Aidan Kelly

OpenAIs graphs against humanity for their release of GPT-5 is just *chefs kiss*. We've achieved PhD level intelligence but watch us make graphs where 50 < 47

What a ridiculous state of affairs. Terence Tao one of the greatest mathematicians of his generation getting his grants suspended 💩

Brilliant news and about time!

Build a Career in Data Science Podcast · Jacqueline Nolis and Emily Robinson · Build a Career in Data Science teaches you what data science courses leave out: from how to land your first job to the lifecycle of a data science proje...

Listened to the "build a career in data science" whilst making the transition from academia to data science. Found it an extremely useful resource and would recommend to anyone starting out on their journey. Even now I've been relistening and still v relevant: open.spotify.com/show/78Nft51...

Estimation of energy efficiency of heat pumps in residential buildings using real operation data - Nature Communications Heat pumps are essential for sustainable heating, but their performance varies widely. Here, authors analyze data from 1023 systems, revealing that 17% of air-source and 2% of ground-source heat pumps...

Plot taken from this paper: www.nature.com/articles/s41... (2/2)

Heatmaps comparing the Coefficient of Performance (COP) for Air-Source Heat Pumps (left) and Ground-Source Heat Pumps (right), plotted against outdoor temperatures (-5°C to 15°C) and supply temperatures (20°C to 60°C). Ground-source pumps consistently exhibit higher COP values (represented by lighter colors) across the temperature range compared to air-source pumps, especially at lower outdoor temperatures.

Love this plot of ASHP vs GSHP in terms of their COPs. Really nice illustration of the outdoor temperature dependency of ASHPs vs that of GSHPs (or lack thereof). Observed performance has been interpolated, but based off 2 years of sensor data for a sample size of 883 ASHPs and 124 GHSPs resp. (1/2)

So...many scenarios in which top tech people are predicting AGI within 5 years. Demis Hassabis thinking it'll be just after 2030, and I just can't help but think about how crazy the next 5 years are going to be if these predictions are anywhere close to the mark..

Mission Radar: emerging signals and innovation trends in low-carbon heating Our first edition of the series shares innovation signals for venture funding, research funding and policy trends in low-carbon heating

What's on your radar? 📡

Excited to share our first Mission Radar – a quarterly look at innovation signals spotted in our foresight work at @nestauk.bsky.social

We explored innovations around reducing home carbon emissions, linked to our sustainable future mission.

www.nesta.org.uk/feature/miss...

Mission Radar: emerging signals and innovation trends in low-carbon heating Our first edition of the series shares innovation signals for venture funding, research funding and policy trends in low-carbon heating

Link to full update: www.nesta.org.uk/feature/miss...

A radar-style infographic titled "Mission Radar: A sustainable future • Q1 2025" highlights innovation signals in low-carbon heating and green technologies. The radar is divided into three labeled segments: "Ventures and investments," "Research funding," and "Policy." Each segment shows blips representing trends or signals, with their size indicating potential impact on home decarbonisation and their position reflecting estimated time to impact (0-2 years, 3-5 years, 5+ years). Key trends in the 'Ventures and investments' quadrant include heat pump venture funding growth, smart energy control for utilities, and heat pump installation software. 'Research funding' trends include high-temperature heat pumps, heat batteries, and zero-emission boilers. The 'Policy' quadrant highlights BUS grants and the prominence of heat pumps in debates. The graphic supports Nesta's broader exploration of early innovation signals in the low-carbon heating space.

Really cool work done by my colleagues at @nestauk.bsky.social where they use AI to scan for innovation trends/signals. This first edition of Mission radar is on low carbon heating! Check it out.

The cover of Andrej Karpathy's youtube channel got me rofling 🤣

Thanks for posting here. At least from my Bluesky feed there is definitely quite an anti-AI vibe and I am trying to find people who are doing interesting AI stuff here but seriously struggling. If anyone got a good list to hand, would appreciate.

I've had a first stab at making a list of GenAI ppl on here but it's proving harder than anticipated! X is still the place to be for that content.

Let me know if you're interested in being added or know anyone who would be a good addition😎
bsky.app/profile/did:...

Feel like they serve slightly different purposes? I am not at the stage where I trust ai for search, whereas I know what my particular use-cases are for genAI. I think they should offer the option of turning it off 🤖.

OpenAI have released a few new models (gpt-4.1, gpt-4.1-mini and gpt-4.1-nano) for their api. I was watching their live release as we've got an application which uses their previous model and wondering about the jump in performance. I'm still guessing based on this plot, where are the values??? 😅

They are also completely different use-cases in my mind. I'll go to my specific AI if I have something I want to explore or get help with. It almost feels like they've gone to great expense to train an LLM and they're going to use it whether you like it or not.

Does anyone find it annoying that Google are force feeding us Gemini in every search. I used to quote that a ChatGPT query was worth 5 google searches in terms of energy. It's now much more accurate to say it's worth 5/6ths of a search. Same goes for Meta's AI feature in WhatsApp. Very in your face.

One of my favourite AI image outputs so far. Farage does love his foscii fuels!

OpenAI's image generator used to be terrible at making edits, so this is quite the leap and personally think this result looks very good. However it's a bit scary once you start thinking about some bad use-case scenarios. The guardrails for this have to be very strong.

I then decided I looked too tan and the Hollister logo has deformed and it gave me the following!

I then asked for it to make it more realistic by making my eyes blue and hair dark brown and it outputted the following.

OpenAI's new gpt-4o image generator/editor is very good. I put in my bluesky profile pic and asked for a Dragonball Z version of it and here's the output!

Big props to the whole team on this work @lizgzil.bsky.social, Roisín Gorman, Seth Caldwell, Daniel Lewis and also for input from @acjsissons.bsky.social, @andymarsden.bsky.social and others!

Onwards and upwards!

GitHub - nestauk/asf_heat_pump_suitability: Project to identify which small areas in Great Britain are likely to be suitable (or unsuitable) for which types of heating technology. Project to identify which small areas in Great Britain are likely to be suitable (or unsuitable) for which types of heating technology. - nestauk/asf_heat_pump_suitability

This provides us with some confidence in the methodology that we've used for scoring areas for heat network suitability.

A reminder that the data, code and tool are open source and we are always happy to hear your feedback on any of it!

Check out the github here

github.com/nestauk/asf_...

A scatter plot of the average heat network suitability scores when a DESNZ heat network zone is present (blue) and when it's absent (red) for several local authorities. We see that the heat network suitability scores are higher when the DESNZ heat network zone is present versus when it's absent.

We compared our scores across many of the local authorities included in the DESNZ work and found that our Heat Network suitability scores are on average higher in areas which include a DESNZ pilot HN zone versus those areas without a pilot zone in the selected local authority.

Well we've also been validating our heat network (HN) scores against outputs from the DESNZ heat network identification pilot. They've developed and implemented a methodology to identify areas suitable for HNs in 28 pilot locations across England.

So that's great! But you might ask how can we trust your model?

This is because we've now added in additional features such as grid capacity, anchor properties as well as conservation status of the area.

For properties in these areas, a multifaceted approach is required for low carbon heating.

However, looking at areas with medium population densities, we don't find this correlation. This is interesting as the suitability scores are calculated using household density as well as related factors such as rural/urban categorisation, garden size as well as proportion of flats.

A scatter graph where the x-axis shows the average suitability of air source heat pumps (ASHP), and the y-axis shows the suitability of heat networks (HN). Each point represents a local authority and the points are coloured by geographic region. We see a inverse relationship trend where local authorities with high ASHP suitability have low HN suitability, and local authorities with low average ASHP suitability have high HN suitability.

We've also published a data story, pulling out some insights from the map data:

www.nesta.org.uk/data-visuali...

Areas with very high or low household densities tend to have high or low suitability for shared heating technologies as one would expect. This is what we see in this figure.