Harrison Pim

Harrison Pim

@harrisonpim.com

head of data science at climatepolicyradar.org | harrisonpim.com

153 followers534 following65 posts

Harrison Pim reposted

truly a wild time to be alive when the cryptography world is split between those worried about quantum computers and those whose private key is “4”

Harrison Pim reposted

NEW | Solar power surge in 2025 stalls rise in fossil electricity worldwide ⚡📈 Clean power growth outpaced global electricity demand last year – with solar alone supplying THREE-QUARTERS of the increase in demand ☀️ ember-energy.org/lat...

Harrison Pim reposted

This is super interesting to read because I was very concerned about the water and AI thing so I went to my one scientist friend who literally worked on water reg/quality/usage and asked her about AI, data centers, water etc. She was far more concerned with agriculture abuses than AI.

Chris Brown@bluecology.bsky.social

Starting this collaborative blog was the best decision I’ve made this year. Carla Sbrocchi puts to words the uncomfortable feeling I have and explains how to move forward on the data centre environmental issue open.substack.com/pub/vitaexma...

Harrison Pim reposted

VICTORY! Government to open up the Land Registry - bringing to an end a thousand years of secrecy shrouding who owns England. I’ve been campaigning for this for ten years: the new Land Use Framework, published later today, makes it government policy. 1/ www.theguardian.com/environment/...

theguardian.comGovernment to lift paywall from large parts of the Land RegistryExclusive: finding out who owns land will become simpler under plans to make the best use of green spaces and hit net zero targets

absolutely baffled by the fact that there seems to be no way of staging individual line changes (rather than whole files) in git without going through an interactive/GUI mode 🥴 why why why can i not `git add my_file.txt --lines=5-10` ???

absolutely buzzzzzzing about the doc-to-lora stuff here. sakana use a hypernetwork to adapt an LLM to doc-specific knowledge in a single forward pass 🤯

Sakana AI@sakanaai.bsky.social

We are excited to introduce Doc-to-LoRA and Text-to-LoRA, two related papers exploring how to make LLM customization faster. By training a hypernetwork to generate LoRA adapters on the fly, models instantly internalize new info. Blog: pub.sakana.ai/doc-to-lora/ Doc-to-LoRA: arxiv.org/abs/2602.15902

Harrison Pim reposted

You can place pancakes, crepes, and scrambled eggs on a simplex where the variables are the ratios between milk, eggs, and flour, (with another small axis for baking soda.) We have explored too little of this manifold. More breakfasts can exist than we have known.

FARTS and SCIENCES

Daniel van Strien@danielvanstrien.bsky.social

Re-OCR'd the complete 1771 Encyclopaedia Britannica (2,724 pages) with a single command on @hf.co Jobs. - 0.9B model (GLM-OCR) ~$0.002/page ~$5 total on an L4 GPU Before (old Tesseract ocr) → After

Harrison Pim reposted

Sure. AI companies have ALWAYS been training their models on Wikipedia content, which under the free and open access model is available to anyone — including AI companies. Agreements like these require AI companies to limit and offset the strain they place on Wikimedia infrastructure.

This post is unavailable.

Harrison Pim reposted

We took a look at the UK's biggest ever offshore wind auction, and dug into what it will mean for energy bills and the "clean power 2030" plan ⬇️

Carbon Brief@carbonbrief.org

NEW – Q&A: What UK’s record auction for offshore wind means for bills and clean power by 2030 | @mollylempriere.carbonbrief.org @drsimevans.carbonbrief.org Read here: buff.ly/uEBxzdK

excellent, excellent round up of all things energy/decarbonisation in 2025. there are a couple of really exciting narratives in here!

Nat Bullard@nathanielbullard.com

Do you like charts? Oh yes you do. I've just published hundreds of them, as I do every year. www.nathanielbullard.com/presentations

Harrison Pim reposted

wikipedia turns 25 today! the last unenshittified major website! backbone of online info! triumph of humanity! powered by urge of unpaid randos to correct each other! somehow mostly reliable! "good thing wikipedia works in practice, because it sure doesn't work in theory" - old wiki adage

Sara Hooker continues to be my fav person pondering the landscape of work in ML. This essay is just as perceptive and provocative as the original hardware lottery paper which it builds upon - i love it

Sung Kim@sungkim.bsky.social

On the Slow Death of Scaling by Sarah Hooker "For the last decade, it has been hard to stray off the beaten path of accepted wisdom that scaling training parameters drives innovation. However, the relationship between training compute + performance is uncertain + rapidly changing."

Harrison Pim reposted

JustHTML by @emilstenstrom.bsky.social is a new Python library (no dependencies) that parses HTML according to the HTML5 specification and passes the 9,200 test html5lib-tests suite It's 3,000 lines of code mostly written by coding agents over a couple of months simonwillison.net/2025/Dec/14/...

simonwillison.netJustHTML is a fascinating example of vibe engineering in actionI recently came across JustHTML, a new Python library for parsing HTML released by Emil Stenström. It’s a very interesting piece of software, both as a useful library and as …

I’ve spent the last 10 years focusing on real-world applied ML, and following the proceedings of major academic conference like this from afar. With that context, it feels like a very personally meaningful milestone to have finally attended this year, with a package of work which bridges both worlds

brilliant news, obviously 🙌 though when compared to what's being done elsewhere, these numbers feel quite disappointing... other countries with similar solar potential have been adding capacity more than 5x faster than we have in the UK!

Harrison Pim reposted

Harrison Pim reposted

Four new models from Mistral today - all Apache 2 licensed, all vision-capable, and one of them is a 3GB model that can run in a web browser and answer questions about things it can see through the webcam! simonwillison.net/2025/Dec/2/i...

simonwillison.netIntroducing Mistral 3Four new models from Mistral today: three in their "Ministral" smaller model series (14B, 8B, and 3B) and a new Mistral Large 3 MoE model with 675B parameters, 41B active. …

Given where our research is going, I’m very excited to chat to folks working on graph machine learning, small NLP model fine-tuning, and mechanistic interpretability. Drop me a message if you’re up for a chat!

Harrison Pim reposted

for all talks of local-first AI, given controversies over datacenters, I don't think people are willing to hear that local LLM inference is *probably* less environmentally friendly than cloud inference the majority of the time.