A bit of space, a bit of AI, a bit of running.

Here.
Joined December 2014
NRights retweeted
Computers are fucking insane man. Cannot believe we made them and they actually work
12
7
132
2,434
Ternary Bonsai 2 9B on my RTX 2060 6GB: 110 tok/s gen, 3.9 GB VRAM at 32K ctx beat Qwen3.5 9B Q4 on my small GSM8K run (18/20 vs 14/20, 3x faster), and Qwen doesn't even fit on the card then it drove @opencode through a real repo task: 8 tool calls, 0 errors, tests green in 65s nice work @PrismML. save this before you buy a GPU
1
1
49
bonus: Tesla P100 (2016, passive cooling) does 55 tok/s on the 9B PTQ1_0 vs 27 for Qwen Q4. Hermes also passed but needed 64K ctx and 42 tool calls, opencode was way more efficient on this one
26
everyone's talking about 0-60 under a second. i'm stuck thinking about the engineers tuning 10 cold gas thrusters on a draining 600L air tank, plus valve delay, plus traction control fighting it. looks fun to debug, from far away
14
What if instead of releasing agents charm labs don't release something like Muse Parrot, Hermes Dodo, GPT Tucan and we get the pirate style
12
2B active params doing agent work at 184 tok/s on a laptop. i keep saying it: the local future is a bunch of small specialized models, not one giant one
4
A stupid test i was doing with Cat Jam from @opencode, i guess is not very smart, or maybe i've been outsmarted
125
NRights retweeted
Replying to @beginbot
We are noticing more and more just how many people were larping when they were saying 'we want the humanity to progress'. Shame.
1
54
969
I was using Hermes via Discord on gpt 5.6 luna xhigh and It was feeling, then i started using Hermes Desktop and felt like i was fully missing out on this amazing experience and now with gpt 6 luna it's incredibile. But now i need the mobile versione @Teknium. @HermesWatcher help
1
3
52
Obama is perfectly displaying the concept of hallucination with LLMs here. He sounds so darn good and confident, while having absolutely no clue wtf he sayin.
🚨 OBAMA ON AI: "If we are thinking about AI just in terms of how do we cure cancer or get better energy, you can do that without having agentic AI and having it just roaming free in the internet. The reason you are doing that is because you have to market a product that people will pay money for. That’s a misalignment between what our society needs and the commercial imperatives that these companies are facing, not because necessarily they’re trying to do bad things, but because they’ve got to justify these valuations. So, that’s one more reason why it is really important for us to have a competent government and a serious bipartisan conversation around this issue, and we have to do it fast. And I would encourage voters to pay attention to this. If somebody does not have a serious plan for how to deal with this, then they’re not meeting the moment, and you should probably look for somebody else."
197
357
34
5,016
363,012
I built a TypeSafe plugin for Hermes Agent. It turns model output into a typed judgment: Choice, Score, or Noul. The plugin validates the result before your host code decides what happens next. 24-second walkthrough ↓
1
26
The same flow, in one frame: Input state → Jev → validated output → your host code. The plugin handles routing, review gates, and bounded retries. Your code keeps control of side effects.
1
13
Today i'm hitting rate limits on muse spark 1.3 every 2 minutes, did we reach agi @alexandr_wang ?
1
19
Introducing Atlas: The world's first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D. Model the world, move the camera, and simulate space & time.
773
3,480
1,537
32,382
10,865,339
Here's a video of our capsule ripping through the atmosphere at mach 25, no renders, raw footage:
75
846
199
4,578
1,798,544
After troubleshooting communications, flight controllers have confirmed Odysseus is upright and starting to send data. Right now, we are working to downlink the first images from the lunar surface.
1,412
3,175
1,100
21,084
16,575,188
NRights retweeted
Everyone knows Tupper's "self-referential" formula, one that draws its own formula. That formula is from a 2001 SIGGRAPH paper. The lore: Tupper is doing king shit, demoscene but in a graphing calculator that he invented Thread of more sick formulas from the paper:
9
345
16
4,099
251,811
Replying to @thenewarea51 @FedEx
Did you see this crazy video of the landing?
23
116
39
776
376,493
You can train Llama 2 7B on a T4 GPU which you get for free on Google Colab 🤯 How? The power of 4bit quantization and QLoRA which are easily accessible via bitsandbytes and the peft libraries! #LLMs #Llama2 #GenAI
1
1
258
The other day I discovered a little environment variable buried in the @huggingface Hub Python docs: 𝙷𝙵_𝙷𝚄𝙱_𝙴𝙽𝙰𝙱𝙻𝙴_𝙷𝙵_𝚃𝚁𝙰𝙽𝚂𝙵𝙴𝚁 It has changed my life! Docs say 2x faster, but in my testing it's 3-5x faster 🚀😍 (and it's just as fast for uploads!)
8
47
7
238
82,420