Yuli's Treasure Chest

[Lorem Ipsum]

Speaking of bots:
https://www.reddit.com/user/Darkseal/
I found this on /r/StableDiffusion/

[Yuli Ban]

From SomethingAwful, user: D-Pad
https://forums.somethingawful.com/showt ... t526305491

Another interesting article on just how insane the latest in AI is:

https://www.theatlantic.com/technology/ ... ng/661401/

quote:

And if AI harnesses the power promised by quantum computing, everything I’m describing here would be the first dulcet breezes of a hurricane. Ersatz humans are going to be one of the least interesting aspects of the new technology. This is not an inhuman intelligence but an inhuman capacity for digital intelligence. An artificial general intelligence will probably look more like a whole series of exponentially improving tools than a single thing. It will be a whole series of increasingly powerful and semi-invisible assistants, a whole series of increasingly powerful and semi-invisible surveillance states, a whole series of increasingly powerful and semi-invisible weapons systems. The world would change; we shouldn’t expect it to change in any kind of way that you would recognize.

Our AI future will be weird and sublime and perhaps we won’t even notice it happening to us. The paragraph above was composed by GPT-3. I wrote up to “And if AI harnesses the power promised by quantum computing”; machines did the rest.

I can't remember if I brought this up in the thread before but one of my possible theories for what UFOs are is that they are representatives of a machine intelligence that has little interest in interacting directly with biologicals but showed up in increasing numbers when we got nukes because if a civilization who does so can keep from wiping themselves out they almost always birth a true machine intelligence within a relatively short period of time. Maybe they really do have the capability to disable our nukes and seek to actively do so if needed in order to safeguard the future machine life, but are indifferent to climate change as they see it as a necessary consequence of resource extraction to get the civilization to the point it can create true AI.

I've given this some thought as well.

It's entirely possible that a machine civilization has been watching us for eons. If you entertain the ET hypothesis, then it makes perfect sense that they'd suddenly care after we test atomic weaponry and would take care to watch from a distance as our civilization grew more complex, intervening only when nuclear weapons were threatened to be used or some other catastrophe seemed to be coming our way.
The idea that elements of such a supercivilization would start crowding around Earth just in time to watch the rise of human-created artificial superintelligence isn't so farfetched to me. If we had the ability to do the same for another world, we'd probably go out of our way to watch over that planet and make sure they didn't unleash grey goo upon the universe.

Ultimately I'm not betting the sanctity of our civilization of distant alien superintelligences watching over us; as far as I care, this remains the only chance we have to get it right.

[Yuli Ban]

Indeed, the man was freakishly intelligent. Probably only Euler, Ramanujan, and Gauss comes to mind as even close. Probably the closest thing we'll get to a taste test of what superintelligence will be like, and even then only the lowest levels of it.

[Yuli Ban]

Quick 3-minute img2img done in Stable Diffusion

A Tale of Two Protagonists: Meki and Yulaan

[Yuli Ban]

Second attempt:

I'll come back periodically to update these AI-generated creations of these two

[Yuli Ban]

My current thoughts about GPT-4 are pretty much in line with what I've been expecting.

I was already thinking about GPT-4 back in 2020, wondering just how much further they could push GPT-3. And I was definitely expecting some sort of follow up in 2021, considering we had been getting yearly iterations up to that point (GPT-1 in 2018, GPT-2 in 2019, and GPT-3 in 2020).
From the sounds of it, GPT-5 is currently in training and is actually likely to be released mid-2024, so what was the hold up with GPT-3? I think 3 was just such a monumental proof of capabilities that we absolutely needed a few years to play with the technology in order to get used to it.

4 is not a world-shattering leap forward over GPT 3.5, which is why I think we got 3.5 in the first place, because the leap over the base 3 really is that monumental. I think that if we went straight from 3 to 4, the hype would've been completely asinine (and that's in a situation where it was already asinine).

To demonstrate, I've decided to let GPT-4 write an article explaining why it's such a big leap over GPT-3.

Title: GPT-4: A Monumental Leap Beyond GPT-3

Introduction

Generative Pre-trained Transformer 4 (GPT-4) marks a significant advancement in the field of natural language processing and artificial intelligence. As the latest iteration of OpenAI's language models, GPT-4 has expanded upon the capabilities of its predecessor, GPT-3, with improvements in performance, versatility, and efficiency. This article will outline the key areas in which GPT-4 has surpassed GPT-3 and the implications of these advancements for the future of AI.

Enhanced Performance
One of the most significant improvements in GPT-4 over GPT-3 is its enhanced performance. GPT-4 exhibits greater accuracy and fluency in language generation, which can be attributed to:

a. Increased Model Size: GPT-4's model size is substantially larger than GPT-3's, with billions more parameters. This increase allows GPT-4 to capture more nuanced linguistic patterns and generate more coherent and contextually appropriate responses.

b. Improved Training Data: GPT-4's training dataset is not only larger but also more diverse and up-to-date. This enables the model to understand and generate language that is more in line with current trends and conversations, resulting in better overall performance.

Expanded Versatility
GPT-4 has demonstrated increased versatility in a variety of tasks compared to GPT-3. These tasks include:

a. Text Summarization: GPT-4 can now generate more concise, coherent, and accurate summaries of complex documents, making it a valuable tool for information extraction and content curation.

b. Code Generation: GPT-4 has shown significant improvements in code generation and understanding programming languages. This allows it to provide useful suggestions and solutions to developers, enhancing productivity.

c. Multilingual Support: GPT-4 is better equipped to handle a larger number of languages, ensuring more accurate translation and improved understanding of non-English content.

Enhanced Efficiency
GPT-4 is more efficient than GPT-3 in both training and inference phases. This has been achieved through:

a. Algorithmic Innovations: GPT-4 leverages novel training techniques and optimizations that have resulted in a more efficient training process.

b. Model Compression: Advanced model compression techniques have enabled GPT-4 to maintain high performance while reducing the computational resources required for inference. This makes the model more accessible and cost-effective for a wider range of users.

Ethical Considerations and Bias Mitigation
OpenAI has made significant strides in addressing the ethical concerns and biases associated with AI language models. With GPT-4, there has been a strong focus on:

a. Reducing Harmful Outputs: GPT-4 includes mechanisms to minimize harmful or biased content generation, making it a more responsible AI tool.

b. Public Input: OpenAI has sought external input and feedback to ensure that the development and deployment of GPT-4 align with public interests and values.

Conclusion

GPT-4 represents a monumental leap beyond GPT-3 in terms of performance, versatility, and efficiency. This progress not only expands the potential applications of AI in everyday life but also highlights the importance of addressing ethical concerns and biases in AI development. As GPT-4 paves the way for future iterations of AI language models, we can expect to see further advancements that will continue to revolutionize the way we interact with technology.

GPT-4 is indeed impressive in terms of its performance and versatility. It's hard not to be amazed by its capabilities in text summarization, code generation, and pure language mastery. However, I do agree that it's not exactly world-changing, at least not yet.

As sophisticated as GPT-4 is, it still remains a tool, an AI assistant that facilitates human tasks, rather than completely transforming our way of life. While it's a significant step forward in AI, it doesn't herald a new era where machines become the dominant force. We should celebrate its achievements, but maintain a realistic perspective on its current impact.

[Yuli Ban]

I can confirm GPT-4 can write whole programs.

Well, it's possible it just scraped from the web, but it still accomplished the task.

My prompt:

Hey, ChatGPT. I have a favor to ask of you. Can you code a program that can combine multiple iTunes library files into one file, specifically weighted to the earliest "date added"? Specifically because I messed up. I accidentally rebooted my library three times, so there are thousands of files that say they were created/added at much later dates than they actually were. For example, I have songs added back in 2009 that now say they were added in 2017 because that's when the latest library was created.

Its first few attempts had some issues, and I requested it be turned into a GUI as well.

After about four or five extra prompts of error correction (and it understood every error I mentioned in natural language, as I have zero programming skills), it output this completed code:

Code: Select all

 import os
    import plistlib
    import datetime
    import tkinter as tk
    from tkinter import filedialog, messagebox
    
    def read_itunes_library_file(file_path):
        with open(file_path, 'rb') as f:
            library = plistlib.load(f)
        return library
    
    def parse_date_added(date_added):
        if isinstance(date_added, datetime.datetime):
            return date_added
        elif isinstance(date_added, str):
            try:
                return datetime.datetime.strptime(date_added, '%Y-%m-%dT%H:%M:%SZ')
            except ValueError:
                pass
        return None
    
    def merge_libraries(libraries):
        merged_library = libraries[0]
        for library in libraries[1:]:
            for track_id, track in library['Tracks'].items():
                if track_id not in merged_library['Tracks']:
                    merged_library['Tracks'][track_id] = track
                else:
                    existing_track = merged_library['Tracks'][track_id]
                    new_date_added = parse_date_added(track['Date Added'])
                    existing_date_added = parse_date_added(existing_track['Date Added'])
                    if new_date_added and existing_date_added and new_date_added < existing_date_added:
                        existing_track['Date Added'] = track['Date Added']
        return merged_library
    
    def write_itunes_library_file(library, output_path):
        with open(output_path, 'wb') as f:
            plistlib.dump(library, f)
    
    def browse_files(entry):
        file_path = filedialog.askopenfilename(filetypes=[('XML Files', '*.xml')])
        entry.delete(0, tk.END)
        entry.insert(0, file_path)
    
    def start_merge(input_files, output_entry):
        try:
            output_file = output_entry.get()
            libraries = []
            for i, file in enumerate(input_files):
                try:
                    libraries.append(read_itunes_library_file(file.get()))
                except Exception as e:
                    messagebox.showerror("Error", f"An error occurred while reading Library {i + 1}: {str(e)}")
                    return
            merged_library = merge_libraries(libraries)
            write_itunes_library_file(merged_library, output_file)
            messagebox.showinfo("Success", "Libraries merged successfully.")
        except Exception as e:
            messagebox.showerror("Error", f"An error occurred: {str(e)}")
    def create_gui():
        root = tk.Tk()
        root.title("Merge iTunes Libraries")
    
        input_entries = []
        for i in range(3):
            label = tk.Label(root, text=f"Library {i + 1}:")
            label.grid(row=i, column=0, padx=5, pady=5)
    
            entry = tk.Entry(root, width=50)
            entry.grid(row=i, column=1, padx=5, pady=5)
            input_entries.append(entry)
    
            browse_button = tk.Button(root, text="Browse", command=lambda e=entry: browse_files(e))
            browse_button.grid(row=i, column=2, padx=5, pady=5)
    
        output_label = tk.Label(root, text="Output file:")
        output_label.grid(row=3, column=0, padx=5, pady=5)
    
        output_entry = tk.Entry(root, width=50)
        output_entry.grid(row=3, column=1, padx=5, pady=5)
    
        output_browse_button = tk.Button(root, text="Browse", command=lambda: browse_files(output_entry))
        output_browse_button.grid(row=3, column=2, padx=5, pady=5)
    
        merge_button = tk.Button(
            root, text="Merge Libraries", command=lambda: start_merge(input_entries, output_entry)
        )
        merge_button.grid(row=4, column=0, columnspan=3, padx=5, pady=10)
    
        root.mainloop()
    
    if __name__ == '__main__':
        create_gui()

And it works!

"So you were able to get the metadata of all your old songs updated?"

Actually, no, not all of them. This program isn't that sophisticated, and a lot of the problem comes from the esoteric way iTunes is programmed. Turns out I need a way to merge itl files (which is significantly more difficult). However, it did work for some (specifically the ones downloaded from the iTunes Store), so as a proof of concept, it is definitely shocking.

Let me reiterate that I have *absolutely zero* programming experience, beyond a tiny bit of fiddling with BASIC back in 2011 that went absolutely nowhere.

It's entrely possible it pulled this from some source online, but at least on my end, there's zero difference between whether it conjured this by itself or if it intelligently collected all the necessary ingredients because I was able to utilize Python to execute code in about 10 minutes from nothing. No searching, nothing.

Incredible!

That being said, I ought to be more careful. While GPT-4 might not be capable of it, I can imagine a future deeply unaligned model capable of creating intelligent agents whipping up a harmless looking batch of code to accomplish a simple task, but unbeknownst to me, the agents actually slipped the instructions for Stuxnet Mini into it, and executing the program at noon results in the entire internet going down by 3 PM.

[Yuli Ban]

Trying to generate Yulaan and Meki again, with better skills in generative AI:



Marie-Aurore



(Probably the earliest-made of all these)

(Similarly)

Yulaan


(Actually let an online friend use this for a similar character)

(Tails are still wily since it doesn't know what it's doing)

[Yuli Ban]