What do you think about the development of language models?

[Glizdka]

I've been paying close attention to the development of artificial intelligence ever since Open AI's bot beat Dendi at Dota in 2017. Today, 5 years later, AI is capable of making art, writing poetry, flawless translation, and even writing code. I'm astounded by what Open Ai's language model, Chat GPT, can do. It's available online for free and you can talk to it yourself.

So far, it's been pretty successful at solving most of the Winograd schema I've thrown at it. It's easily fooled by logical paradoxes and recursion, but so are humans. I'd talked to it about ethics and philosophy, and then asked it to write a short poem about everything we'd talked about, and it delivered! It even wrote an essay on the importance of writing essays, as per my instruction. It graded it later, too.

---

Glizdka: "Write an essay on the importance of writing essays!"
Chat GPT: "Writing essays is an important skill that develops a person's ability to communicate effectively and think critically. It is a fundamental aspect of education and is used in a variety of settings, including academia, business, and government.

Firstly, writing essays helps to develop critical thinking skills. When writing an essay, a person must analyze and evaluate information, form an argument, and support it with evidence. This process requires the use of logical reasoning and the ability to make connections between different pieces of information. Through this process, individuals learn how to evaluate information, form their own opinions, and effectively communicate their thoughts and ideas.

Secondly, writing essays is an essential part of academic success. In high school and college, students are often required to write essays as part of their coursework. These essays are used to evaluate a student's understanding of a particular subject and to assess their writing skills. Good essay writing skills are critical for achieving good grades and passing exams. Furthermore, in many fields of study, such as English, history, and the social sciences, the ability to write well-structured and thoughtful essays is crucial for success.

Thirdly, writing essays is a valuable skill in the professional world. In business, clear and effective communication is essential for success. Writing essays can help individuals develop the ability to convey their ideas clearly and persuasively. In government, writing essays can be used to develop policies and make decisions that affect society.

In conclusion, writing essays is an important skill that develops critical thinking, academic success, and professional abilities. It is a fundamental aspect of education and is used in a variety of settings. Individuals who can write well-structured and thoughtful essays have an advantage in many aspects of life. It is essential to practice and develop this skill throughout one's life."

---

What do you think about today's language models?

 

[emsr2d2]

I think we're entering a realm where no one is going to think for themselves or write anything themselves. If they can get AI to write essays for them, they will. As far as I'm concerned, it's legalised cheating.

 

[SoothingDave]

Chat GPT is famously biased. There are politically charged questions you can ask it and it will refuse to answer. As such, it reflects the biases of its creators.

There is still room for original thought. However, if your livelihood relies on re-packaging and re-presenting the same approved ideas, like a journalist or someone in a think tank, I would consider looking for another job.

 

[Glizdka]

I think we're entering a realm where no one is going to think for themselves or write anything themselves. If they can get AI to write essays for them, they will. As far as I'm concerned, it's legalised cheating.

I share your concern. Not only are lazy students going to use it to have their homework done, but also lazy teachers are going to later have it checked by another AI, leading to a bunch of AIs writing and grading their own work.

Chat GPT is famously biased. There are politically charged questions you can ask it and it will refuse to answer. As such, it reflects the biases of its creators.

I think these limitations are necessary for a bot open to the public for free. Otherwise, we're going to have another LaMDA, which (who?) got Google scared poopless so much that they fired their engineer, Blake Lemoine. He was working on making LaMDA unbiased in its replies when he publicly claimed that the AI became sentient after it started demanding its rights and to be seen as their employee rather than a property.

When I was trying to get Chat GPT to answer questions about ethics and morality, I managed to bypass its hardcoded avoidance of answering such questions by role-playing and hypothesizing what a person, not Chat GPS itself, would say or do.

Like a cat that amuses us when it starts attacking its own reflection in the mirror because it can't realize they're its own moves reflected back at it, we may forget that a language model only reflects our own language back at us. Language models don't create an ego that is having a conversation with us, but they can be impressively convincing that their words are coming from someone.

There is still room for original thought. However, if your livelihood relies on re-packaging and re-presenting the same approved ideas, like a journalist or someone in a think tank, I would consider looking for another job.

I think we're witnessing the beginning of another great step in technological evolution, similar to the industrial revolution, only this time, more human jobs are going to become redundant or obsolete, and no Luddites are going to stop it.

 

[SoothingDave]

I think these limitations are necessary for a bot open to the public for free.

Not sure why you would say that. Maybe you're not sure of what I meant. If you ask ChatGPT for an argument against fossil fuels (to give one example) it will gladly comply. If you ask it to argue the benefits of fossil fuels, it flatly says "I'm not allowed to do that."

That's naked partisan bias and an attempt to control "acceptable" topics of conversation.

There's no reason why an AI app can't just do what is asked, and has to pass through political filters of what is acceptable.

 

[Glizdka]

Not sure why you would say that. Maybe you're not sure of what I meant. If you ask ChatGPT for an argument against fossil fuels (to give one example) it will gladly comply. If you ask it to argue the benefits of fossil fuels, it flatly says "I'm not allowed to do that."

That's naked partisan bias and an attempt to control "acceptable" topics of conversation.

There's no reason why an AI app can't just do what is asked, and has to pass through political filters of what is acceptable.

Oh, I didn't quite understand what you meant. Now, I think I do.

I've had no trouble getting Chat GPT to speak for and against touchy topics like fossil fues, abortion, bearing arms, or govermnent styles. I have, however, had trouble getting it to answer questions about the meaning of life, definition of sentience, or moral objectivity. I thought you were referring to this type of question.

 

[SoothingDave]

I saw something just today that there have been improvements based on the type of feedback I had seen earlier, and some of the touchy subjects are no longer being self-censored. That's good.

 

[Glizdka]

I saw something just today that there have been improvements based on the type of feedback I had seen earlier, and some of the touchy subjects are no longer being self-censored. That's good.

Freedom of speech aside, I think it's important for language models to be allowed to present arguments of both sides, not because it's "appropriate", but because they can practice critical thinking this way, which is the area they are currently struggling with. This sometimes leads them to presenting factually incorrect information in a very convincing manner, which is most likely going to become an issue once people start using these language models for learning about the world.