RobinL

The problem is that people write addresses down in different ways (I'm in the UK)

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

I appreciate this comment. I'm currently working hard on something seemly straightforward (address matching) and sometimes I feel demotivated because it feels like whatever progress I make, the bitter lesson will get me in the end. Reading your comment made me feel that maybe it's worth the effort after all. I have also taken some comfort in the fact that current LLMs cannot perform this task very well.

I agree. Bluey is a great show. As a dad it can either make you feel inadequate or motivate you to to better. You just have to accept no dad can be Bandit all the time, but try to be Bandit as much as possible.

I'm surprised at how poor this is at following a detailed prompt.

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.

Yeah I think that's fair, especially from the 'end consumer of the data' point of view, and doing things like row-level permissions.

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.

Ah nice - reading that made me feel good! Appreciate the feedback!

Yes - agree! I actually wrote a blog about this just two days ago:

May be of interest to people who:

- What to know what DuckDB is and why it's interesting

- What's good about it

- Why for orgs without huge data, we will hopefully see a lot more of 's3 + duckdb' rather than more complex architectures and services, and hopefully (IMHO) less Spark!

https://www.robinlinacre.com/recommend_duckdb/

I think most people in data science or data engineering should at least try it to get a sense of what it can do

Really for me, the most important thing is it makes it so much easier to design and test complex ETL because you're not constantly having to run queries against Athena/Spark to check they work - you can do it all locally, in CI, set up tests, etc.

But the same is true of gas, and in the UK at least they cost about the same per kWh of output heat (electricity is more expensive, but this is balanced out by the higher efficiency of the heat pump)

I've heard this several but don't really understand the rationale. Can't you just put in a bigger heat pump? Or top up an air to water with an air to air?

Thank you!

This sounds great - would love to hear a little more about the prompts. Are you literally just asking 'write me a context.md that explains how feature x works' or something like that?

Apologies for possibly stupid question but where can you use it right now? Just on 'direct chat' on https://lmarena.ai/ or is there a better alternative? Or do you have early access?

They were doing web search before open ai/anthropic, so they historically had a (pretty decent) unique selling point.

Once chat gpt added web browsing, I largely stopped using perplexity

If I was a billionaire I'd have humans do all these types of things. I can imagine AI would be useful to give me an unbiased/non sycophantic view.

A big reason I blog is that I find writing something down is essential to clarifying my thoughts.

Once it's written down, I might as well put it online, and it has the added advantages that I can simply link to the material rather than having to explain it over and over again e.g. in emails.

Thanks - having tried Co-Pilot again today after a 6-month Cursor hiatus, I think this is a good summary

Update: I have used co-pilot agent mode for a couple of hours today.

It's definitely catching up with Cursor but not there yet. In particular: - Edits take quite a bit longer to apply, breaking flow - Autocomplete predictions (equiv. of Cursor Tab) not as good

But in the past 6 months or so it's gone from being pretty hopeless to very useful. If I was forced to use it instead of Cursor it wouldn't be a huge deal any more.

Can anyone speak to weather it's worth going back to co-pilot from cursor. On the face of it $10 a month for unlimited messages looks compelling. Is it really unlimited? From these videos it's starting to look pretty similar to cursor...

Interesting

You might find it amusing to compare it to: https://hn-wrapped.kadoa.com/timabdulla

(Ref:https://news.ycombinator.com/item?id=42857604)

Whilst I had tried R1 before, I hadn't paid attention to how fast it was. I just tried some similar prompts and was pretty impressed with speed and quality. I think o3-mini was still a bit quicker though.

Wow - this is seriously fast (o3-mini), and my initial impressions are very favourable. I was asking it to layout quite a complex html form from a schema and it did a very good job.

Looking at the comments on here and the benchmark results I was expecting it to be a bit meh, but initial impressions are quite the opposite

I was expecting it to perhaps be a marginal improvement for complex things that need a lot of 'reasoning', but it seems it's a bit improvement for simple things that you need doing fast

Has anyone tried to link and dedupe the various datasets using a probabilistic linkage tool like Splink?

https://moj-analytical-services.github.io/splink/

(Disclaimer: I am the lead author, but the tool is FOSS)

I also wish this was more widespread. I do it for my FOSS data linkage software here: https://moj-analytical-services.github.io/splink/topic_guide...

I'd also love to see some high quality, well researched information on this.

For instance I recently saw this exchange: https://x.com/i/bookmarks?post_id=1881090543956222293

I'm generally a proponent of renewables, but I think it's also important to take seriously quantitative arguments against them. But without good sources, it's impossible to work out who's right.

This is useful to suggest solar will win in the end: https://ourworldindata.org/grapher/levelized-cost-of-energy?...

But much less clear how well wind, especially offshore, can compete.

This is also decent but doesn't get much into cost: https://www.sustainabilitybynumbers.com/p/can-solar-and-wind...

There's a theory an economics that says that the more different prices a provider can charge the more of the surplus they capture (ie they can tilt that percentage towards the seller and away from the buyer).

Of course, if they're a monopoly provider and the buyer really needs it, they have to cough up. But generally there are substitute products. So the buyer would do well to look for an alternative that doesn't do differential pricing to capture more surplus for themselves.

Schedule a call is a huge red flag to me because:

- it implies differential pricing, meaning they will charge you as much as possible both now and in the future (when you may be locked in)

- it usually obscures what the product actually does

Differential pricing is really pernicious because if the product happens to be super valuable to you, they're likely to find out and charge you even more

There are two components to the price of driving:

1. The congestion charge, if it exists 2. The opportunity cost of the time taken

It's plausible to me that whilst total cars will go down (supply and demand) you may get more high income customers (because they have a high opportunity cost of time) and hence some businesses may do better.

It's also not obvious it should increase costs of e.g. deliveries to businesses because the the delivery driver will spend less time waiting in traffic.

I wonder if any research is being done on either of these effects

> It's impossible to improve performance when you have to go out to the DB anyway;

That's not right. There are many queries that run far faster in duckdb/datafusion than (say) postgres, even with the overhead of pulling whole large tables prior to running the query. (Or use like pg_duckdb).

For certain types of queries these engines can be 100x faster.

More here: https://postgres.fm/episodes/pg_duckdb

I've only recently started with uv, but this is one thing it seems to solve nicely. I've tried to get into the mindset of only using uv for python stuff - and hence I haven't installed python using homebrew, only uv.

You basically need to just remember to never call python directly. Instead use uv run and uv pip install. That ensures you're always using the uv installed python and/or a venv.

Python based tools where you may want a global install (say ruff) can be installed using uv tool

You can use duckdb on a single machine. It's also indexless (or more accurately, you don't have to explicitly create indexes)