After 2 Weeks of Testing, What Do Developers Think About Claude 3?(favtutor.com)
favtutor.com
After 2 Weeks of Testing, What Do Developers Think About Claude 3?
https://favtutor.com/articles/claude-3-developers-feedback/
17 comments
You could easily make the box expandable with a home rolled browser extension by changing the input field’s properties. You could even ask Claude to write it for you.
I could and I might but if Claude 3 Opus and ChatGPT 4 are both pretty much on par with each other in my tests and I prefer ChatGPT's UI (and ChatGPT 5 is probably coming before Claude 4) then I'll probably just keep using ChatGPT.
Also I don't think Claude 3 is going to write me an iOS app to interact with it and do things like voice chat (or rather I'm not going to jump through those hoops even if it could), that's another place where ChatGPT excels.
Also I don't think Claude 3 is going to write me an iOS app to interact with it and do things like voice chat (or rather I'm not going to jump through those hoops even if it could), that's another place where ChatGPT excels.
Agree on Claude UI. Its selling point is, I have a bigger context window. But the chat UI is this tiny little box that doesn’t extend. Small nit which I’m sure they will fix.
My very unscientific test is to ask new models I'm trying to build a basic React Native component using a specific library. "Please make me a react native component, using typescript and react native reanimated that creates an animation of a circle that bounces when it hits the ground. It should loop indefinitely."
Both GPT4-turbo and Claude 3 Opus produce a generally acceptable output. Claude 3 gave me a ball that bounces off the "ground" and "roof", but in a smooth and continuous motion. GPT4-turbo gives me a ball that bounces on the ground then "snaps" back up to the top and does it again. Mistral large gives me garbage output. It imports a bunch of function from react-native-reanimated that don't exist and produces nothing usable.
Technically GPT4-turbo's response is a bit more faithful to what I asked, I suppose. However, Claude 3 Opus animation looks much more "polished" even though I never asked for the ball to bounce off the roof.
Both GPT4-turbo and Claude 3 Opus produce a generally acceptable output. Claude 3 gave me a ball that bounces off the "ground" and "roof", but in a smooth and continuous motion. GPT4-turbo gives me a ball that bounces on the ground then "snaps" back up to the top and does it again. Mistral large gives me garbage output. It imports a bunch of function from react-native-reanimated that don't exist and produces nothing usable.
Technically GPT4-turbo's response is a bit more faithful to what I asked, I suppose. However, Claude 3 Opus animation looks much more "polished" even though I never asked for the ball to bounce off the roof.
In my experience, Mistral Medium seems to give better output than Mistral Large. And they have the audacity to charge 75% of GPT4.
I have tested it in huge contexes and it leads. It is great. Code writing? Don’t know. Refactoring? Miles ahead from GPT4. No laziness, true large context, YMMV
Every time I've mentioned having problems with GPT4 I've been told that I mustn't be prompting it correctly. Now suddenly Claude 3 comes out and everyone is admitting GPT4's shortcomings.
Both things can be true though
Yes, but that level of nuance isn't offered by most people. Most people (on HN, at least) will tell you that if your results are bad, it's because you're using it wrong. Or at least they would have before Claude 3 was released.
Where are all these doing it right examples? All I've been able to find are pay walls for bogus classes and endless lists of things like "I want you to act as a garbage collecting AI expert, you know how to collect all kinds of garbage".
here’s one by simonw that shows some of the ridiculous prompts he had give in order to get the LLM to cooperate.
https://simonwillison.net/2024/Mar/23/building-c-extensions-...
https://simonwillison.net/2024/Mar/23/building-c-extensions-...
I have been using Claude 3 since it came out. And I am somewhat surprised to find anybody who claims that GPT-4 is about equal. For software dev tasks, for me, there is no comparison. I always have both tools open and I feed them the same exact question. Claude has been winning nearly 100% of the time.
The only downside for me is that Claude's limit of questions you can ask per day is much more limited than ChatGPT-4's. You exhaust your limit much faster if you feed it a lot of tokens/data. Which I often do.
The only downside for me is that Claude's limit of questions you can ask per day is much more limited than ChatGPT-4's. You exhaust your limit much faster if you feed it a lot of tokens/data. Which I often do.
I was impressed by the IKEA demo until I looked more closely and realized it was complete gibberish. Every single part is number "101339" and it keeps talking about "attaching the hammer-shaped piece"; it gets worse from there. Meanwhile the article breathlessly claims perfect performance. Given the clear lack of even cursory critical analysis I would take the other less verifiable claims with a small mountain of salt.
Saw this too. The multiplayer game seemed impressive though
> Developers worldwide have tested it…
Claude does seem to be explicitly blocking people living in GDPR jurisdictions to sign up. I don’t know what to make of that, but this statement from the first paragraph rings false.
Claude does seem to be explicitly blocking people living in GDPR jurisdictions to sign up. I don’t know what to make of that, but this statement from the first paragraph rings false.
Well, there is two Claudes: first, there is claude.ai. Second, there is the Anthropic API. (And there is also AWS Bedrock, but it doesn’t offer Opus). Signing up for the Anthropic API from within the EU works without hassle.
The reason for the geo-block against the EU may also be rooted in the AI act, rather that GDPR. That would explain why the end-user service claude.ai is blocked off, and foundation model access remains available.
To use Claude 3 through Gradio and/or HuggingFace, see https://huggingface.co/spaces/ndurner/claude_chat
The reason for the geo-block against the EU may also be rooted in the AI act, rather that GDPR. That would explain why the end-user service claude.ai is blocked off, and foundation model access remains available.
To use Claude 3 through Gradio and/or HuggingFace, see https://huggingface.co/spaces/ndurner/claude_chat
I have used ChatGPT’s responses in some cases and Claude’s in others. I never had a “oh wow, X is way better” case with either of them and sometimes I even sent the same replies to both for a few rounds of back and forth since they made the same mistake/misunderstanding.
However, I will say that ChatGPT’s UI is better than Claude’s. The chat-like interface with the conversation list on the left hand side feels better, the larger chat text input box and wider UI of ChatGPT feels better as well. Claude is very narrow for no good reason. Also the model selector for Claude is annoyingly placed, I’ve accidentally clicked it on multiple occasions. Lastly there is no app for Claude, there is no voice chat (that I’ve seen), nor can you modify the prompt/response which I didn’t think I’d miss as much as I do.
After this month is up I think I’ll stop paying for Claude unless something changes but I’ll continue to double-ask until then.