Egregoros

Signal feed

Timeline

Post

Remote status

Claude AI:
-zeroeth impression: Abysmal and invasive onboarding. You have to provide a legitimate (non-throwaway) email, your phone to even register. I wanted to activate a "free" trial so I had to give full billing details, CC and address and name, too. (I spoofed all these)

-first impression: Chatting with Claude in browser while Claude Code is installing. Pic related. Every time an answer starts. Not exactly inspiring confidence.

Replies

9
first reasoning tasks I tested it with (I'm not too stupid to figure these out, I want to test the AI), writing down negatives I noticed (4.5 sonnet):
-VM software choice, two of the recs were MS Hyper-V and WSL2 ("If you just need linux, it's pretty good!" jesus bro...)
-I told it about virtualbox auto-capturing my keyboard and asked how to fix so it only does it when I press the host key. It keeps saying the option is in the machine settings. It's not. It's in the global preferences and I have already turned it off.
To cross-check I googled it. It's on the first page and the google AI overview also has it right.

NB: The absolutely ABYSMAL "nuclear" option it suggested. This would do the OPPOSITE. It would capture my input permanently and make it harder to disable.

my guy absolutely FLUNKING his first exam
@john_darksoul Grok has been responding way better to my queries when I give it tech things, ymmv
Maybe it's prompting style. It's very mid at research, but I haven't compared alternatives.
ChatGPT is very bad for basically anything. It's very unreasonable. Like if you compare two products it will compare straight up costs, and not cost / whatever is reasonable. I expect Grok to do better but I can't objectively conclude that it does.
Haven't tested Gemini outside search AI mode

it's free. Try throwing the emoji thing at it!