Experience with AIs - GPT4, DeepSeek, Claude

[Bobby_2021]

Share you experiences with different AIs and how it helped you with specific use cases.

For me the most important aspect of Claude is how much it feels like communicating with another human. So this has a latent learning effect since it always feels like some superior teacher is explaining things to you. This is a genius design decision from Anthropic.

This is different from OpenAI where it simply throws a lot of stuff at you as it feels like a machine talking.

[Bobby_2021]

On my experiences with DeepSeek:

I recently had a real life technical problem. I fed the problem into claude 3.5 sonnet and DeepSeek. 

Sonnet hinted at the correct solution but felt like it was doubting itself. I gave it a little more information and then I asked it to reconfirm itself. Then it thought through the prompt and eventually fixed on the right answer.

The same problem I fed into DeepSeek and it came to the right answer in just one prompt. It's reasoning skills are superb. It asserted the right answer in one go.

But the API is under severe load from all the users so that is a bummer for the moment. I hope they fix it so that I can use DeepSeek for the most hardest technical problems and Sonnet for more human like problems.

[Bjorn K Holmstrom]

I feel like ChatGPT is more cheery, sometimes over the top. Claude is more demure.
They both are sometimes better sometimes worse at solving my coding issues.
I haven't gotten enough experience with DeepSeek yet to judge it's tone or coding capacities, DeepSeek translated a long JSON snippet better than ChatGPT, which missed some parts (I don't waste my precious Claude tokens on such easy tasks).

[ryoko]

Ask CHATGPT for it's chain-of-thought, it'll deny you outright. This is where deepseek shines.

Best way to use deepseek is locally, using ollama and open-webUI. You need a GPU, for faster responses, it's not worth it to run on CPU.

Also, deepseek have less flexibility compared to ChatGPT, in the sense that ChatGPT can quickly change topics, but deepseek can't. It is good in some sense, but not always. It's actually an accurate name, deep-seek. The model can easily hit the depth, but have trouble jumping around into different topics.