Hardware performance counters are great but you usually only use them on one-off whole program debug runs. I've prototyped sampling them at every OxCaml Runtime Events span instead, at almost no cost. Eventually this could surface issues across a whole fleet of services. toao.com/blog/free-pe...

Tessera: A Temporal Foundation Model for Earth Observation with Sadiq Jaffer and Frank Feng

Last week Frank Feng and I joined Robin Cole on his satellite image deep learning podcast to talk about Tessera: how it works, how it differs from existing models, and our future plans. Thanks for hosting us, Robin!

www.satellite-image-deep-learning.com/p/tessera-a-...

It still surprises me you can get good performance with even tiny models on top of TESSERA embeddings. Here's how to find solar farms in the UK with a small ~42k CNN: toao.com/blog/earth-o...

There's relatively little LLM training data for niche languages and this causes poorer coding agent performance. I think this is an existential threat for smaller language communities like OCaml.

My talk at the OCaml workshop gave some actionable steps to mitigate that: toao.com/blog/ai-exis...

Every OCaml talk needs a pun, and @sadiq.toao.com is no exception #icfpsplash25

Our lightning talks session opens with @sadiq.toao.com demonstrating TESSERA, their new geospatial foundation model that is FAIR and global #icfpsplash25

Can AI detect hedgehogs from space? Maybe if you find brambles first. Cambridge researchers use satellite-based bramble detection as a proxy for mapping hedgehog habitats.

Not how I expected to make my @arstechnica.com debut but I'll take it arstechnica.com/ai/2025/09/c...

Fun field trip today trying to validate a colleague's bramble detecting model: toao.com/blog/can-we-... with @anil.recoil.org

A good point. Was being generated but not linked anywhere. Fixed now. Thanks!

Some fun OCaml GC projects here with @sadiq.toao.com and @kcsrk.info if any students are looking for projects involving programming languages toao.com/blog/ocaml-0...

The most incredibly fun part of this Nature comment on evidence synthesis we published today is that the cartoonist (David Parkins) also did Beano and Dennis the Menace (!) A true legend. www.nature.com/articles/d41...

The rapid rise in AI-generated fraudulent academic papers is "poisoning" scientific literature, say Cambridge researchers in Nature magazine today. But though AI is the problem, it could also help in ensuring the integrity of scientific discovery... buff.ly/AuSNcGd
@anil.recoil.org @sadiq.toao.com

I'm pleased to announce OxCaml!

OxCaml is Jane Street's branch of OCaml. We've given it a new name and a snazzy logo, and done a bunch of work to make it easy for people to try.

Careful design of Large Language Model pipelines enables expert-level retrieval of evidence-based information from syntheses and databases Wise use of evidence to support efficient conservation action is key to tackling biodiversity loss with limited time and resources. Evidence syntheses provide key recommendations for conservation deci...

New paper out today on how the careful design of LLMs is crucial for expert-level evidence retrieval in conservation (but with implications for any evidence synthesis pipeline across other fields) 🌍 doi.org/10.1371/jour... and anil.recoil.org/news/2024-ce... for a summary

One thing I probably should highlight more in the post is that the proprietary models (like Claude and Gemini) that most students currently have access to can already ace the assignments.

This is a thorny question and mostly comes down to what we're trying to teach. I wonder if a progressive approach where at early stages of teaching there is no automatic tooling but as critical skills are learnt more can be automated. It's a bit of a moving target at the moment though.

Just how good are locally hostable code models on Cambridge first year OCaml assignments? @anil.recoil.org , @jon.recoil.org and I wanted to find out, so ran some tests. TL;DR Qwen3 means we might need new assignments. toao.com/blog/ocaml-l...

If you are using llama.cpp, here's a workaround using grammars for getting JSON structured output from Deepseek R1 and distills: toao.com/blog/json-ou...

The potential for AI to revolutionize conservation: a horizon scan Artificial Intelligence (AI) is an emerging tool that could be leveraged to identify the effective conservation solutions demanded by the urgent biodiversity crisis. We present the results of our horizon scan of AI applications likely to significantly benefit biological conservation. An international panel of conservation scientists and AI experts identified 21 key ideas. These included species recognition to uncover 'dark diversity', multimodal models to improve biodiversity loss predictions, monitoring wildlife trade, and addressing human–wildlife conflict. We consider the potential negative impacts of AI adoption, such as AI colonialism and loss of essential conservation skills, and suggest how the conservation field might adapt to harness the benefits of AI while mitigating its risks.

Part of our @ai.cam.ac.uk project on AI in Conservation was published in TREE today. We gathered conservation scientists and AI experts and looked at the key conservation areas AI could revolutionise: www.cell.com/trends/ecolo...

Working to surface challenges faced by folks at the coal face.

Data in research contributions from @orbenamy.bsky.social @sadiq.toao.com @scotthosking.bsky.social Stefan Scholtes, Vasco Carvalho, Mireia Crispin and a foreward with Jess Montgomery @dianecoyle1859.bsky.social @ginasue.bsky.social

New preprint from our work on using LLMs to accelerate conservation evidence synthesis across millions of papers. We crosscheck 3 retrieval strategies against 10 LLMs and benchmark against human experts and find quite a bit of variance https://www.researchsquare.com/article/rs-5409185/v1