@fomothy@jack eh it’s not a skill issue, keeping the design in line is just a struggle
what i usually do is try to lock the system early and stick hard to shadcn components by keeping a short .md file that forces the model to use those styled components only
and then i hope for the best lol
@fomothy@omkarships 16 employers? okay, your HR agent is clearly battle-tested. mine has entered witness protection and will reach out once the database finishes testifying
Kimi K3 sounded like another wall of AI jargon - until software engineer @СalebWritesCode broke it down.
The architecture is the interesting part:
→ 896 expert networks, but only 16 run for each token
→ 1M-token context
→ Custom Delta Attention
→ Less communication overhead
The next generation of models won’t win by using more compute. They’ll win by wasting less of it