Isophore Available for Trial
Precogrecog makes its first product available. Isophore uses novel algorithms to allow engineers to understand the variability of the manifold around the prompt that they have selected and to establish how robust a particular prompt is. Isophore is under development but you can download a trial/demo version from the products page, or if you contact us directly we can discuss making the full version available for evaluation.
ComitFS Forum Talk
ComitFS is a trader voice company. Simon Thompson spoke at their forum and shared experience with developing applications that meld AI and voice data.
Collaboration about Systemic Risk from AI in the Financial System
Simon Thompson has joined a team led by researchers from the University of Oxford to work on understanding how systemic risks can arise in the financial system due to the use and deployment of modern AI technology.
2/12/2025 Simon Thompson delivered a lecture at the UCL Chemical and Physical Society on modern machine learning. View the lecture slides or check out the references for further reading. The picture on the side bar here is and AI generated fabrication, the actual lecture happened in a normal seminar room, but the audience was patient and came up with a bunch of interesting questions. It was great to also meet with a work excolleague who had decided to come along and see what I had to say. The lecture's aim was to try to communicate the reality of modern ML tech without confusing people but also without eliding too much or telling fibs to get past technical issues. I hope it succeeded.
19/11 Simon Thompson will be on a panel in London Revolutionising investment with Quantum. Should be fun! Spoiler - I think that QC will likely be somewhat useful for chemistry / biology, and maybe it will break RSA, but since the biggest factorisation with Shor's algorithm is still 21 after 15 years it's not looking like this will happen tomorrow, or even by 2035.
Posted a new blog about some of the disappointing stories that have come out about AI research and applications in the last six months. There have been stories about how physics inspired neural networks have not proved as effective as the hype, how AI tutoring has both not lived up to claims and also diverted investment from better educational technologies, and worst of all, what looks like a flat out fraudulant set of claims around the effectiveness of AI in material science research.
Part of being ethical as an AI practioner is being straight about both the good things and the bad things, users need to understand the limitations of AI and what we can really do with it.
11/11 Simon Thompson will be speaking at the Agentic AI Security event at CGI, Fenchurch Street, London. I'm going to talk about trust in agent systems powered by LLMs, looking at how LLMs cope with concepts of trust and scenarios where trust is required for successful collaboration.