When releasing information under the Freedom of Information Act, public authorities often attempt to redact documents by obscuring sensitive text using black boxes. When this is done incorrectly, t...

In my last post, I mentioned that I’d finally managed to crack how to extract meaningful data from ~3.2 million emails that I’d scraped from WhatDoTheyKnow. I thought it was worth jotting down. Ov...

FOI data After years of chipping away at it, I finally worked out how to extract the answers from FOI responses in a way that is not susceptible to hallucination, and that, as a welcome byproduct, ...

This is not quite a one armed bandit, but the closest I could get in a Gradio space 😁. https://huggingface.co/spaces/HMC83/request-generator It’s running one of the OLMo fine-tunes at the moment....

I recently trained a ~645M parameter LLM from the ground up. This was by far the largest model that I have attempted to create to date. I wanted to train a “good enough” base model that was fully m...

Искам да запиша за малките неща. Върнах се към този език, който бях изоставила. С езика се връща и ентусиазмът. Няма бързината и топлината на родния език, но има нови думи и нов начин да видя нещат...

This week I wanted to explore whether a model that is smaller than the average Android app could outperform billion-parameter LLMs on a specialised task. To test this out, I trained a 22 million pa...

Objective Travel on as many different types of public transport as possible within a 12 hour window, starting and finishing in the same location. Ground Rules For the purposes of this challenge, I...

At the start of the year I scraped data from over 800,000 FOI requests from WhatDoTheyKnow to look for patterns in the data. This sent me down a number of rabbit holes, including experiments with a...