ChatGPT? On the cusp of a revolution, or just a fancy toy?

So we’ve been trying for ages to figure out how to do a planning application to replace the windows on our grade II listed farmhouse and never got past the various documents you have to provide as we could figure out what to write. ChatGPT just did it in seconds 😂

Q: Can you provide an example of a design and access statement to replace the windows on a grade two listed farmhouse which is compliant with UK planning law?

A: Design and access statements are documents that provide information about the design of a proposed development and how it will be accessed. They are often required as part of the planning application process in the UK. Here is an example of a design and access statement for new windows on a grade II listed farmhouse:

Design and Access Statement: New Windows for Grade II Listed Farmhouse

Introduction:
This design and access statement has been prepared in support of a planning application for the installation of new windows at a grade II listed farmhouse in [location]. The purpose of this statement is to provide information about the design of the proposed windows and how they will be accessed, in order to demonstrate how the proposal complies with relevant planning policies and guidelines.

Design:
The proposed windows will be designed to match the style and appearance of the existing windows as closely as possible, in order to preserve the character and appearance of the listed building. They will be made from timber, with double glazing to improve energy efficiency.
In order to minimise the impact of the new windows on the listed building, they will be installed in the same locations as the existing windows. The size and proportion of the new windows will also be the same as the existing windows, in order to maintain the balance and symmetry of the building’s facade.

Access:
The new windows will be installed by a team of specialist contractors who have experience working on listed buildings. They will be accessed using scaffolding, which will be erected in a manner that minimises the impact on the surrounding area. The scaffolding will be removed once the installation is complete.

Conclusion:
The proposed installation of new windows at the grade II listed farmhouse will enhance the energy efficiency of the building and improve its appearance, while also preserving its character and appearance. The design of the windows and the method of access have been carefully considered in order to minimise any impact on the listed building and the surrounding area, and to ensure compliance with relevant planning policies and guidelines.

Q: Can you provide example heritage statement which applies to a Grade two listed farmhouse for an application to replace windows?

A: Sure! Here is an example of a heritage statement for a Grade II listed farmhouse that is seeking planning permission to replace its windows:

Heritage Statement
The subject property is a Grade II listed farmhouse, located in the rural village of XYZ. The farmhouse is a well-preserved example of a traditional vernacular building, with a timber-framed structure and clay lump walls. It is thought to date from the 17th century, and is a rare surviving example of a type of building that was once common in the area.

The farmhouse is of architectural and historical significance, and is listed on the National Heritage List for England as a building of special interest. It is a key element in the local historic landscape, and makes a positive contribution to the character and appearance of the village.

The proposed works are for the replacement of the farmhouse’s windows. The existing windows are original to the building, and are in poor condition. They are single-glazed, and are not energy efficient, resulting in high heating costs for the property. The proposed replacement windows are of a traditional design, and will match the appearance of the existing windows as closely as possible. They will be double-glazed, and will be more energy efficient, reducing heating costs for the property.

The proposed works will have a minimal impact on the character and appearance of the farmhouse. The replacement windows will be of a similar design and material to the existing windows, and will be in keeping with the traditional vernacular character of the building. The use of energy efficient windows will also have the added benefit of reducing the building’s carbon footprint, and will contribute to the sustainability of the property.

In conclusion, the proposed works for the replacement of the farmhouse’s windows are necessary to address the building’s poor energy efficiency, and will have a minimal impact on its character and appearance. The works are therefore considered to be appropriate, and are recommended for approval.

It’s just a glorified copy and paste that, no? ChatGPT looks more and more like a natural language interface to a search engine to me. Valuable… but the results still need sense checking, just as with any search engine that crawls content and selectively offers it up as if fact.

It’s just a glorified copy and paste that, no?

Possibly, but we tried (or rather Mrs Daz did) finding example statements on the web with not a lot of success. It’s definitely a lot more than an interface to a search engine though. Even if it is it’s saved hours of research and even more in trying to write the damn things.

– it’s taught me how to manipulate geotiffs in python!

I had to write a GeoTiff decoder once, horrible file format IIRC..

Even if it is it’s saved hours of research and even more in trying to write the damn things.

How do you know if it’s any use to you though? A real expert system would be asking you further questions to ensure its tailored answer is applicable to your situation.

I had to write a GeoTiff decoder once, horrible file format IIRC..

Indeed. It does have an insane kind of logic to it tho! “We need a data format that has a grid of squares with each square being associated with a different measurement level from 0-X…… sounds a lot like an image!”

Thankfully someone has already made some tools to manipulate them (gdal, rasterio). GDAL is a horror show, but GPT has shown me the way.

How do you know if it’s any use to you though? A real expert system would be asking you further questions to ensure its tailored answer is applicable to your situation.

You can provide it with further information in subsequent prompts and it’ll modify it’s answer accordingly.

How do you know if it’s any use to you though?

I’ll report back on the resuts of the application 😀

Seriously though, all we needed was a template to get us started and now we have that. I suppose I could’ve spent ages rooting around the planning portal and searching on google but that took 10 seconds. Took a couple of minutes to edit the heritage statement and we’re submitting it as is so we’ll see.

Good luck!

You can provide it with further information in subsequent prompts and it’ll modify it’s answer accordingly.

Yes, but what’s lacking there is the system helping you by it leading the refinement of the answer. As an expert, or expert system would.

Yes, but what’s lacking there is the system helping you by it leading the refinement of the answer. As an expert, or expert system would.

Ah right yes see your point.

How does it learn without getting feedback? For example was the planning application accepted? If not why not?

How does it learn without getting feedback?

Not really how machine learning works, it gets re-trained on new / larger data sets periodically and they add it specific data sets to correct specific issues.

Eg I was reading somewhere that every time a Tesla driver does something that the car’s auto pilot doesn’t expect (auto pilot runs in the background on all cars all the time, just disconnected), the car uploads all the sensor feeds (video) of that incident. They then retrain it with the enhanced data set, trying to improve it’s match between how it behaves and how humans behave. So if auto pilot fails to spot cyclists, but drivers keep braking for something it doesn’t understand, it will ‘learn’ that new behaviour over time…

Yeah, Tesla isn’t a car company…

It’s going to be a game changer in the workplace. I work in finance and I’ve been trying it over the last couple of days. It saves sooooo much time on things that take ages if I have to start from scratch, especially with my increasingly short attention span. A few things I’ve used it for in work – write a job description for role x, rewrite this paragraph in bullet points, correct the grammar in this letter, explain ESG to a five year old, write a policy for collecting debts. In my job I think it’s the biggest time saver since Excel. In a year everyone will be using it or (Google/MS versions) for everything. Can’t be long until the singularity now 😳

How does it learn without getting feedback?

You can give it feedback. And it does learn from it. You can get it to accept and repeat falsehoods to keep you happy sometimes, but it’s ready for that… you have to be cunning/persuasive. I don’t think that anything learned this way is accepted beyond your session though. Would be interesting to try getting the same falsehood accepted in many sessions… and see if that changes that. Hmm… ChatBot wars…

explain ESG to a five year old

I’d start with this track…

Seems the site is at capacity. Will it be easy to cater for the expected surge in demand for this kind of site? Especially if it’s to become so intrinsic to the workplace.

if you’re asking, “has it achieved sentience?” then yes, I’d say it’s passed the Turing Test 😀

Just wait until it inhabits an ambulatory doll designed to act as a protector and companion for a grieving child…

ChatGPT looks more and more like a natural language interface to a search engine to me. Valuable… but the results still need sense checking, just as with any search engine that crawls content and selectively offers it up as if fact.

From what I’ve been reading, that seems to be the case, but it’s got Google sweating a bit, because it’s better than anything they’ve got.

It’s just a glorified copy and paste that, no?

Isn’t that what a lot of human writing is?

I’ve just done a project health check report for a client, basically I was just looking for stuff they’d done wrong and telling them why it’s wrong. That could clearly have been done by an AI like this, and better because it could have examined the entire codebase.

Seems the site is at capacity. Will it be easy to cater for the expected surge in demand for this kind of site? Especially if it’s to become so intrinsic to the workplace.

It won’t remain free. It must be using quite a bit of data centre resource to run. They will either start charging for it or it will become riddled with adverts.

It’s just a glorified copy and paste that, no? ChatGPT looks more and more like a natural language interface to a search engine to me.

Yeah, but we all know that if Daz googled it these days, he’d have had page after page of bot created shit with addresses like http://www.everythinganswers.info that all said:

Many people are interested in Design And Access Statement. This website will give you all the Design and Access Statement information you will need. Some of the most common questions about Design and Access Statement are: 1. How can I find out more about Design and Access Statement….

Google have ruined their own product IMO. I’ve found ChatGPT really good as a supplementary search engine – in cases where you don’t really know what to Google for in the first place, it’s really really useful IMO

Google have ruined their own product IMO.

Agreed, because of monetizing it. That’s not happened to this… yet..

Looks like the STW massive are overloading it right now : ChatGPT is at capacity right now

Google have ruined their own product IMO. I’ve found ChatGPT really good as a supplementary search engine – in cases where you don’t really know what to Google for in the first place, it’s really really useful IMO

just be aware that it very definitely isn’t a search engine (at the moment anyway). It doesn’t have access to the web, it knows what it knows as it’s been trained on a (web-sourced) dataset, but doesn’t know much about anything that happened after 2021 apparently! That said I was using it a lot over the weekend to help with programming, it was my first port of call for everything and solved 95%+ of my questions without resorting to Google!

Looks like the STW massive are overloading it right now : ChatGPT is at capacity right now

you should be able to refresh/try again in a few mins & get in!

It won’t remain free. It must be using quite a bit of data centre resource to run. They will either start charging for it or it will become riddled with adverts.

they are already charging for their existing product (GPT-3), you buy “tokens” which are consumed as it generates replies (longer replies costing more tokens) which is how a lot of API access works. But whether they go down this route again, adverts, or one of the big boys buys them out, who knows! But yeah it can’t remain free forever. It’s worth it for them at this stage as they’re getting hugely valuable feedback/testing out of it. I’d gladly pay though for a version that was instantaneous & didn’t have capacity problems!

I sure it is great for churning out code etc, if you accept that there are bound to be errors and it will need debugging like any other code.

My big concern with stuff like this is that people will treat what it says as gospel. Can you imagine the poor GPs having to deal with people self diagnosing with this and marching in armed with the ‘truth’.

It is only as good as the dataset it is trained with. A large amount scientific knowledge is hidden behind paywalls or owned by private companies. It simply won’t have access to this and will be generating answers based on superficial snippets in the public domain. It will then produce plausible sounding superficial snippets.

I sure it is great for churning out code etc, if you accept that there are bound to be errors and it will need debugging like any other code.

My big concern with stuff like this is that people will treat what it says as gospel. Can you imagine the poor GPs having to deal with people self diagnosing with this and marching in armed with the ‘truth’.

But that already happens! Been happening for years, as I’m sure some of our resident GPs will testify – ChatGPT is no different to Google in that respect.

Regarding code, if you know your stuff it’s probably easy to see where it goes wrong. But as an armchair dabbler who occasionally knocks up a webpage for internal work purposes, I’ve been very impressed. And on the one occasion it served up buggy code, I pointed this out and it debugged it by itself…

My big concern with stuff like this is that people will treat what it says as gospel. Can you imagine the poor GPs having to deal with people self diagnosing with this and marching in armed with the ‘truth’.

yeah, as said, that’s a human problem really, no different to people believing everything they read via Google/Facebook/tabloid or hear down the pub etc!!

But as an armchair dabbler who occasionally knocks up a webpage for internal work purposes, I’ve been very impressed. And on the one occasion it served up buggy code, I pointed this out and it debugged it by itself…

the fact that it takes feedback so you can ask it to do something differently, etc is very cool. I’ve found it’s great at fixing obvious-but-I-can’t-see-it syntax errors as well which seem to plague me for far too long when I’m tired 😂

btw has anyone tried their other tool for creating AI-generated images? That’s @binners out of a job too, then 😂

Yup. It can do amazing things. The results are like an image smoothie though, scraping together many pieces of genuine human creativity, blending it together, and smearing it across the screen.

Think it’s broken 😩

Annoying cos I’ve got a load of work for it to do and some past chats I need to access. Victim of its own success I guess.

btw has anyone tried their other tool for creating AI-generated images?

I’ve done a fair few… they always come out slightly… deranged.
I’ll find some

This was ok

but.. erm

@desperatebicycle that dog one is great! I can totally see a London-hipster-type spending proper money on that in one of many little pop-up galleries in our trendy little seaside tourist town 😃 I’ve certainly seen a lot worse pictures for sale! 😂

I can only take credit for the description: “Two dogs wearing glasses and smoking a cigar, digital image” 😆

Another slightly odd result

It’s worth reading what Gary Marcus has to say about large language models such as ChatGPT. I don’t agree with all of his opinions, but it’s interesting and a counterpoint to all the hype. The blog post on ChatGPT errors is a good use of 5 minutes.

One of my neighbours used ChatGPT to write some articles. She gave it ideas, it wrote articles, she edited them and published them. It saved her a bunch of time and she was happy with them.

I asked her if she gave ChatGPT credit and she told me that she asked it if it should have credit and it said no.

The blog post on ChatGPT errors is a good use of 5 minutes.

Sure. It’s definitely not infallible (and those guardrails are annoying, but you can see why they exist, at least for the time-being) but it’s still a useful tool and I’m not going to write it off just because it sometimes makes stuff up 😂

I’ve experienced errors – notable when I asked it for Scrabble words with 2 “U”s and it returned a list of words, half of them patently featuring only one 😃 And I’ve had some incorrect code, but I was able to fix the part that was wrong easily (the rest was perfect).

I suppose it would be different if the errors were the type I couldn’t easily spot. Still, you have to remember it’s really early days still for this kind of thing. I’d say the hype/bandwagon about ChatGPT being “useless” is probably bigger then the hype in favour of it actually on sites like Reddit etc (which is the same with any new technology, to be honest, you only have to check it immediately after Apple make any kind of announcement whatsoever 😃)

I can only take credit for the description: “Two dogs wearing glasses and smoking a cigar, digital image”

the results (for Dall-E specifically!) seem equal parts “prompt engineering” (a fairly new field obviously, a lot of people I’ve seen criticising CGPTs effectiveness though are just bad at this!) and luck… re-running the exact same prompt gives wildly different results (I’ve burned through my free credits pretty quick just messing around with this!)

Another slightly odd result

3 hands? Trippy 😃

Judge used ChatGPT to help with making a court decision.
https://www.vice.com/en/article/k7bdmv/judge-used-chatgpt-to-make-court-decision

Challenge accepted!
I will get ChatGPT to replace JHJ.
To be fair I could probably do that on a Spectrum but this will get more youtube views…