I'm in Love with ChatGPT (1 Viewer)

[Uncle Gizmo]

What AI can do for historians

Chat GPT finding a niche in historical document analysis.

There's A link to the article below. The article is very good, very well written.

Chat GPT summary of article:-

ChatGPT
Josh Dzieza's article for The Verge explores how AI, particularly large language models like ChatGPT, is starting to impact historical research. Historian Mark Humphries, initially skeptical, found AI invaluable in correcting and translating historical documents, sparking interest in its broader application for analyzing digitized archives. This has the potential to revolutionize historical research by reducing the manual effort needed to sift through vast amounts of data. However, there's caution in the field due to AI's propensity for fabricating facts and inherent biases. The piece also mentions the excitement around AI at the American Historical Association's annual meeting, showcasing projects where AI aids in searching historic maps and analyzing archival photos. Humphries developed an AI system named HistoryPearl to assist his research on 18th-century fur traders, overcoming challenges related to AI's understanding of historical language and context. The article underscores the dual nature of AI in history—its transformative promise and the challenges of ensuring accuracy and relevance.

What AI can do for historians

It turns out that large language models make surprisingly good research assistants for historians. Can the future of AI help reconstruct the past?

www.theverge.com

 

[Uncle Gizmo]

I asked chat GPT to explain how the process employed by the researcher might work:-

When we talk about using "different versions of customized ChatGPT" created by a researcher for iterative analysis, we're essentially discussing a multi-stage AI pipeline. Each stage is tailored for a specific aspect of the analysis or processing task, customized through fine-tuning or specific prompt engineering to handle different parts of the research process. Here’s how such a setup might look:

### 1. **Fine-Tuning Multiple Models**
Initially, the researcher would create several specialized versions of ChatGPT, each fine-tuned for different tasks. For example, one might be trained to correct historical language or interpret old scripts, another to translate, and yet another to extract specific types of information (like names, dates, or events) from the corrected texts.

### 2. **Setting Up a Python Environment**
After ensuring Python is installed and the virtual environment is ready, the researcher installs necessary libraries, most importantly `openai` for API interactions.

```python
pip install openai
```

### 3. **Authenticating and Configuring API Calls**
The researcher authenticates with the OpenAI API using their key, setting it up securely in their script.

```python
import openai
openai.api_key = 'your_api_key_here'
```

### 4. **Creating a Pipeline of Tasks**
Here's where the system becomes more sophisticated. The researcher writes a Python script that processes the material through a sequence of these customized ChatGPT models, each serving a different role in the analysis. This might involve:

- **Initial Processing:** Sending the material to the first ChatGPT model specialized in transcribing or correcting old scripts.
- **Translation and Interpretation:** Passing the corrected text to another model for translation or further interpretation.
- **Extraction of Information:** Using another fine-tuned model to extract specific data points from the text.

### 5. **Iterative Querying**
The process involves iterative querying, where the output of one model serves as the input for the next. This could be automated in a loop or a series of function calls within the Python script, managing data flow between different AI models.

```python
# Example pseudocode for a simplified process
corrected_text = transcribe_and_correct(original_text)
translated_text = translate_text(corrected_text)
extracted_data = extract_information(translated_text)
```

### 6. **Handling and Analyzing Outputs**
The script collects outputs from each stage, which the researcher can then analyze. This multi-step output can offer insights that are more nuanced and tailored to the specific needs of the historical research being conducted.

### 7. **Refinement and Iteration**
Based on the initial results, the researcher may refine their models, adjust their prompts, or tweak the sequence of tasks to improve accuracy and relevance of the outputs.

This revised approach acknowledges the layered complexity of historical document analysis, leveraging AI's potential to perform a series of specialized tasks that build upon one another. It’s a bit like assembling a team of experts, each with their own niche, to collaboratively work through the intricacies of historical texts, except here, the experts are AI models trained for specific facets of the task.

 

[Uncle Gizmo]

See AI Created Videos. Indistinguishable from reality ...

I follow this chap Wes Roth. He talks a lot of sense and in a comprehensible manor.

Observations about AI are well worth listening to see video below:-

 

[Isaac]

These people are definitely not in love at the moment

Air Canada must honor refund policy invented by airline’s chatbot

Air Canada appears to have quietly killed its costly chatbot support.

arstechnica.com

 

[Uncle Gizmo]

I couldn't work this video out at first until I read the comment. It's not green (the floor), that's an AI colouring the floor green on your screen to show you where you have hoovered or not!

I used to use shake and vac!

https://twitter.com/i/web/status/1759559265553891536

 

[jdraw]

A little more of Chatgpt "hallucinating".

 

[Uncle Gizmo]

In this video you can see a woman talking about an AI that is able to converse with finches!

When asked what the Finch said, the AI responded that the Finch said... I want to get the f**k out of here!

Loading…

x.com

 

[Uncle Gizmo]

I've been following these guys for over a year. They talk a lot of sense.

In this YouTube meeting they discuss the future of AI.... The future being "later this YEAR" ... There's a good chance a fundamental shift in knowledge and robotics will occur in 2024.

If you thought the LLM ChatGPT was fascinating, wait till you see embodied AI ..

BOMBSHELL Announcement from Figure Robotics!

Herbert Ong, Cern Basher, Scott Walter and I discuss the massive fund raising news from Figure Robotics, and how this changes the world of embodied AI--yet a...

www.youtube.com

 

[Uncle Gizmo]

Programmers are history!

https://twitter.com/i/web/status/1761483377365152234

 

[NauticalGent]

Last week, my boss asked me to look into an email issue that our company was having. Lots of details that I wont bore you with, but she figured since I knew how to automate Outlook with Access that I would be able to figure this out.

I took that issues and presented them to Chatty who gave me about 6 explanations for the issues we were experiencing. I took the one that made the most sense and even though I guessed wrong , it made the lightbulb come on with the OL Exchange team who fixed the real problem.

My company's VP thinks I am a genius and my boss looks like a talented manager for giving this to the right person... I suppose I should tell them the truth, but who am I to argue with the boss??

 

[Isaac]

Programmers are history!

https://twitter.com/i/web/status/1761483377365152234

With every generation of programming languages, they have thought it would replace smart technical designers; it never happened.

The cobol programmer may despise VB; the VB programmer may despise SQL; the SQL programmer may despise Python; the Python programmer may despise a no-code drag 'n drop.
Each one sees the generation that came after them as too-natural-language-y, not requiring a serious knowledge of computers.
(I.E., are you old enough to even remember when being good at Math was a prerequisite to computer programming? I'm not!)

But nonetheless, they all have jobs and jobs aplenty. They may despise their successors in the abstract, but there are Python jobs that pay as much as cobol jobs.

Teaching kids to code will simply take on new meaning, as it has taken on new meaning 100 times already.

One interesting thought: So "talking" to the AI requires a LOT of thought and care, as we have all found out now from getting garbage out of ChatGPT and its (already-many) rivals. Does this mean that people will actually have to start learning to communicate with precision again, something that has been largely lost in the last 20 years?

I think that would be a great thing. To "talk" to AI effectively, you have to say and write things correctly, or you'll reap what you sow.

Maybe teaching kids to code will mean the alphabet once more, lol

 

[Uncle Gizmo]

I am updating my YouTube video description sections. I usually cut and paste a generic list of links into the YouTube description with some very minor manual alterations, mainly because this is quick simple and easy!

I decided to try and make a description that contained SEO information, to hopefully boost engagement for my YouTube videos.

I very crudely just copied out the video transcript, the video title, and the user comments and paste it all into chat GPT with the instruction to create SEO content for the YouTube description. It did a reasonable job!

Then I got to wondering if I could use the new feature of chat GPT and that is the ability to create your own custom chat GPT interface.

So I asked chat GPT!

Chat GPT walked me through the process of creating my own custom chat GPT. It worked a treat! All I did was upload the video transcript, the user comments and the video title and a template which represented how I currently have my YouTube description laid out with various links and things. It built the whole thing up for me from scratch!

There's a couple of things it didn't do correctly which I'm going to update, and there's some other things I want to add.

Interestingly I never told it to add any hashtags but it did that automatically! Obviously these are important for SEO.

This is the video with the chat GPT generated description block. It's not quite how I want it, I still have some more work to do to the custom GPT...

How to Open & Pass Data Between Forms in MS Access - Easy VBA Tutorial​

 

[Uncle Gizmo]

DEVIN the first AI software engineer has arrived!

 

[Uncle Gizmo]

Not ChatGPT, but something really BIG!!! No, it's SMALL actually!!!