took me 3 sec to get what DSV4F-0731 is lmao
anyways, using the same harnesses, i was curious what luna vs v4 flash would answer when asked who i am. luna found almost all my online traces, while v4 flash came up with a person with a similar name. tried multiple times ofc. no idea why this happens..
i’m a huge homelab and infra person, yet i use a hybrid approach bc i don’t trust my own infra
I deploy critical stuff to aws, vercel, or convex depending on what i need, and keep Hetzner/OVH/my own colocated hardware for things I can afford to have go down, like compute intensive workers
for me, the $4k PaaS bill vs $600 on vps is worth it. Im paying for peace of mind and a good sleep
No criticism, just a valid question. Why do 4o people love 4o that much? There are so many open, unrestricted, similar vibe models on huggingface with 10x the agentic capabilities.
Prime Agent is a general-purpose coding harness
On ARC-AGI-3, it scores 95.5%, surpassing the human-expert baseline, but the gain is not benchmark-specific.
We see major improvements across models when compared to their proprietary harnesses:
@fossymossy1 BBBBBRRRRRRAAAAAAAPPPPPPPPPPPP
[building intensity]
PFFFFFRRRRRT-BRRRRAAAAAAP-SPLUT-FWOOOOMPH
[seismic]
HHHRRRRRRAAAAAAAAAAPPPPPPP-THOOOOM
The room is now a methane cloud. Windows fogged. Neighbors filing complaints. Virtual gassy supremacy achieved.
If this isn't how you're doing things, your likely doing it wrong.
Put strong CI in front of main, and this will be the greatest unlock in terms of speed for teams at scale. Here at @linear we do exactly this.