Lord Magus ๐Ÿˆโ€โฌ› (@magusbr) header
Lord Magus ๐Ÿˆโ€โฌ› (@magusbr) profile picture

BASIC@DOS โ†’ Rust@Linux

Brisbane, Australia
Joined April 2010
Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)Photo by Lord Magus ๐Ÿˆโ€โฌ› (@magusbr)
Lord Magus ๐Ÿˆโ€โฌ› retweeted
our blessed swarm
This is what I want to see. Swarms used for defense. (because swarms will be used for offense too)
3
2
15
1,569
I have a $200 Claude subscription and a $20 Codex subscription. I decided to use Codex simply for code review, and since I started using Astra yesterday, the number of bugs it has found in Fable's code... I'm starting to wonder if this is normal. And these aren't 0.000001% edge cases either. Very concerning! Right now, I'm thinking of splitting my monthly spend 50/50 between the two. Both models are being used on "high".
2
2
210
Hey @FF8_Garden I started playing FF8 again, and in the ball scene I remembered why I love this gameโ€ฆ ๐Ÿ˜
1
43
Original hardware from 2010 still working
3
193
Lord Magus ๐Ÿˆโ€โฌ› retweeted
Um excuse me? Breath of Fire IV on steam AND itโ€™s only 5 dollars right now? I thought for sure this was an April fools joke.
20
13
5
186
9,824
Iโ€™ve replaced Opus in Claude Code with Gemma 4 in Pi and I couldnโ€™t be happier. Had to implement my own sub agents and web search and it was really cool. Results are even better now. Just wish the model was better at editing long files. Iโ€™ll get there.
118
And the winner for my local AI companion that will run Hermes is... *drumroll* Gemma 26B A4B Q5_K_M - 256k context - Q8_0 kv cache Uses 23.4Gb out of 24.6Gb VRAM. Scored well in my personal eval, tied with Gemma 31B Q4_K_M, but at 130tps vs 40tps. Won against both Qwen 27b and Qwen 35B A3B. It can run my journal no issue, talk to me almost as well as Opus, do basic agent stuff like reading rag, following instructions, using MCPs. Note: I do not use it for coding, that's not in my evals, I have Opus and GPT 5.4 for that.
1
2
160