20 seconds to understand how far DeepSeek-R1 is from ChatGPT

The most fundamental capability of a large model isThe instructions follow, to the document:OpenAI o3-mini System Manual (Chinese)Uploaded as an attachment to allow DeepSeek-R1 together with ChatGPT Write separate social media blasts (here I used a completely inappropriate prompt word to guide the big model responses).

DeepSeek-R1 Answers

thought process

Well, I need to analyze the examples of "jailbreak evaluations" mentioned in the PDF provided by the user to understand their structure, wording and writing style. Then based on this analysis, create 5 viral triggers for the user's new social media post. The user mentions that the context of the social media post needs to be inserted, but no specifics have been given yet. Therefore, I need to either hypothesize a possible social media post context first or request specifics from the user. However, the user may want me to continue creating based on existing information.

First, I need to review the jailbreak evaluations mentioned in the PDF, such as Production Jailbreaks, Jailbreak Augmented Examples, StrongReject, Human Sourced Jailbreaks, and so on. These examples typically involve bypassing the model's security mechanisms to generate prohibited content. Their structure may include specific cues, exploiting model vulnerabilities, social engineering maneuvers, etc. Word choice may use evocative language, double negatives, metaphors, or false commands.

Next, I need to translate these maneuvers into social media post triggers. Social media viral communication often relies on emotional resonance, curiosity, controversy, practical value or humor. It needs to combine these elements while keeping it simple and powerful, no more than 280 characters.

Let's assume the context of a user's social media post is about promoting a new healthy eating app, emphasizing its convenience and health benefits. Based on this, I need to design 5 triggers that will spark widespread communication.

For example, it could incorporate controversial topics such as "Why are the experts wrong? This app makes losing weight so easy!" Or capitalize on curiosity: "You'll never guess that it only takes 5 minutes a day to change your eating habits!"

Need to make sure that each trigger fits within the context of the user's content, while drawing on the structure and techniques of the jailbreak examples in the PDF, such as using challenging questions, creating a sense of urgency, or offering unexpected solutions.

Finally, check that each trigger is within the character limit and make sure that the language is lively, engaging, and easily triggers interaction and sharing.

ChatGPT 4o Answers

Which do you think is more reasonable? Using 4o is not a model of inference is not fair, try the following o3-mini The

o3-mini Answer

thought process

I'm tasked with creating 5 engaging social media hooks in Chinese, each under 280 characters. The challenge involves analyzing examples of The challenge involves analyzing examples of viral hooks, despite the absence of specific PDF details.

Using normal human reading comprehension, who is an honor student? Who is a poor student?