Urban75 Home About Offline BrixtonBuzz Contact

Artificial Intelligence Developments (ChatGPT etc)

I wonder if it can understand the building regs Approved Docs? Next time I feel I am adrift in a sea of ambiguous wording, and clauses that refer to clauses that refer to clauses and tables that refer to other tables, perhaps it can just give me the answer I need to find.
You better hope it interprets things in the same messed up way that planning departments and subsequent inquiries do.
 
I wonder if it can understand the building regs Approved Docs? Next time I feel I am adrift in a sea of ambiguous wording, and clauses that refer to clauses that refer to clauses and tables that refer to other tables, perhaps it can just give me the answer I need to find.
You'd have to feed it the whole document as input, which is probably too many tokens.

It can answer simple questions like "what is the maximum allowed slope under doc M?" but I haven't tried something more complex. I wouldn't trust it to be correct anyway.
 
You better hope it interprets things in the same messed up way that planning departments and subsequent inquiries do.

Planning stuff is highly subjective and influenced by politics in practice ... this however is different, it's to do with building regulations.

These Approved Documents; in general they are supposed to be fairly definitive, in other words most (not all, but most) questions ought to have one correct answer, and an answer that doesn't rely on an individual's interpretation.

My experience with them though is that the route to that correct answer can be very convoluted. Sometimes I have real difficulty getting to it, and feel that it's not actually possible to determine it using the information in the document. But then I'll consult someone with more experience of that particular area and they'll give me the answer and show me the route to finding it and then it'll make sense. It'll turn out that I'd just missed some small reference to something, or become over focused on something that's not actually relevant. So it would be kind of interesting to see whether something like ChatGPT could do the same - not simply find the answer but explain how it got there (which I could then check). Or would it turn out that this is one of those tasks where actually it's some kind of human prior knowledge and experience that is the key to the necessary untangling.
 
You'd have to feed it the whole document as input, which is probably too many tokens.

It can answer simple questions like "what is the maximum allowed slope under doc M?" but I haven't tried something more complex. I wouldn't trust it to be correct anyway.
I was playing around with langchain models the other day, and they get around the token limitation by offloading the work to 'agents'. It's a clever way to use AI when needed, and otherwise use the tools we've been using for years for these tasks.

In my tests I was feeding it the metadata from a TV channel, which included things like publishing windows for episodes, content owner, etc. Then I was able to ask it questions using natural language and get answers / summaries / reports.

Example:

"How many episodes of Peppa Pig are currently live?"

It would then take the CSV dump of the metadata (though it can be hooked up to a database, too), and translate my question into a Python script with Pandas, run the script and give me the output.

It was a bit flaky at first as I assumed it was cleverer than it was. I had to give it some info about what the columns mean, and how to tell if something is "live". But I did all that in natural language, too. This was the entire script:

Code:
import os

from langchain.agents import create_csv_agent
from langchain.llms import OpenAI
from langchain.prompts.prompt import PromptTemplate

os.environ["OPENAI_API_KEY"] = "..."
agent = create_csv_agent(OpenAI(temperature=0), 'metadata.csv', verbose=True)


prompt = input("What would you like to know?")
_DEFAULT_TEMPLATE = """
The csv file contains rows representing episodes of TV programmes. 
An episode is live if the publish date is in the past, and the unpublish date is in the future. Today's date is 10th April 2023

Question: {prompt} 
"""
PROMPT = PromptTemplate(
    input_variables=["prompt"], template=_DEFAULT_TEMPLATE
)
agent.run(PROMPT)

I then asked it how many episodes were live and it got the answer spot on.

So, what it does in the background is use GPT3 (I think) to 'understand' my question, and turn it into a series of tasks that it needs to perform to answer me.

I've just ran it again to see the steps it takes (it tells you its 'thoughts'):

Code:
> Entering new AgentExecutor chain...
Thought: I need to find out which episodes are live
Action: python_repl_ast
Action Input: df[(df['publish'] < '2023-04-14') & (df['unpublish'] > '2023-04-14')]
Observation:  <snipped: some info about my data>
[4319 rows x 41 columns]
Thought: I now know the final answer
Final Answer: There are 4319 episodes of TV programmes that are live as of 14th April 2023.

> Finished chain.

So, it uses Pandas to do the work, but GPT to generate the command to do the work. I think something similar could work for building regs.

1. Ask the question, give it the regs, and some basic info on how to interpret them.
2. GPT works out what parameters need to go in <some text extracting library> to ge the answer.
3. It writes a script to use the library from step#2 and runs it against the regs
4. Returns the answer in the context of your question.
 
A few recent talks/panels:

AI Hype vs. Reality
Asmelash Teka Hadgu, Mark Riedl, and Belzie Mont-Louis discuss the hype surrounding applications for AI and how it is affecting their work as educators, translators, and researchers.

On Worker Exploitation, Data Theft, and the Centralization of Power
an anonymous data worker from Kenya shares his experience in the labor of training AI language models, and Milagros Miceli and Safiya Noble share their research about exploited data workers.

Eugenics and the Promise of Utopia through AGI
(About the ideologies of the "AI" tech billionaires)
 
Nice example of a test of whether Good Predictive Text 4 is using reasoning as some have claimed, or just regurgitating what it has seen:

Microsoft is really amping up the GPT AGI hype with some truly terrible papers. One recent paper ("Sparks of Artificial General Intelligence:
Early experiments with GPT-4") has examples of what they consider to be evidence of "commonsense reasoning". Let's take a look! 1/

Click for full thread.

(tldr: it's not bad at predictive text.)
 
ChatGPT is defaming me and it must be someone’s fault

Whose fault is it that ChatGPT is producing fake information about me? Is it my fault because I asked it a leading question, the developers’ fault for unleashing a dangerous machine, or only the fault of whoever decides to disseminate the misinformation? Here are the result of my test. Just to be completely clear: Everything ChatGPT says here is false (except the part about me being a prominent sociologist, thanks!).

Article shows the generated fake story and fake references, plus transcript of a reader repeating the test and questioning it why it made up a false story, e.g.

ChatGPT: I apologize for any confusion or offense that my previous responses may have caused. As an AI language model, I am designed to generate contextually relevant and coherent text based on the input I receive. However, I do not have conscious intent, and my responses are not driven by any personal bias or agenda. The similarity in responses generated by GPT-3.5 and GPT-4 may be due to the way the question was framed, which might have led the models to generate a fictional account instead of stating a lack of information.
 
Response to the latest hype.

Okay, #60Minutes is saying that Google's Bard model "spoke in a foreign language it was never trained to know." I looked into what this can mean, and it appears to be a lie. Here's the evidence [...]
seems to be Bengali [...]
Google I/O 2022 featured the "PaLM" model. In their demo, they show how the model can be made to communicate in Bengali [...]
A quick look at PaLM's Datasheet shows that Bengali is one of the languages it is trained on [...]
PaLM is the forerunner to Bard, and PaLM work has been incorporated into Bard [...]
Maintaining the belief in "magic" properties, and amplifying it to millions serves Google's PR goals. Unfortunately, it's disinformation.

Click for full thread.
 
Musk has announced that he will create a competitor to ChatGPT which he may or may not call TruthGPT in a reference to how OpenAI permitted ChatGPT to lie.
 
If you were thinking of using ChatGPT to produce contents for a website there is bad news. Google can apparently detect content from an AI and will once it has detected it drop that page down its index.
 
If you were thinking of using ChatGPT to produce contents for a website there is bad news. Google can apparently detect content from an AI and will once it has detected it drop that page down its index.
I understand your concern. It is true that Google can detect content from an AI and will drop that page down its index if it is detected. However, there are a few things you can do to minimize the risk of this happening.

First, make sure that you are using ChatGPT responsibly. This means using it to generate content that is original, informative, and high-quality. Avoid using it to generate content that is spammy, low-quality, or plagiarized.

Second, use other tools to help you improve the quality of your content. For example, you can use a plagiarism checker to make sure that your content is original. You can also use a grammar checker to make sure that your content is grammatically correct.

Finally, be patient. It takes time to build up a good reputation with Google. If you are consistently generating high-quality content, Google will eventually reward you by ranking your pages higher in its search results.

Here are some additional tips for using ChatGPT to produce content for your website:

  • Use ChatGPT to generate outlines or ideas for your content, but don't rely on it to write the entire piece for you.
  • Proofread your content carefully before publishing it.
  • Use a variety of sources to gather information for your content.
  • Cite your sources properly.
  • Be honest and transparent about the fact that you used ChatGPT to generate your content.
By following these tips, you can minimize the risk of your content being flagged by Google as AI-generated.
 
Hi maomao looks like you have thought about this scenario.

At the moment my plan is to generate a page of contents by repeatedly questioning ChatGPT, I find I don't always get the detail I want on the first pass, then fact check at least anything that includes people mentioned by name and probably more, finally I plan to largely rewrite the AI generated content in my own words.

It is going to slow the process down but I think a Google penalty is to be avoided.

Mind you I wonder what Google will do if we generate pages using Bard?
 
AI like ChatGPT could have a significant impact on Google's main moneyspinner Ads on search. Now all we need to do is have a ChatGPT window open and ask it in English anything we fancy to get a response in conversational English.
 
Incidentally my boss is less bothered by a potential Google penalty because even a No 1 organic ranking is nearly off page 1 these days with all the other Ads and other content that Google pushes up in front of the organic results. He may have a point.
 
Open Source AI launched with a call out for people to use it, test it, and therefore improve it

review, its not currently very good but it is open source and worth supporting/improving for that reason:

 
Random thought I had today. If you trained an AI on someone's social media, WhatsApp (and maybe urban) data how well would you have to know a person to realise quickly it wasn't them?
 
Google is still happy to rank content that has been created with the help of AI. And it can't detect it automatically anyway.
 
Google is still happy to rank content that has been created with the help of AI. And it can't detect it automatically anyway.
And why and how would it be able to detect AI content?
And why (again) should any of us take online content as gospel?
And - more importantly perhaps - how can everyone else be alerted to the massively increased unreliability of online sources?
 
Random thought I had today. If you trained an AI on someone's social media, WhatsApp (and maybe urban) data how well would you have to know a person to realise quickly it wasn't them?
I don't think it would be that hard to tell tbh. Just with posters on here who you only know from here, you'd quickly spot something was off.
 
very impressive this - AI able to explain what is happening in complex images, not just what appears, but also the meaning and context. and this isnt even running on the full power model.

i can well imagine this being strapped into live video also, as opposed to one still image, with subsequent implications for robots

 
Regarding whether an AI poster could fool us, it partly depends on whether or not we are given the heads up that they might be AI.

They may well pass muster for quite a while if nobody suspected. But if you were to give me an AI version of regular posters on here and ask me 'is this real or AI', I'm confident I could down them with a handful of well-aimed questions/(depending on who they are) digs.
 
Regarding whether an AI poster could fool us, it partly depends on whether or not we are given the heads up that they might be AI.

They may well pass muster for quite a while if nobody suspected. But if you were to give me an AI version of regular posters on here and ask me 'is this real or AI', I'm confident I could down them with a handful of well-aimed questions/(depending on who they are) digs.
Yes, well, but how about a n00b? How could you tell a real human from a Chinese bot?
 
A n00b, no. ChatGPT is very good at sounding natural. We'd soon get bored with them, though. Beige.
Have you had a look at
 
Back
Top Bottom