Hey guys this was a Vision AI project that I have been working on for recording live chess games via AI. It has perfect move reconstruction on a dataset of roughly 50 unseen videos. The data collection, simulation, training, move tracking, etc. was quite involved.
Interesting. Any way of running this on a desktop with multiple videos simultaneously? Or an Android version that can run on cheap phones and send the timing and moves to a server?
Models above 1T params make the argument moot. You need infra to actually serve it. The scale of serving infrastructure alone will keep AI labs in the lead.
"Meanwhile, I’m over here asking ChatGPT to rewrite the same paragraph for the third time because it keeps defaulting me into ‘LinkedIn wisdom post’ mode. GARH."
Did they tell it to write it differently? You can literally make it sound like a pirate if you want. You can also make it be conservative, non-hype. Just ask.
Thanks! :) I feel like there is a dreadful lack of local-only apps that are runnable on a single simple server, now that everything is overly distributed. Should we bring back more P2P apps?
Go’s restraint in adding new language features is a real gift to its users. In contrast, Swift feels like a moving target: even on a Mac Studio I’ll occasionally fail to compile a simple project. The expanding keyword list and nonstop churn make Swift harder to learn—and even harder to keep up with.
I watched the search for the Higgs Boson and the search for Cold Dark Matter carry on in parallel for decades.
The former was clearly actual science: they had a theoretical particle, they knew what it did, it had a place that made sense in the Standard Model, they had an estimate for the energy range in which they could find it, they built an instrument to look for it, and they found it.
The latter... well, it was clearly epicycles. Endlessly tweakable, with six free parameters, not in the Standard Model, a bunch of different guesses as to what it actually was, a bunch of different energies at which it might be found – oh dear, not there, well it must be at a much higher energy then – always on the brink of discovery but never actually discovered...
And then, as I began researching my book on cosmological natural selection, I could see that an evolved, fine-tuned universe was going to have startling emergent-looking properties built into its developmental process. Baryonic matter was going to pull off some weird shit, as the interaction of extremely fine-tuned parameters led to highly unlikely-looking outcomes. These would look like inexplicable anomalies, if your fundamental assumption was that we lived in a random and arbitrary one-shot universe.
And cold dark matter started to look awfully like the kind of think you would have to invent to save the old paradigm...
So as I developed my approach, I assumed dark matter was an error, and did my best to explain everything using fine-tuned parameters, and baryonic matter only.
"Though BGP supports the traditional Flow-based Layer 3 Equal Cost Multi-Pathing (ECMP) traffic load balancing method, it is not the best fit for a RoCEv2-based AI backend network. This is because GPU-to-GPU communication creates massive elephant flows, which RDMA-capable NICs transmit at line rate. These flows can easily cause congestion in the backend network."
