Dawid

@djvdq
168 Followers
358 Following
2.6K Posts

Software developer, father, human being.

I'm writing about everything I find interesting. So probably mainly politics, both local and international, but other things as well.

Currently mainly concerned about demographic collapse.

I'm trying to be active here in English only, but I'm not very good at it 😅 so activity in Polish is also possible

For Polish: @djvdq

@rasterinterrupt these models are not that great with things that happened recently, so I'm not surprised.

I was reading a lot on Reddit how people shit on Claude Code that they use whole 5h usage with few prompts, but didn't really believe it, as I use it at work and was never even close to using it all.

Until today.

On daily basis I use Claude Sonnet 4.6 and it's ok. But I switched to Opus today to check something.

I have 2 observations:
1. It's really fast, faster than Sonnet for me
2. With 2 not that complex prompts I used 30% of my 5h usage 🤣

#claude #claudecode

@whopper yes, definitely. I managed to get 3 people to switch to Signal + 1 friend was also using it.

But I still have to use WhatsApp, because many people are still there, group chats for my kids' kindergarten are there etc.

How is it that China can speedrun and massively build, grow and use their own tech (both hardware and software) while we in Europe can just talk for years about the same thing and do nothing?

#eu #europe #europeanunion #china #technology

It looks like you're formatting your drive in FAT32. Guess who else got FAT after 32.

It's been around 7 years since #vim and later #neovim are my main text/code editors and to this day I don't get this micro management and/or gamification of configs. Spending hours working on it just to cut 17ms from startup time.

I'm doing updates (including plugin updates) to my config very rarely, as I don't treat it as yet another project to care about but as my editor that I want to just use.

LLMArena and specifically their coding leaderboard, where 27b Qwen model is 20 positions higher than 675b Mistral model shows really good, how slow Mistral is and how much they are lagging behine Chinese open source competition, not even mentioning American SOTA models.

#mistral #mistralai #qwen #qwen35 #ai #artificialintelligence

@DGutie thanks. I know what you write, and while I personally was able to do bigger steps, I was trying to do them as small as possible with everyone around. Looks that just making those super teeny tiny steps was way too much for everyone around me.

@theron29 my reasoning is that I have to fight with literally everyone, about both (de)googling and using big tech products and services in general.

In current form (which I'm not able to change, although I've tried) I need to live inside Google ecosystem, use whatsapp, have fb account etc.

@[email protected] yep, I know that. But this is the thing - I basically was able to leave just gmail (which I was using as spam email anyway) and that's all. Which isn't degoogle at all.