Seems to do the job and reduce usage; I just ran Astra for ~5 hours (using a goal) and it used the last 30% of my usage. And now they released GPT-6 Sol and Luna (which is basically 5.6 Sol and Luna, but a bit better and also 50% cheaper) ;_;
Perfect, I will incorporate this as default as well as the commit from the other guy into my own codex fork https://github.com/AmazingTurtle/codex btw. I'm rebasing on 0.156.0 right now
Taalas (Acquired by AMD, back in August) created Jimmy[0], a little chat app that runs on a POC chip with ~14k tps. Yes, 14,000 tokens per second. Sure, it's just a 8B model or so (Llama 3.1 8B), but I can imagine that having a 1.58-bit model might be helpful for their next chip.
Heck, what would happen if you used a dLLM (d for diffusion)?
I used GPT-6 yesterday[1] to generate a paladin, based on a generated texture. The front looks impressive, but then you look from the side, and... well: https://imgur.com/a/VN2g34D
I'm hoping this will do better than my setup.
[1]: using the BlenderMCP + a bunch of skills I found online
That being said: I can imagine that IF any of them generates the design of a new drug that cures _any_ cancer right now, it will still take several years before seeing results - creating the drug must be feasible, then it must be lab-tested on animals, then on humans, get approval, etc, etc.
So I'm crossing my fingers on this one, but I won't hold my breath because 5+ years is a long time.
Hizballah and Hezbollah are different spellings of the same thing, with Hizballah probably being more accurate and the one regularly used by the US government.
It clarifies what, exactly? That there are a myriad of spellings for a group whose name is originally in Arabic and couldn't care at all about how we spell it? I was just correcting the person who thought it was inaccurate to use the spelling "Hizballah".
Check the <li> near the bottom. Some are named, such as Armed Proletarian Justice, Revolutionary Class Self-Defense, and Jane’s Revenge, but some unnamed ones as well (at least their actions have been described).
Gotta love how you provide a clear, cogent answer to his question, and so people downvote you.
A/I wasn't like protonmail; they specifically provided IT support to terrorist groups, who would often be violent and would sometimes kill people. I mean to thoroughly vet your users and then make the services anonymous is a wild thing to do, all while encouraging violent users to use your site.
In the energy world, this should be such a boon over the classic NWP (Numerical Weather Prediction; complex ML models), but I've not seen it implementated. Anyone with experience of these models over classic NWP?
Its not there because a lot of the providers werent providing until recently, and higher spatial resolutions, referesh times and higher compute costs meant that no one used them in prod. We tried with ECMWF-AIFS however ENS + IFS was bettter better.
> Problem is that the WN3 grid is still quite rough (5km) - but a that's a brutal improvement for many places compared to many other global models.
That's a pretty apples-and-oranges comparison. One would almost always use a high-resolution regional model if you needed certain details for different forecasting applications like renewable energy.
It's also worth noting that the 5km outputs are from a model decoder head that was trained against temperature and dewpoint at surface stations. According to the Rasp et al (2026) preprint, this head was designed for continuous sampling; the choice of a 5km grid is arbitrary. What we don't actually know is how well the model handles shocks like a frontal passage or impacts from things like outflow from storms - or even evaporative cooling from precipitation. We are limited to the output that DeepMind publishes; we can't run the model and stress test these things on our own.
That's all a long way to say that the 5km resolution is (a) limited to temperature fields, and (b) we don't know if the "additional" resolution has any impact whatsoever on the phenomena that one would typically use a mesoscale-resolving forecast for.
This would not likely be a great idea since you reduce your ability to understand inputs except for a few parameters. Explainable inputs become very important for many down the line processes used by government and industry alike, because said inputs and their predictive certainty can be quite informative, even critical, for accurate mesoscale prediction.
If model members were available, with all the usual measures, thats a fantastic place to start looking at serious inclusion. It doesnt seem thats the case, however.
There's negative levels for faster decoding and compression. level -1 uses half the CPU of level 1 and with a custom window size can be really good for streaming compression. I think the negative levels go down to -7 if I remember correctly?
Speaking of pigz, I've run into pigzpp[1], or rather its paper: "pigzpp: Fast, Parallel, Portable Compression for the Whole Stack"[2].
Turns out we can squeeze quite a bit more compression performance out of DEFLATE - ~10x in certain instances, 2x as a base minimum (read the paper for details).
reply