RobinL

This resonates. Moreover, it's very easy for architects to assume that because different areas of the business use data about the 'same' thing, the thing must be the same.

But often the analysis requires a slightly different thing. Like: we want a master list of prisons. But is a prison a building, a collection of prisoners (such that the male prison and the female prison on the same site are different prisons), or the institution with that name managed under a particular contract?

There's a kind of inversion in how you do things: e.g. rather than navigating to settings until you find the screen time controls and changing the settings, you explain in natural language what to do, the agent asynchronously proposes the solution, and you just click accept or reject.

So I'm perfectly happy with a system of higher education that strongly rewards this behaviour

Unless I'm missing something, it's amazing how few free, _high quality_ materials are online.

Ultimately I'm interested in two things: genuinely fun games that make you do some maths, and quality visualisations that help make concepts easier to learn

To be honest I think both are quite limited by context length (in that they try to limit the context they send to the LLM and hence cost), and so I find myself using Gemini 2.5 in AI studio with the 1m context length, and asking it to generate instructions for Copilot (which seems to work pretty well)

It's personal taste, but this is significantly better than the last couple of fiction books I've read (which were both well reviewed).

I think it's good enough that it's hard to argue is emotionally vacuous, unless you define that to mean 'it was written by a machine'

I think increasing we'll find AIs are extremely good at emotional 'manipulation' (which I mean in the same sense has how a good tearjerker is in some sense emotionally manipulative).

I've used sqlglot to do this in the past, but doing it natively would be nice

input_data = duckdb.sql("SELECT * FROM read_parquet('...')")

step_1 = duckdb.sql("SELECT ... FROM input_data JOIN ...")

step_2 = duckdb.sql("SELECT ... FROM step_1")

final = duckdb.sql("SELECT ... FROM step_2;")

For context, I've tried it! I've been working on a free library for geocoding UK addresses quickly and accurately. It comes with the caveat that you need access to the dataset of all addresses you're geocoding against - which could be your own list, or a commercial product like addressbase: https://github.com/robinL/uk_address_matcher/

That said, a 100 year wait for social housing in Westminster is not surprising, and not as bad as it sounds. A typical professional couple in London could not afford a family home in Westminster, and it's extremely common for people in their 30s to move out of the centre when they have kids. I don't think this should be any different for social tenants: they should certainly not have to wait 100 years, but it seems reasonable for them to be offered housing outside of Westminster. The Guardian article I read about the 100 year wait specifically mentioned 3 and 4 bed family homes, and my immediate reaction was that no one else can get them either!

I think the bigger problem is the lack of affordable housing particularly for young families (social or otherwise) outside of London

Someone I know recently had a referral - it's pretty light touch, you just get a prompt appointment, and they do a minor op to remove the mole, and send it to the lab for testing. Luckily in their case, it wasn't cancer. But nothing in the process seemed weird, it was just the way of the GP escalating it because they couldn't be sure. Hypothetically, if the AI had been able to diagnose with higher certainty than the GP, all of this could have been avoided, so definitely room for improvement.

My perception is that it's caught up a lot. It still seems a bit slower to run agent, and slower to apply diffs. As the auto complete is a little worse. But it's no longer a no brainer to spend the extra on the cursor subscription.

For harder problems, I've also found myself reaching for long-context Gemini 2.5 pro using files-to-prompt, meaning I've been relying on copilot/cursor for less complex tasks

I can't help feeling like this is a vicious cycle - the lack of community facilities is causing greater isolation, causing a rise in health needs and so on.

- A library for fast, accurate matching (geocoding) of UK addresses that uses Splink under the hood: https://github.com/RobinL/uk_address_matcher

- An npm library that generates maths mental arithmetic problems that align to the UK national curriculum, that can be used to power maths games: https://github.com/RobinL/maths-game-problem-generator

- A breakout maths game that uses the above: https://github.com/RobinL/maths-game-problem-generator

> Gemini 2.5 barely made a ripple in the broader tech conversation.

But my sense (vibes from using it) is it's the very best model right now (though I have not tried the $200 a month exclusive openai offerings)

For example, I asked it to render a few lines of text on a medieval scroll, and it basically looked like a picture of a gothic font written onto a background image of a scroll

You will typically have a master/canonical list of addresses from an official source. In the UK that's Ordnance Survey's AddressBase.

You will then have 'messy addresses' that humans have written down. For simple addresses, they'll often be the same as the master version (e.g. 5 Rainbow Road, Hemel Hempstead, AB1 2BC).

But there are many harder addresses that exhibit lots of variations, especially flats and subunits.

These may all be the same: Flat A 1 High Street vs 1A High Street vs Basement Flat, 1 High Street

There no guarantee there will be a number in the address. In: 'THE OLD FARM COTTAGE PAD FARM BADGERCROFT ROAD PIKING', 'THE' and PAD FARM' may be missing, which doesn't seem like a problem until you find out there's also a PAD FARM COTTAGE on Badgercroft Road. There's no guarantee tokens will be in the same order.

FWIW, my work is open source, and it's here: https://github.com/RobinL/uk_address_matcher

It seems capable of generating a consistent style, and so in that sense quite useful. But if you want (say) a regional UK accent it's not even close.

I also find it confusing you have to choose a voice. Surely that's what the prompt should be for, especially when the voices have such abstract names.

I mean, it's still very impressive when you stand back a bit, but feels a bit half baked

Example: Voice: Thick and hearty, with a slow, rolling cadence—like a lifelong Somerset farmer leaning over a gate, chatting about the land with a mug of cider in hand. It’s warm, weathered, and rich, carrying the easy confidence of someone who’s seen a thousand harvests and knows every hedgerow and rolling hill in the county.

Tone: Friendly, laid-back, and full of rustic charm. It’s got that unhurried quality of a man who’s got time for a proper chinwag, with a twinkle in his eye and a belly laugh never far away. Every sentence should feel like it’s been seasoned with fresh air, long days in the fields, and a lifetime of countryside wisdom.

Dialect: Classic West Country, with broad vowels, softened consonants, and that unmistakable rural lilt. Words flow together in an easy drawl, with plenty of dropped "h"s and "g"s. "I be" replaces "I am," and "us" gets used instead of "we" or "me." Expect plenty of "ooh-arrs," "proper job," and "gurt big" sprinkled in naturally.

For the ETL side, where often whole-table access is good enough, I find Spark in particular very cumbersome - there's more than can go wrong vs. DuckDB and it's harder to troubleshoot.