My thoughts after publishing video. A few weeks ago, Anthropic released Claude 3.5 Sonnet with Artifacts. Having experimented with similar concepts in ChatGPT since summer 2023, I dove into an in-depth comparison with ChatGPT. Video came out long but I put work in to providing good chapters. As I was making chapters I was keeping score of results: ChatGPT 12, Claude 5 (with some draws). This score does not give Claude justice though. Sonnet 3.5 impresses with speed and quality. Claude gets good answers in 3-5 fewer responses than ChatGPT. Artifacts UX is addictive. However, ChatGPT's CustomGPTs and 3rd party API calls offer immense and broad utility. And that is where ChatGPT got most of the points. After video I succeeded at finish physics toy with ChatGPT You can compare results here: Claude: codepen.io/wonderwhy-er/full/ExzMxXJ ChatGPT: codepen.io/wonderwhy-er/pen/OJYKeoY In the end. Claude's limitations vs ChatGPT: - No internet search - Rewrites whole code instead of iterating - Due to that hits 4000 words or 800 lines limits in size of apps - Doesn't see errors in Artifacts - Limited 3rd party library support in Artifacts - No 3rd party API calls - Can't use uploaded files in Artifacts (e.g., pictures, full CSV files in charts) While Claude excels in speed and quality for text generation, ChatGPT's additional features make it more valuable for my needs. I'll likely cancel my Claude subscription this month. What are your thoughts on these AI models? Which features do you find most useful?
Thanks! I am glad it helps! I am in the process of making the next video. I did cancel Claude by now. It's cool but ChatGPT is more useful in many ways.
@@outofordermedia played with it a bit. Very impressive toy but it did not seem useful as you can't tell it what you need. It can't iterate on making a game or am I wrong?
@@EduardsRuzga WebSim can iterate on previous prompts. I made your favorite gravity balls simulator with a single prompt and then added controls for gravity and the ability to move the gravity balls around with a couple more prompts. I know that it's "packaged" like a game, but don't be fooled by this-it’s extremely powerful if you treat it as a single-file HTML5 prompt coding system. It uses Claude Sonnet 3.5 and somehow it's free for now, so I recommend you check it out while "free" lasts. In WebSim, you can see previous prompts by clicking on the simulated URL line. This shows you up to four previous prompts and lets you go back to that point in the project-at which point you can go back even further if you want. For professional prompt coding, I highly recommend Aider, which appears to be state of the art these days. It has the capability to understand your entire local git repository and execute changes across multiple files with a single prompt. It uses Claude 3.5 Sonnet and GPT-4 to return file diffs instead of entire files, which saves on speed and token cost.
@@EduardsRuzga WebSim can iterate on previous prompts. I made your favorite gravity balls simulator with a single prompt and then added controls for gravity and the ability to move the gravity balls around with a couple more prompts. I know that it's "packaged" like a game, but don't be fooled by this-it’s extremely powerful if you treat it as a single-file HTML5 prompt coding system. It uses Claude Sonnet 3.5 and somehow it's free for now. In WebSim, you can see previous prompts by clicking on the simulated URL line. This shows you up to four previous prompts and lets you go back to those points in the project-at which point you can go back even further if you want. For professional prompt coding, I highly recommend Aider AI, which appears to be state of the art these days. It has the capability to understand your entire local git repository and execute changes across multiple files with a single prompt. It uses Claude 3.5 Sonnet and GPT-4 to return file diffs instead of entire files, which saves on speed and token cost.
@@outofordermedia Share a link to websim physics toy, what was the prompt? As for Aider AI, do you use it often/ How much API dollars do you burn daily?
the guy is basically a baseball bat. even can't understand he writes.his english is really bad and writes everything like "i want gravity to be weaker" instead of "make gravity weaker". bro don't write everything like you are a spoiled 5 year old. Probably the AI is angry with you haha. Plus just say. "Make a slider that edits gravity"
Yeah, but that edge was never thinner :D Moment they add function calls to 3rd party APIs it may as well become non existent though. Well Dall-e 3 is still good too. But one can connect Clade to Stable Diffusion service then.
Project is basically useless. Do not do what I did and start uploading tons of docs. I was so excited to launch a new project I was doing with Claude Sonnet 3.5 when I heard about Projects. However, it did not take long before I hit a limitation. I was just getting started. I only uploaded 20 percent of my stuff into Claude. Waiting the next day does not help either. For me, this becomes useless. Unless they allow me to upload docs with no limitation and allow me to continue to chat all day, it is unusable. So I had to unsubscribe. I might go to Gemini as I can use google docs and upload with no issues. Just be careful as the stupid Claude limitation can completely break your project.
@@EduardsRuzga Each Project has a set limitation. Once you reach that limitation, you cannot continue with the project. You can however open up a new chat windows and start completely over again which completely defies the whole purpose of using a project. I hit my limitation within 2 days. My conclusion is Claude, as good it may seem is just a DEMO. It it not very practical for real use cases...at least for me I feel.
@@senju2024 I see, looks like Projects have 200k token limitation as they don't use retrieval like OpenAI Memory but dump whole your documents in to 200k context window they have. Yeah 200k even though large is not really that much. Its like 100 small articles or so. On level of an organisation it will be completely useless.
@@EduardsRuzga FYI - You can ask ChatGPT4o to create a markup file (DM) for any code, process, etc. which is very similar to Artifact window. Not only that, ChatGPT4o will provide a link to download the DM file it creates. Open that DM file and it your personal artifact with no dam restrictions. 🙂
![WE GOT ACCESS TO GPT-3! [Epic Special Edition]](http://i.ytimg.com/vi/iccd86vOz3w/mqdefault.jpg)
My thoughts after publishing video.
A few weeks ago, Anthropic released Claude 3.5 Sonnet with Artifacts.
Having experimented with similar concepts in ChatGPT since summer 2023, I dove into an in-depth comparison with ChatGPT.
Video came out long but I put work in to providing good chapters.
As I was making chapters I was keeping score of results:
ChatGPT 12, Claude 5 (with some draws).
This score does not give Claude justice though.
Sonnet 3.5 impresses with speed and quality.
Claude gets good answers in 3-5 fewer responses than ChatGPT.
Artifacts UX is addictive.
However, ChatGPT's CustomGPTs and 3rd party API calls offer immense and broad utility.
And that is where ChatGPT got most of the points.
After video I succeeded at finish physics toy with ChatGPT
You can compare results here:
Claude: codepen.io/wonderwhy-er/full/ExzMxXJ
ChatGPT: codepen.io/wonderwhy-er/pen/OJYKeoY
In the end. Claude's limitations vs ChatGPT:
- No internet search
- Rewrites whole code instead of iterating
- Due to that hits 4000 words or 800 lines limits in size of apps
- Doesn't see errors in Artifacts
- Limited 3rd party library support in Artifacts
- No 3rd party API calls
- Can't use uploaded files in Artifacts (e.g., pictures, full CSV files in charts)
While Claude excels in speed and quality for text generation, ChatGPT's additional features make it more valuable for my needs. I'll likely cancel my Claude subscription this month.
What are your thoughts on these AI models? Which features do you find most useful?
Great video, This video is enough to make decisions whether you want to continue gpt or use claude.. Thanks for efforts ❤
Thanks! I am glad it helps!
I am in the process of making the next video. I did cancel Claude by now. It's cool but ChatGPT is more useful in many ways.
Have you explored WebSim AI yet?
@@outofordermedia played with it a bit. Very impressive toy but it did not seem useful as you can't tell it what you need. It can't iterate on making a game or am I wrong?
@@EduardsRuzga WebSim can iterate on previous prompts. I made your favorite gravity balls simulator with a single prompt and then added controls for gravity and the ability to move the gravity balls around with a couple more prompts. I know that it's "packaged" like a game, but don't be fooled by this-it’s extremely powerful if you treat it as a single-file HTML5 prompt coding system. It uses Claude Sonnet 3.5 and somehow it's free for now, so I recommend you check it out while "free" lasts.
In WebSim, you can see previous prompts by clicking on the simulated URL line. This shows you up to four previous prompts and lets you go back to that point in the project-at which point you can go back even further if you want.
For professional prompt coding, I highly recommend Aider, which appears to be state of the art these days. It has the capability to understand your entire local git repository and execute changes across multiple files with a single prompt. It uses Claude 3.5 Sonnet and GPT-4 to return file diffs instead of entire files, which saves on speed and token cost.
@@EduardsRuzga WebSim can iterate on previous prompts. I made your favorite gravity balls simulator with a single prompt and then added controls for gravity and the ability to move the gravity balls around with a couple more prompts. I know that it's "packaged" like a game, but don't be fooled by this-it’s extremely powerful if you treat it as a single-file HTML5 prompt coding system. It uses Claude Sonnet 3.5 and somehow it's free for now.
In WebSim, you can see previous prompts by clicking on the simulated URL line. This shows you up to four previous prompts and lets you go back to those points in the project-at which point you can go back even further if you want.
For professional prompt coding, I highly recommend Aider AI, which appears to be state of the art these days. It has the capability to understand your entire local git repository and execute changes across multiple files with a single prompt. It uses Claude 3.5 Sonnet and GPT-4 to return file diffs instead of entire files, which saves on speed and token cost.
@@EduardsRuzga You can iterate
@@outofordermedia Share a link to websim physics toy, what was the prompt?
As for Aider AI, do you use it often/ How much API dollars do you burn daily?
❤
Make gravity weaker, make balls bigger mean nothing, you have to talk to AI in technical terms to get the best results.
@@edenassos I am interested in AI use cases for not technical people so when I test I intentionally start with non technical questions.
the guy is basically a baseball bat. even can't understand he writes.his english is really bad and writes everything like "i want gravity to be weaker" instead of "make gravity weaker". bro don't write everything like you are a spoiled 5 year old. Probably the AI is angry with you haha. Plus just say. "Make a slider that edits gravity"
Great video, very timely, thank you! For me, neither model is there yet but I agree that ChatGPT probably has the edge at the moment.
Yeah, but that edge was never thinner :D
Moment they add function calls to 3rd party APIs it may as well become non existent though.
Well Dall-e 3 is still good too.
But one can connect Clade to Stable Diffusion service then.
Project is basically useless. Do not do what I did and start uploading tons of docs. I was so excited to launch a new project I was doing with Claude Sonnet 3.5 when I heard about Projects. However, it did not take long before I hit a limitation. I was just getting started. I only uploaded 20 percent of my stuff into Claude. Waiting the next day does not help either. For me, this becomes useless. Unless they allow me to upload docs with no limitation and allow me to continue to chat all day, it is unusable. So I had to unsubscribe. I might go to Gemini as I can use google docs and upload with no issues. Just be careful as the stupid Claude limitation can completely break your project.
@@senju2024 by limitation you mean messages per hour or how much documents you can upload?
@@EduardsRuzga Each Project has a set limitation. Once you reach that limitation, you cannot continue with the project. You can however open up a new chat windows and start completely over again which completely defies the whole purpose of using a project. I hit my limitation within 2 days. My conclusion is Claude, as good it may seem is just a DEMO. It it not very practical for real use cases...at least for me I feel.
@@senju2024 I see, looks like Projects have 200k token limitation as they don't use retrieval like OpenAI Memory but dump whole your documents in to 200k context window they have.
Yeah 200k even though large is not really that much. Its like 100 small articles or so.
On level of an organisation it will be completely useless.
@@EduardsRuzga FYI - You can ask ChatGPT4o to create a markup file (DM) for any code, process, etc. which is very similar to Artifact window. Not only that, ChatGPT4o will provide a link to download the DM file it creates. Open that DM file and it your personal artifact with no dam restrictions. 🙂