The Art of Computer Prompting.
www-cs-faculty.stanford.edu/~knuth/paper...
@mircomusolesi
Professor of Computer Science. Machine Intelligence Lab, UCL AI Centre, Department of Computer Science, University College London. Artificial Intelligence | Machine Learning | Computing | Books https://www.mircomusolesi.org
The Art of Computer Prompting.
www-cs-faculty.stanford.edu/~knuth/paper...
Teaching "Generative Artificial Intelligence" this term. Started today's class with an example celebrating our founder. #UCL200
New paper in Transactions on Machine Learning Research with
Giorgio Franceschelli: "DiffSampling: Enhancing Diversity and Accuracy in Neural Text Generation".
Paper+code+video here: tmlr.infinite-conf.org/paper_pages/...
More formally, the probability is uniformly distributed across all possible weather conditions.
The overall website about the OECD AI Capability Indicators project and policy work can be found here: aicapabilityindicators.oecd.org 6/6
The OECD AI Capability Indicators Technical Report has now been published and is available here: www.oecd.org/en/publicati... 5/6
We focused on defining AI creative capabilities - not only in the artistic sense but, perhaps more importantly, in terms of problem-solving and original thinking. This work is based on Giorgio's PhD thesis and current projects. 4/6
In essence, the question is how to assess the extent to which an AI system can perform a given task with a level of proficiency comparable to that of humans. 3/6
The current goal of the Working Group is to devise a series of metrics to quantify AI capabilities on a scale from 1 to 5 (similar to the βautonomyβ scale used for assessing the capabilities of self-driving cars) in order to derive a global standard for governments, businesses, and other orgs. 2/6
During the past year, our lab (in particular, Giorgio Franceschelli and myself) have been involved in the @oecd-ocde.bsky.social Working Group on AI and the Future of Skills. 1/6
Rightmove added the strapline βbelieve itβ to its logo. The reference to what an estate agent tells you is clear (donβt).
Comic. Someone is claiming to predict the exact date of a future earthquake. Should you listen? [flowchart: start] β NO (There are big earthquakes constantly, so if anyone ever *does* figure this out, it will be immediately obvious that their method works and then the worldβs seismologists will not shut up about it. You wonβt need this flowchart.)
Earthquake Prediction Flowchart
xkcd.com/3165/
This blogpost by Cal Newport is pretty good: "Forget Chatbots You Need a Notebook". calnewport.com/forget-chatb...
In fact, these look like insects and maybe they look like that to attract predators that will be able to spread them elsewhere. I would be very interested in an opinion from an expert. 2/2
More wonders of the natural world. Now some flowers of the plant in my office appear as if suspended in the void. This looks like a βtechniqueβ to invite pollinators. Or maybe it is a self-pollinating plant and these flowers contain seeds (?). 1/2
The Singularity is Not Near.
Comic. [two connected spacecraft hovering in space above the Earth] Houston, the view is superb. We can see the continents spread out below us, right where theyβve been since the Earth formed. [caption] I still canβt believe we developed spaceflight before we figured out that the continents moved.
Continents
xkcd.com/3159/
Link to the source: www.bbc.co.uk/news/live/c5...
And, by the way, regarding the value of a Computer Science degree in the "vibe-coding" era: one thing you learn in a good CS programme is how to design and build systems that are resilient and fault-tolerant. 2/2
Re the AWS outage, itβs quite cool that the BBC interviewed Ken Birman, who knows a thing or two about distributed systems. 1/2
We received the ACM UbiComp 2025 10-Year Impact Award for our paper βTrajectories of Depression: Unobtrusive Monitoring of Depressive States by means of Smartphone Mobility Traces Analysisβ co-authored with the great Luca Canzian. Paper here: www.mircomusolesi.org/papers/ubico... #ubicomp2025
New preprint with Charles Westphal and Stephen Hailes: "A Generalized Information Bottleneck Theory of Deep Learning". In this work, we introduce the Generalized Information Bottleneck framework, a synergy-based reformulation of the Information Bottleneck theory. arxiv.org/abs/2509.26327
Lots of headlines claim graduates are more likely to be unemployed than non-graduates. Graduate unemployment panic makes good headlines, but the data says otherwise. Great analysis by @jburnmurdoch.ft.com. www.ft.com/content/ea9e...
.
New preprint: βComplexity-Driven Policy Optimizationβ. The paper discusses a novel complexity-driven policy optimisation solution for efficient and robust approximation in Reinforcement Learning. Essentially, entropy alone is not sufficient, we also need "structure".
Received a copy of βThe Library of Lost Mapsβ by my colleague James Cheshire at @uclgeography.bsky.social in the internal mail. Really beautiful (and very interesting) book. Thanks a lot James!
For the geeks among you: just discovered the Apple Calculator app has a "Programmer" mode, complete with binary, hex, and related operations.
The problem is still relevant - we now have Amazon delivery vans instead to play with. 2/2
I was thinking about this some weeks ago. The Travelling Salesman Problem (finding the shortest path through a list of cities, visiting each only once), a classic in Computer Science/Maths, is becoming something of the past without actual salesmen on the roads. 1/2
Cactus.
Wonders of the natural world. I inherited an office cactus from a colleague who left. In a matter of days, out of nothing the cactus has developed a very long stem (with flowers).