Pinned Tweet
One time I was at my buddy's house and I set his DVR to record everything with "gay" in the description and it wasn't easy to figure out what had happened so like a **year** later I got this exasperated call because his wife legitimately was questioning him and he knew it was me
Couple things from S1 I'm seeing differently now
Father Intintola is intolerable
Why is Carmela so mad about a female psychiatrist? Why would Tony go though the trouble of finding a doctor to cheat with
The world is so completely different now it may as well be set in 1950
Can't sleep, rewatching Sopranos (S1E3) for the 100th time, watching Meadow sing with her high school, realizing I'm older than Tony now
3
289
Now Joan Holloway (my Meta Muse agent) is using the phone fleet for all of the features I've outlined here and more. She and Claude are going back and forth getting things fixed.
Update - the host orchestrator is now a Pi 400 with a tiny screen for a HUD. So far we have Qwen .6B (no QLoRA yet), embedder, tts/stt, reranker, decision model, OCR, and more. The orchestrator exposes an API that allows for work to be dumped in and handled async, no problem
4
200
We got a switch 2 for my girls delivered from Walmart on launch day and it was a guy in his 60s, met him in the driveway. I joked with him about his relative risk and he laughed about having a firearm with him, great memory
Target same day delivery guy for my PS5 just straight kept it. Got within 5 minutes of my apartment, said delivery wasn't possible, and deleted his entire delivery account.
Honestly? Fair play, I respect it. Refund for me, PS5 for him, have at it let's run GTA online together
9
312
Can't sleep, rewatching Sopranos (S1E3) for the 100th time, watching Meadow sing with her high school, realizing I'm older than Tony now
Update - the host orchestrator is now a Pi 400 with a tiny screen for a HUD. So far we have Qwen .6B (no QLoRA yet), embedder, tts/stt, reranker, decision model, OCR, and more. The orchestrator exposes an API that allows for work to be dumped in and handled async, no problem
I just bought 2x burner phones ($30 each) and I'm going to try to use them for decision model inference. If the Mali-G57 MC2 GPU can work with llama.cpp, compelling toy for hosting tiny decision models
slickdeals.net/f/20280060-st…
This also puts (good) pressure on me to get serious about finetunes because that's basically the only knob I have to improve performance beyond scaling to more devices. It also can be used across all phones and can be swapped almost instantly as the need arises
1
1
48
I had a similar experience last week and I flew into a mindless rage
In that moment I was just like the anti-ai furry troons and I had to eat a @SNICKERS to regain my composure
1
3
143
I know they're different parts of the country but I always imagine the family from Home Alone having Vin as the father
2
198
(AI influencer posts "tip")
( First reply) "Hi, you wouldn't want to do that it makes no sense and would use a lot of credit pointlessly"
AI Influencer: "Very interesting! Thanks for the tip!"
(Leaves original "tip" post up)
4
138
Just saw a post about how OpenAI gives you an allotment of free tokens every day if you're cool with them using your prompts/responses to train. Yes, you wouldn't want to use this generally but it's a huge deal if you're processing public datasets for training
1
5
282
Going to set up an N8N flow to dump work into the endpoint every day until it's exhausted and then wait until the next day. Finally a good reason to learn RLVR (Reinforcement Learning with Verifiable Rewards) and RLAIF (Reinforcement Learning from AI Feedback)
1
1
71
If there's any one thing that I love LLMs for it's figuring out tedious configuration bullshit for me. I have a bunch of Intel Arc cards (A770, A750, A580, 3xA380) and they've been a pain to use beyond games/transcoding. Now I can give them all kinds of jobs.
1
6
254
A750/A580 are about the same, best use for me is Gemma 4 E2B in Omarchy as an always-on local worker. Very capable. You can try that 4B version big but with only 8GB onboard it'll choke on fan-out
huggingface.co/OpenVINO/gemm…
1
25











