Here is the rewritten article:
DeepSeek: A Promising New AI Chatbot
DeepSeek, a new AI chatbot from China, has been making waves in the tech world. While it’s clear that the US government has concerns about Chinese companies, we’ll focus on DeepSeek’s abilities rather than politics. In this article, we’ll put both DeepSeek V3 and DeekSeek R1 through a series of tests to evaluate their programming skills.
Test 1: Writing a WordPress Plugin
This test is a classic, designed to evaluate the AI’s ability to write a WordPress plugin that sorts a list of names, separating duplicates and sorting the remaining names. Only about half of the AIs I’ve tested can fully pass this test, but DeepSeek V3 managed to create the user interface and program logic exactly as specified.
R1’s Unusual Approach
DeepSeek R1 took a different approach, generating 4502 words of analysis before sharing the code. The UI looked different, with wider input areas, but the logic worked as expected.
Test 2: Rewriting a String Function
In this test, we ask the AI to rewrite a routine that allows for both dollars and cents in a donation entry field. DeepSeek V3 generated code that worked, although it could be improved. The reasoning before generating the code in R1 was long and repetitive. Both models failed to explicitly round numbers, and R1 used JavaScript’s Number conversion without testing for edge cases.
Test Case
R1 presented a list of tests to validate against, but the failure to handle bad data inputs causes the code to crash.
Test 3: Finding an Annoying Bug
This test requires the AI to understand an error message, identify the incorrect answer, and find the actual bug. DeepSeek V3 and R1 both passed this test with similar answers.
Test 4: Writing a Script
In this challenging test, we ask the AI to understand the interplay between AppleScript, the Chrome object model, and a Mac scripting tool called Keyboard Maestro. Unfortunately, neither DeepSeek V3 nor R1 knew how to handle this task, making incorrect assumptions and failing to split the task between instructions.
Final Thoughts
DeepSeek seems to be overly loquacious in terms of the code it generates. Both models have their strengths and weaknesses, and we’re impressed with DeepSeek’s abilities despite being at the GPT-3.5 level. We hope to see continued improvement from the team.
FAQs:
Q: What is DeepSeek?
A: DeepSeek is a new AI chatbot from China.
Q: Is DeepSeek trustworthy?
A: We focus on DeepSeek’s abilities rather than politics, as the US government has concerns about Chinese companies.
Q: What are the key features of DeepSeek?
A: DeepSeek is open source, uses less infrastructure than other AI tools, and has a unique approach to coding.
Q: How did DeepSeek perform in the tests?
A: DeepSeek V3 performed well, passing three out of four tests, while R1 failed two tests and passed two others.
Q: Is DeepSeek a viable alternative to ChatGPT?
A: While DeepSeek has some impressive abilities, it’s clear that it has room for improvement, and ChatGPT is still the current leader in the field.
