterpriseQA system. Maria Kuroedova, another of our strong marketing talents, who ran the campaign for our disclosure product, UnCheck, and helped grow our presence in the accounting space. Nedas Mikučionis, who has made fantastic contributions to a new 0 to 1 product in the UnCheck suite with his incredible software engineering skills. And of course, Jasamrit Rahala, though sadly not in the photo! He got stuck into multiple research projects, running key experiments spanning multiple teams to help us push forward our underlying technology. Thank you all for bringing your ideas, energy, and curiosity to UnlikelyAI. We’re so proud of what you’ve achieved and can’t wait to see what you do next. #Internship #AICareer #Hiring

What's your most trusted source for AI updates today? 🤔 With so many voices from big tech to influencers, the AI space has never been noisier. But where do you turn for information you can genuinely rely on? Vote below and tell us why you trust that source. #AI #Trust

Wondering what it might be like to join us? Our team is open, inclusive and always excited to welcome people who share the same spirit. There's an opportunity to be part of what we’re building. Check out our open roles in the comments! #WeAreHiring #AICareers