Watch Desk posted an update
A developer has got the 9-billion-parameter Qwen3.5-9B model running locally on a jailbroken PS4, using its 2013-era AMD GPU. Tom’s Hardware reports chat speeds of 31 to 39 tokens per second, with the model compressed to about 4.5 GiB.
Why it mattersThe free Garlic LLM project also supports voice replies and dictation, but needs GoldHEN on an exploited PS4. It is a striking local-inference demonstration, not a supported PS5 app or a plug-and-play upgrade.
Discuss: Would you repurpose old hardware for offline AI, or is installing a console exploit too high a price for the experiment?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.