UnlikelyAI builds a neurosymbolic AI approach for reliable and explainable enterprise AI, led by William Tunstall-Pedoe (co-creator of Amazon Alexa)
London
Joined June 2018
- Tweets30
- Following6
- Followers265
- Likes14
Most assume regulated industries are the hardest place to put AI to work. Too many laws, too much governance. It might be the opposite.
The industries most exposed to AI's reliability problem are also the ones with the most to gain from solving it.
They already have to follow the law. So the AI they use has to follow rules too. That's hard, because most AI systems today are probabilistic and unpredictable.
Watch @williamtp 's take on why.
The industry is bolting reliability layers (guardrails, evals, human review) onto LLMs to shave the error rate down, but that's polish, not a fix. A statistical process is still the one making the call underneath.
For any use case with real consequence, the reasoning needs to run deterministically, and the model only gets a say where its probabilistic nature is a strength, not a risk.
@williamtp says it better in the video.
For regulated, high-stakes AI use cases, teams choose between symbolic approaches that are accurate but slow and expensive, or LLMs that are capable and fast but hard to trust.
STRIDE, which stands for semantic translation into deterministic execution, removes that trade-off. It turns complex policy into decisions that are fast, explainable, cost-efficient, editable by policy experts, and auditable, running inside live systems while keeping LLMs' capability.
UnlikelyAI's founder and CEO, @williamtp, dives deeper on STRIDE below.
How do you define what makes an AI decision high-stakes? Money at risk, regulatory exposure, or reputational damage that takes years to undo... you name it, but it's really case by case.
But one thing that doesn't change: the current architecture of AI runs into a real problem in high-stakes environments - LLMs operate with an intrinsic error rate, and they can't reliably distinguish between what they know and what they've made up.
Listen to @williamtp, @UnlikelyAI's founder and CEO.
@williamtp has set out the full thinking in a two-part essay series: The first one on why AI has hit this wall, and the second one on the architectural approach he believes moves the field past it.
The Trust Ceiling: williamtp.substack.com/p/the… Through AI's Trust Ceiling: williamtp.substack.com/p/bre…, CEO and founder of@UnlikelyAI
Fluency turned out to be the easy part in the AI race, but trust is the hard one. It's not about making today's models bigger, it's about fixing the architecture underneath them.
@williamtp, CEO and founder of @UnlikelyAI, has set out the full thinking in a two-part essay series: The first one on why AI has hit this wall, and the second one on the architectural approach he believes moves the field past it.
The Trust Ceiling: williamtp.substack.com/p/the…
Breaking Through AI's Trust Ceiling: williamtp.substack.com/p/bre…
LLMs are black boxes. While research in explainable AI is advancing and new regulations demand transparency, a perfect explanation for every complex model decision is still far from reach.
Join the webinar here to explore a practical framework : events.zoom.us/ev/Alugqb-wNC…
Building something truly great? Our founder William, the mind behind Amazon's Alexa, shares his top lessons on talent, vision, and why you should build for scale from day one!
Listen to his insights 👇
#AI #Startup #Founder #Scaling #UnlikelyAI #Alexa #TechCEO #BusinessGrowth
LLMs amaze us daily. But in finance, insurance or accounting…is that enough?
Here, Alex explains how our neurosymbolic AI delivers precision you can trust!
#TechInnovation #Startup #AI #AIExplained #Automation #MachineLearning #artificialintelligence
Would you trust an AI to handle your claims right now? As Ailish explains, today’s models simply cannot deliver the precision and confidence insurers need to avoid costly mistakes. That is the exact problem we're addressing and solving here at UnlikelyAI!
#InsurTech #AITrust #AI
How did UnlikelyAI come to be? For our CEO & Founder William, it came from a deep first-hand understanding that AI still has a long way to go to become valuable, reliable, and trustworthy.
Here he shares what led to our creation!
#ExplainableAI #FutureOfAI #AIInnovation
Bringing AI into accounting is a challenge.
Most AI tools don’t provide sufficient rigour or explainability to be truly utilised.
That’s why we’re bringing trustworthy AI tools to accountants to help them do more in less time, with complete audit traceability.
The AI you are using lies about how it thinks.
Traditional AI fabricates its reasoning and remains a black box. Explainable AI provides a full audit trail of it's answers. In regulated industries explainability is everything.
The AI headlines you're reading are mostly hype. Before you believe the headlines as yourself:
Is this actually innovative?
How does it impact YOUR life?
Think critically, not emotionally 🧠
Headlines about AI progress make it seem more revolutionary than it actually is.
Before you believe the next headline about AI development think critically, not emotionally 💭
UnlikelyAI put the top AI reasoning models to the test.
The results were shocking, and they should concern every enterprise leader.