yesterday:
- test ChatGPT's°hard word new "custom°hard word voices" feature°hard word. 🔥 can it do a xantham-y voice?
The voices are ... good, but not°hard word good enough. I can't°hard word see myself building a product°hard word on top of this version°hard word. Maybe the next version°hard word. The API°hard word is simple enough that it would take about 2 days to come up with an MVP°hard word.
- test Gemma°hard word3:4b and Gemma°hard word3:1b locally°hard word
The results are good, but not°hard word magical°hard word. They are tuned°hard word for chatting°hard word, but struggle°hard word more than Gemma°hard word2 on mechanical°hard word tasks 💡 copy the misspelled°hard word word from this sentence. They do fairly well on the accuracy°hard word / time tradeoffs°hard word 💡 because the hardware°hard word is the same, "time" is an accurate°hard word way of measuring°hard word "cost°hard word"
- get the greenland°hard word quals°hard word to be available°hard word on the public internet 🔥 I still haven't°hard word chosen°hard word a domain°hard word for it. Should earlyversion°hard word have file°hard word hosting? Probably not°hard word.
- look at some old EA°hard word game repos°hard word ⚙️ https://news.ycombinator.com/item?id=43197131 for commentary°hard word and download°hard word some to universe.
----
What is universe?
It is my archive°hard word of Git°hard word repositories°hard word, or similar.
1372_ea°hard word_games % du°hard word -hs°hard word
286M CnC°hard word_Red_Alert°hard word
673M CnC°hard word_Renegade°hard word
51M CnC°hard word_Tiberian°hard word_Dawn°hard word
----
for today:
- a "watchdog°hard word" for Atacama°hard word? the machine has become wedged°hard word twice in 3 months. both fixes were of the "just restart°hard word the server" persuasion°hard word. ⚔️ well, actually, the problem is that the lexer°hard word has an infinite°hard word loop°hard word on a single unpaired°hard word asterisk°hard word.
- more blogging°hard word, less technical°hard word work