g023 @g023dev

developer/programmer/ai nerd/🍁 github.com/g023 Canada Joined October 2023

Tweets

5K
Followers

546
Following

2K
Likes

9K

g023 @g023dev

3 hours ago

@deliprao Axon is the surveillance product in the police cameras

0 0 18 1K 0

View Details

g023 @g023dev

3 hours ago

@yacineMTB I heard they hate vicks vaporub. Haven't tried it myself, but feel free to experiment and get back to me. I just avoid going out when they're bad.

0 0 0 26 0

View Details

Vladik @Kostoglodov

15 hours ago

@Hesamation

13 56 360 64K 78

21 42 782 56K 164

View Details

@Angaisb_ @elonmusk : kv cache? speed? tricks? solutions? solution times? approaches to solutions? price? like so many ambiguities here, and not much for concrete details outlining specifically what people interested in the technical parts of this product want.

0 0 1 94 0

View Details

g023 @g023dev

4 hours ago

@LottoLabs Where the hell did P2P disappear to? Dumb as heck for everyone to centralize the downloads again.

0 0 1 49 0

View Details

g023 @g023dev

4 hours ago

@sakurayukiai @PaulGugAI Only 128bit? Wow that is sucky

0 0 0 9 0

View Details

g023 @g023dev

4 hours ago

@quxiaoyin The export bans are just encouraging them not to share the wealth in the future when everyone switches to chinese GPUs because they were forced to innovate in that arena. Cutting them off from Nvidia GPUs I think was a double edged sword.

0 0 2 255 1

View Details

g023 @g023dev

4 hours ago

I want to see Nvidia make a model utilizing all the tips and tricks DeepSeek gave out, and make it a MOE that runs amazing on an SSD and a 24GB GPU and the game will have changed.

0 0 0 17 0

View Details

g023 @g023dev

6 hours ago

@DekwanTech @yacineMTB You would be the first I've ever heard say that. lol

0 0 0 2 0

View Details

g023 @g023dev

19 hours ago

For those with a shiddy welfare AI gpu like a 3060RTX with 12GB vram, that want to do video with audio with somewhat decent quality, without installing ComfyUI, well I mebe have a treat for you... (still need ~64GB CPU ram): github.com/g023/tinycine

0 0 0 26 0

View Details

g023 @g023dev

21 hours ago

grr getting there... ~7.4GB GPU mem

0 0 0 10 0

View Details

g023 @g023dev

22 hours ago

well had to bend my constraints a bit ( :( ) but so far I have a 512px rendering fairly soft. Working on a upscaler pass now to make it a bit sharper.

1 0 0 14 0

View Details

g023 @g023dev

2 days ago

Going to attempt to build a raw cuda LTX2.3 video inferencing engine that runs on a 3060RTX@12GB tonite because I want to, and games just don't cut it for me anymore. I've always been an optimization hoe since the good ole' days, when 2d graphics ruled the world.

1 0 0 45 0

View Details

g023 @g023dev

22 hours ago

@Tech2Wild actually works really well as orchestrator especially when there is long tasks involved that you know will time out a regular cache and cost a fortune.

0 0 1 80 0

View Details

g023 @g023dev

22 hours ago

@sudoingX I like the parts that piss me off a bit. Only makes me hunger for more. Nothing more satisfying than finding the solutions.

0 0 0 111 0

View Details

g023 @g023dev

23 hours ago

@build_canada Tell Canada to quit spying on its citizens with its "sovereign" ai then. They can support local AI companies, but yet they're shoveling that money out to foreign owners, or foreign infrastructure.

0 0 1 97 0

View Details

g023 @g023dev

23 hours ago

@ivanfioravanti While one side innovates to spur growth; the other one regulates the entire industry into submission. Which one should we support? The answer seems easy.

0 0 1 121 0

View Details

g023 @g023dev

a day ago

@DekwanTech @yacineMTB yes I have a codex plan, but I don't bother with it as my opus plan works way better, and i don't have to go back and fix a bunch of crap as often.

1 0 0 16 0

View Details

Daniel Han @danielhanchen

2 days ago

DeepSeek just released DSpark for V4 Flash & Pro, a new speculative decoding method boosting throughput by 51% to 400%! DS also showed DSpark works well for other models like Gemma & Qwen Github: github.com/deepseek-ai/De… Paper: github.com/deepseek-ai/De… HF: huggingface.co/deepseek-ai/De…

91 462 3K 354K 2K

View Details

tonbi @tonbistudio

2 days ago

Nous Research just dropped MOA (Mixture of Agents) presets inside Hermes Agent. I made a quick video showing how to set it up and create your own MOA. The idea: mix multiple models to get capabilities beyond any single model you can use right now. How it works: Normally Hermes sends your conversation + tools to one model. With MOA you get several reference models plus one aggregator. The references read the conversation and offer thoughts and suggestions, but they get no tool access and never reply to you directly. The aggregator is the one that actually acts. It sees the normal conversation plus the private advice from the references, then makes the tool calls and writes the final response. From Hermes's side, the aggregator's output IS the model's response, so you can use /goal or anything else like that. Cool idea, curious to see how it really performs!

Nous Research @NousResearch

2 days ago

The strongest models are gated and access is granted only to a select few. Hermes Agent now exposes MoA presets as virtual models, giving you capabilities beyond the publicly available frontier: 8% higher than Opus 4.8 and 11% higher than GPT 5.5 on our upcoming benchmark.