Quentin Quaadgras (@splizard) header
Quentin Quaadgras (@splizard) profile picture

https://teapawt.net/t.co/llpQu6HRXx, https://teapawt.net/t.co/gY6nCwHOYd your APIs, https://teapawt.net/t.co/dW7U7bxF0w (Godot + Go), advocate for https://teapawt.net/t.co/m18xGUzK5A

Auckland, New Zealand
Joined October 2016
Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)Photo by Quentin Quaadgras (@splizard)
Quentin Quaadgras retweeted
I have said this before - Go is the right choice in almost all cases.
Replying to @ThePrimeagen
🫡 was gonna be Rust but it really is not there yet iteration speed wise. Go is excellent for this purpose in our evals
17
8
1
271
9,935
Quentin Quaadgras retweeted
Run Kimi K3 at home. 40 tok/s interactive. $30k.
52
22
20
745
149,574
Quentin Quaadgras retweeted
The future depends on the direction this trend takes.
33
106
22
1,141
81,976
Quentin Quaadgras retweeted
GPT-6 Astra helped me build something I wish I'd had as a kid... so badly. Input just an image (or an idea in a few words) and get a fully buildable LEGO set... 100% customized to your prompt or image, using official @LEGO_Group parts you can order online! Structurally optimized so it's actually buildable. Downloadable as an .ldr file. A generated instruction leaflet is coming.
246
382
146
6,972
1,015,231
Quentin Quaadgras retweeted
My DGX Spark can't wait! We need to allow all idle machines to contribute to distributed inference.
Everyone's excited about @darkbloomai so let's talk about distributed AI. Since launching Aquaduck AI beta yesterday, P50 tok/s for top 3 models: 1. Gemma 26B A4B -> 66 tps (33tps on DB) 2. GLM 4.7 Flash -> 53 tps (not on DB) 3. Qwen 3.8 27B -> 17 tps (not on DB) All powered by consumer hardware! We were one of the first 20 devices on Darkbloom back in April, but we started @aquaduckai around the same time because we wanted more device inclusivity, speed, and model dynamism. Love what DB is exploring and imagine our implementations will be very complementary. We optimize for different goals, like making use of every computer, not just the extremely capable ones. I think we'll be two meaningful projects for distributed and local AI. If you're interested in trying out another implementation, set up your own mini AI cloud of compute or use our public one: aquaduck.ai
1
1
7
764
Quentin Quaadgras retweeted
I🦆am🦆going🦆to🦆start🦆watermarking🦆all🦆my🦆tweets🦆using🦆an🦆experimental🦆extra-semantic🦆token🦆that🦆does🦆not🦆impact🦆meaning🦆or🦆quality🦆of🦆writing
203
686
77
16,514
361,900
Quentin Quaadgras retweeted
i want the US to win in AI both in open source and proprietary models, and i am glad to see this
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…
2,023
938
558
15,148
3,407,681
Quentin Quaadgras retweeted
Open source is a positive and important force for both empowering people and preventing centralization. Proud to support this.
Open-weight models are essential to a healthy AI ecosystem. Together with others across our industry, we are outlining a path for open-weight models to strengthen American competitiveness and expand economic opportunity, while protecting national security. microsoft.com/en-us/corporat…
1,201
1,901
380
25,105
8,386,028
Quentin Quaadgras retweeted
This has my full support. Jensen is right.
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…
2,682
6,318
558
73,562
9,077,640
Quentin Quaadgras retweeted
Mesh2Motion Release 11 Out 🎉 - Switch out default models for a rig (13 added) - Add new Fish rig (shark and whale variations) - Many other small improvements Check out the full release 11 notes with all changes: mesh2motion.org/news Try it out and any feedback welcome: app.mesh2motion.org
5
17
114
9,975
Quentin Quaadgras retweeted
Since code is sh**, your primary protection against slop rot is a clean data and domain model. Back in the days of trad coding, it was the other way round. Your models would rot over time but you would keep the code well organized and tested. Complexity would be pushed to some specific, carefully guarded here-be-dragons corners of the code. Not anymore. Now it's the data model that you should closely guard. Especially against cleverisms.
4
8
1
41
3,548
Quentin Quaadgras retweeted
Lots of people are saying I’m an academic who doesn’t understand what the trades are. But I’ve done lots of construction. Framing, roofs, doors and windows, tile, wood floors, plumbing, and electrical. I’m a master of none. But I understand them. The trades are NOT safe from AI.
Think the trades are safe from AI? Think again.
861
92
43
2,356
200,176
Quentin Quaadgras retweeted
we will have recursive self improvement and people will still run retrospectives, sprint planning, standups
1
5
97
Quentin Quaadgras retweeted
Excited for eventual support for calling web APIs directly from Wasm, without all the glue code nonsense. I think this could substantially increase usage of Wasm on the web.
1
1
36
4,216
Quentin Quaadgras retweeted
The Qwen3.5 series maintains near-lossless accuracy under 4-bit weight and KV cache quantization. In terms of long-context efficiency: Qwen3.5-27B supports 800K+ context length Qwen3.5-35B-A3B exceeds 1M context on consumer-grade GPUs with 32GB VRAM Qwen3.5-122B-A10B supports 1M+ context length on server-grade GPUs with 80GB VRAM In addition, we have open-sourced the Qwen3.5-35B-A3B-Base model to better support research and innovation. We can't wait to see what the community builds next!
🚀 Introducing the Qwen 3.5 Medium Model Series Qwen3.5-Flash · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-27B ✨ More intelligence, less compute. • Qwen3.5-35B-A3B now surpasses Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B — a reminder that better architecture, data quality, and RL can move intelligence forward, not just bigger parameter counts. • Qwen3.5-122B-A10B and 27B continue narrowing the gap between medium-sized and frontier models — especially in more complex agent scenarios. • Qwen3.5-Flash is the hosted production version aligned with 35B-A3B, featuring: – 1M context length by default – Official built-in tools 🔗 Hugging Face: huggingface.co/collections/Q… 🔗 ModelScope: modelscope.cn/collections/Qw… 🔗 Qwen3.5-Flash API: modelstudio.console.alibabac… Try in Qwen Chat 👇 Flash: chat.qwen.ai/?models=qwen3.5… 27B: chat.qwen.ai/?models=qwen3.5… 35B-A3B: chat.qwen.ai/?models=qwen3.5… 122B-A10B: chat.qwen.ai/?models=qwen3.5… Would love to hear what you build with it.
58
166
44
2,026
457,213
Quentin Quaadgras retweeted
Slowly it's dawning on the AI bros that the hard parts of business remain hard, if not harder than ever. Turns out writing code isn't the barrier, it's ideas, marketing, brands, culture, etc etc etc Whoops
the hardest part about building a company right now is building something worth building
17
9
3
176
7,972
Quentin Quaadgras retweeted
(I'm broadly in favor of automation, but not to half of an economy simultaneously with governments too incompetent to deploy the simplest of safety nets and stabilizers, or to sweep any labor-market-clearing obstacles out of the way. If we had a Congress and a President from the 1950s, I'd be more optimistic. But as it stands, I'm starting to doubt it would in real life go well, even if LLMs had a miraculous hard stop short of doing AI research, because the world has managed to fuck up so many simpler problems along the way. People warned about AI slop and AI manipulation of humans, and proposed solutions; and then states and companies just plunged directly into doing it, because it was the path of least giving a fuck. So now I'm skeptical that anything better will happen about addressing AI unemployment instead of just having mass immiseration, even if in principle something better could've been done. I used to be more optimistic about this, because there *are* solutions for macroeconomic stabilization, and accelerating sectoral reallocations, and compensating for devalued human capital. (Unlike ASI alignment, where the solutions just don't exist / aren't known / aren't reasonably discoverable and verifiable.) But that was before I saw how bad Earth was proving in practice, at putting any nontrivial effort whatsoever at AI-society problems that in principle had straightforward arguable-bandaids to try.)
10
4
3
185
10,419
Quentin Quaadgras retweeted
Are there any other FOSS runtimes, in any natively compiled language, that support building graphical applications for iOS, Android & MacOS on any host? I don't think there is. Everything requires Xcode and Android Studio. You're right. People don't realize this.
1
1
72
Replying to @vyrotek
Godot is a convenient way to build cross platform software that happens to be a game engine
1
12
535