Claude (https://claude.ai) is a new conversational AI assistant that shows promise for coding and technical discussions. While a generalist model, its training is focused on programming topics. In my testing, its capabilities seem on par with GPT-4, and it outperforms the free GPT-3.5 version.
For now, Claude is completely free to use, but access is limited to the US and UK (I am sure you can think of what to do here). Compared to OpenAI’s ChatGPT-4, Claude has some unique features like allowing you input book-size text. It also handles file uploads natively as part of the model.
With several more tests I feel confident to say that Claude is more reliable than GPT3.5 but it is NOT as good as GTP4. It hallucinated quite badly in few occasions. For example, to update Debian it suggested to install and launch “debian-updater”, which does not exists. However when I submitted code, it was quite good.
I think the developers are still training the model incorporating these chats in the training set.
For web search for now I prefer to have a traditional search engine, and mix the two (search results and chat with the LLM) by myself.
Because the incentives are completely broken, google is a shadow of the former self. Between SEO (Search Engine Optimization) by web owners and advertisement agencies paying google for their ads, the system is polluted and becoming almost useless. I am testing now a for-paying product that should not follow the same practices. It is https://kagi.com. So far I see that it is a superior engine compared to present day google, but I think google of 10 years ago was still superior.
