Claude 3 Opus vs GPT-4 for Coding - My Experience

00:00 Introduction
00:46 Test Case 1 - Prompt #1: GPT-4 Misses the Point of Refactoring
04:35 Prompt #2: Testing Claude with Multiple Instructions
07:27 Prompt #3: Claude Maintains Code Conventions
08:20 Prompt #4: Asking for Advanced Refactoring
09:23 Test Case 2 - Prompt #1: Claude Demonstrates Superior Conciseness
13:27 Prompt #2: Claude Displays Higher Accuracy
14:51 Prompt #3: Claude Outperforms Once Again
18:23 Final Verdict: Which AI Model is Superior for Coding?

[yonatan]

"Claude is more helpful and correct" that's the consensus from the web.

[lancemarchetti]

Claude Sonnet helped me fix the issue with my code which I've been struggling with for a few months. Other coding LLMs came close.. but no win.

[arunachalpradesh]

claude 3 codes are also very professional, some codes are even optimized like its written by real programmer, that was quite surprising. as chat gpt and google gemini mostly just copy paste from trained websites.

[adamwood]

I'm getting better results with Opus optimizing my SQL queries than GPT-4. Though it will frequently suggest removing DISTINCT from queries when it shouldn't. Asking it to double check its work generally catches these mistakes. Sometimes it does better when asking it multiple times to make minor improvements to the code over and over again. I also tend to get better results when I start my prompt with, "I'm told you're expert in SQL, can you assist me with ...?"

[smnomad]

That was the best Claude vs GPT comparison on YouTube. Your content is underrated! I will subscribe and keep an eye on comparisons you will be making as newer versions of these LLMs keep coming out.

[iCrimzon]

After the Claude 3 Opus news, i waited a week for a comparison between the 2 and have been using Sonnet while waiting, Sonnet is quite good as well, was trying to write a JS script but i never write frontend code so when my backend and frontend werent working well together, Sonnet managed to solve the issue where GPT4 just keep giving me the same troubleshooting advice and rewriting the same script and backend code, wont use GPT4 as much anymore but glad to see my switch to Opus is worth it even if its just a slight upgrade

[avi]

GPT-4 and Opus are both very good at code. They seem to excel at different things though. I've found Opus is very good at conceptualizing and GPT-4 is better at implementation details (like your autoDownloadIfMissing example). GPT-4 is better at analyzing error messages and fxing them, but Opus is better at analyzing code and suggesting pertinent refactors. A couple times I've been using Opus and thought, oh, I hadn't thought of that, that's actually a really good idea, whereas that doesn't really happen with GPT-4. It's less "creative" one might say. Essentially, they seem to be complimentary tools. My plan is to use Rivet to build a pipeline flow where GPT-4 and Opus argue over code until they agree on a solution together. And I'll probably use it once as I spend a dollar generating 50 lines of code.

[apache]

Opus is definitely better for coding but GPT-4 is still better for casual use (like random questions) because Claude safety trash

[cristianpereira]

🎯 Key Takeaways for quick navigation:

00:00 *🌟 Experiencia con Cloud 3 Opus y comparación con GPT-4 para programación*
- Experiencia personal utilizando Cloud 3 Opus y comparación con GPT-4.
04:38 *💡 Comparación de refactorización de código entre Cloud 3 y GPT-4*
- Comparación de respuestas de Cloud 3 y GPT-4 al refactorizar código.
- Evaluación de la legibilidad y la lógica de las soluciones propuestas por cada modelo.
09:38 *🔄 Revisión de una solución alternativa propuesta*
- Propuesta de una solución alternativa por parte del usuario.
- Comparación de la respuesta de Cloud 3 y GPT-4 ante la nueva propuesta y sus implicaciones.
15:06 *💬 Modificación de lógica y comprobación de cambios*
- Evaluación de cómo Cloud 3 y GPT-4 manejan los cambios propuestos por el usuario.
- Análisis de las modificaciones realizadas y sus efectos en el código final.
18:33 *🏁 Conclusión y recomendación*
- Conclusión basada en la experiencia de uso de Cloud 3 y GPT-4.
- Recomendación sobre cómo utilizar ambos modelos de manera complementaria en el proceso de desarrollo de software.

Made with HARPA AI

[risehit]

Two weeks have passed, do you have broader conclusions? I am thinking of acquiring Claude to work hard, but gpt 4 is better?

[davidconanan]

which is better in generating codes in prompt? I am mostly using it in trying to create cloudformation environment and creating kubernetes environment. thanks subscribing to you.

[preend]

I have no coding experience, and i run into a problem with gpt4 and i am loosing my mind. Lol i hope Claude can fix code.

[fitybux]

I am also with Team Claude on code. GPT-4 took a poo poo on your code. 😂 I have also used GPT-4 for months prior. OpenAI just lost my monthly subscription. (Don't forget to export your ChatGPT conversations.)

[manuelgonzalez]

Interesante probare claude 3 opus pagaré los 20 dólares jaja

[bippytime]

Using GPT api via vscode or cursor is extremely bad compared to even just using GPT+. I guarantee if you redid this test, while claude would have it's moments, you would have vastly different results. I will say though. Claude does just "get it" with less instruction, while GPT seems to need more direction. But overall GPT will always be right more than claude. most of these tests have better results with copilot using regular VSCODE, assuming your codebase is cached.
I can appreciate this video and your efforts, but I do feel like this wasn't a great comparison as cursor doesn't work great with a direct openai API key. I bet copilot would have done better than openai's api integration with cursor there. Also cursor would do better with a subscription than it would with openapi key directly as it doesn't reference the codebase using the same methods. also if you do tests on the results from cursor, you will notice that it doesn't actually even reference the entire cached codebase when it's using it for suggestions, even when you ctrl enter etc to reference the entire codebase, or even a singular file at that. I would like to see your comparison under the same conditions for both. I assure you cursor isn't great for this comparison whatsoever in the context in which you are using it.

[covalentbond]

Can u use AI to change ur accent like some indians do

[justrobin]

The AI voice was too difficult to listen to.