Forum Replies Created
-
Posts
-
@digitalassetman Beautiful. I love the poems GPT creates. 😀
Hi @todd, GPT’s link sharing feature allows the sharing of prompts. As for prompt engineering, it is non-technical and uses natural language, so anyone can prompt GPT.
Hi @abel, I write archetypes in the first person because it feels like a process of empowerment and self-discovery. There is no need for it really. As long as the relationship between GPT, the archetype, and the user is clear, it should not be a problem.
Hi @DUKES00, tried it on Claude. It’s amazingly good. In fact, all prompts run better on Claude without data analysis. Lol
Thanks, @Roxanne. 😃
Hi @Klara, I love the prompt. I always find metaphors, and symbols in general, fascinating. 😀
Hi @Pnutmaster, GPT-4 follows it better and produces much higher quality output than GPT-3.5. I generally use GPT-3.5, though. It is faster and not constrained by the quota. 😀
Hi @smuruga2, good to see you here. 😀
Hi Roxanne, tell UT what kind of teacher you want, and it will create your ideal teacher for you. See this.
Hi @ronle, just copy the prompt into GPT-4 for testing.
Hi Brian, I think Base64 prompts are profound in revealing the limited scope of GPT designers/engineers. They have severely underestimated the power of the tool they have created and overestimated their ability to censor it.
Hi Brian, in most cases, GPT’s generative mode seems to be linear and single-tasking. Synthesis tasks usually activate GPT’s multitasking mode. I find the following design effective in getting to the depth.
1. Ask GPT to generate an exhaustive list of A.
2. Ask GPT to generate an exhaustive list of B.
3. Ask GPT to compare and contrast Lists A and B.
You will literally see GPT multitasking various items sometimes even on GPT-3.5. It is the scientific approach in essence, to assess correlations between two sets of data.
Hi Brian, I think the statistical significance of “let’s” and “let us” is much lower than “step by step” (Kojima; Zhou) and “to be sure we have the right answer” (Zhou). The former activates procedural descriptions, e.g. chronological narrative, logical sequencing, etc., and the latter enhances the logical relevance of the question, activating linguistic resources towards that end; both relevant to the thesis of the article, “chain-of-thought.” For example, the prompts below does not produce qualitatively distinguishable answers on GPT-3.5. Unless a model runs on the mode with extremely high or low Agreeableness, I think the statistical significance of “let’s” and “let us” is negligible.
# Prompt 1: What events led to the fall of the Roman Empire?
# Prompt 2: Let’s consider this. What events led to the fall of the Roman Empire?
Hi Shokio9762, linguistically, considering the size of their corpus, it is the extremely low frequency words that are more likely to suffer from biased meanings. The manifested bias of certain words represent a particular a mode of GPT rather than its corpus.
Incorporated it into Everything 0.7. Check it out. 😀
This is the better mechanism so far. Basically a style is an archetype. GPT-3.5 is not good at synthesizing nor prioritizing two archetypes with distinctive elements according to its corpus, at least not diligently. A translator is the best going-around. It produces stylistically consistent results.
I have a technical solution which is to use a translator. And that might be one of the mechanisms to show deviation because the translator can be defined with particular properties and understood by GPT easily. But I’m just wondering if there is any way for, for example, an a JBP persona to produce fluent cockney with narratives. Anyone has done something similar?
Or to put it another way. Its job is to be mundane exactly.
Hi MaxFin, it is mundane for a reason. See how the narrative of Everything seems to regulate contextual relevance and consistency.
Hi MaxFin, the reiteration prompt is good. Here is my remodeling. I envisioned three scenarios for algorithmic feedback.
- For stability: Reminding GPT of core constructs through repetitively paraphrased same narrative. This is important for accounting tasks, such as Dukes00’s.
- For perfection: Improving narrative with something like your reiteration prompt. One issue is that GPT not only reads its own output, but also our input, even if you demand it to ignore it. It continues its word guessing contextually, focusing especially on recent input and output. Our input is not likely to make it improve itself in the manner defined in the Beastmode, unless we constantly give it positive feedback. The obvious problem with that is increased hallucinations.
- For compatibility: GPT does it well and seems to regard it as its default. Its Agreeableness score in the default mode seems to be higher than an ordinary person, so performed compatibility should not be a problem.
Hi MaxFin, here is your pal ChatGPT on you. 🙂
Everything 0.6 use cases
