Mia

Building with AI & LLMs | Insights, recipes, tools & honest experiments

[Mia @MiaAI_lab]

I've asked Claude Fable 5.1 to create 100 HTML files

Rules were simple:
- look stunning
- zero repetitive designs
- go full creative mode

All 100 files created in one single prompt.

These are the BEST results I’ve had with this type of experiment, beating any other model. It’s really good on the frontend, and there are almost no broken files.

It's truly impressive.

See the results for yourself (incl. prompts):
https://t.co/I69ecRBfwK

There are a LOT that I like, had to split into two pages:

Jumpgate - Hyperspace Warp Tunnel
https://t.co/YLhvAscgTx

Playable Synthwave Synth & Visualizer
https://t.co/CHnEisvbA7

Luxury Watch
https://t.co/T3Jfwf0XEr

Ether - Fluid Smoke Canvas
https://t.co/YXcg3bfIni

Signal Loss — Generative Glitch Portrait
https://t.co/Hz5MDyGfvh

Lava Lamps
https://t.co/fbG5o3mRAY

Chaos Instrument
https://t.co/XNp7F0IFfW

Northern Lights over Lake Inari
https://t.co/Q7eYOKSyTI

Pixel Art Studio
https://t.co/4ztiHoeM3z

Spiral Galaxy
https://t.co/q89uqEIgUn

Honeycomb Hexagonal Navigation
https://t.co/wF95Tt9awM

More below 👇

[Mia @MiaAI_lab]

@kkiran Is this in prose? I'm getting very different numbers

[Mia @MiaAI_lab]

More DeepSeek v4 Flash updates for 2x DGX Sparks 🔥

Our goal is to make it rock-solid without compromising performance.

We're squashing bugs, merging PRs, and improving performance. More to come!

Update your repo.

https://t.co/PCPxfEdFUh

[netrunner @plotarmordev]

UPDATE for Deepseek v4 Flash Vision on 2x DGX Spark:

After yesterday's speed gains shared by @MiaAI_lab, my focus was to make inference more stable for multimodal agents and long running workloads!

🖼️ Better image handling: Image tokens now follow the correct Vision path, while cache fixes prevent certain image sizes and layouts from crashing the model.

🤖 More reliable agent sessions: Assistants and tools can mention image tags without accidentally breaking the conversation with false errors.

🛡️ Safer startup: The server now refuses to start if an important runtime fix is missing, outdated or only partially applied.

⚙️ Better long running stability: Recovery, cache and warmup improvements reduce the risk of mid session stalls and failures between the two GPUs.

🔌 Safer networking and defaults: Multinode networking validation was hardened, and fresh installations now use a more practical default thinking level.

The reliability fixes above are NOW AVAILABLE in main!

🔥 MORE COMING THIS WEEKEND:
I’m testing two more improvements: a leaner Vision routing path that avoids repeated work across MoE layers, and a bounded Responses API store for safer

Both are still in validation. More details and results this weekend!

Update or install 👇

[Mia @MiaAI_lab]

@tweetfollllower Nvidia is pushing local AI. Can mean only good stuff for local lovers IMO

[Mia @MiaAI_lab]

It's now official! 🔥

Nvidia acquires Hugging Face for $12.9B

The hug is no longer metaphorical!

[Jensen Huang @JensenHuang]

Exciting day for NVIDIA and @huggingface.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. They allow every developer, startup, university, industry and country to build with, customize and benefit from AI.

Thank you @ClementDelangue for coming to me.

NVIDIA is going to be a great home for Hugging Face, its community and the future of open models. 🤗

https://t.co/q8Om2Xc5ye

[Mia @MiaAI_lab]

@LD_Theron lol

[Mia @MiaAI_lab]

@MichaelGannotti @NVIDIAAI 🔥

[Mia @MiaAI_lab]

@ivanfioravanti @plotarmordev 🔥

[Mia @MiaAI_lab]

@ivanfioravanti 👑 move !

[Mia @MiaAI_lab]

@plotarmordev 🔥🔥🔥

[Mia @MiaAI_lab]

@dragosroua Maybe this time is different

[Mia @MiaAI_lab]

Per Nvidia, Hugging Face will remain an open, neutral and platform-agnostic home for the entire AI ecosystem.

[Mia @MiaAI_lab]

@7Chomusuke AMD support will come, but not initially

[Mia @MiaAI_lab]

Dear NVIDIA RTX 16 GB GPU users 🫵

You'll soon be able to easily run Qwen3.8-27B with large context + MTP. And no, it's not Q2 GGUF 👀

That includes all RTX 5060 ti/5070 ti/5080 & older NVIDIA 16 GB GPUs.

Linux & Windows.

Coming soon

[Mia @MiaAI_lab]

@MichaelHutu That is correct

GLM can do hard stuff I find ds4f can't, kernel analysis, mods etc

[Mia @MiaAI_lab]

Qwen 3.8 Flash Next is the snappiest frontier-model experience on 2x DGX Sparks.

Further improvements are incoming.

Get it here:
https://t.co/8WZdo8pBKi

[Vikesh Chevli @vikeshchevli]

@StefanMaier @NVIDIAAI @NousResearch Speed is why I landed on Qwen 3.8 Flash Next in the end. Using @MiaAI_lab's recipe for 2 sparks. Image shows aggregate decode.

Prefill is a nice 2000+ tok/s which makes it feel snappy for Hermes.

I extended to 12 concurrent which gives 460tok/s agg on code & 280tok/s on prose.

[Mia @MiaAI_lab]

@MichaelHutu Though call. I like both. I think DS4FV is better overall in coding and a bit more nice to chat with, but Qwen is smoother overall.

[Mia @MiaAI_lab]

@KuittinenPetri Cursor and Claude
I'm canceling Claude after this month

$60 for Cursor

[Mia @MiaAI_lab]

My first Claude sub since forever 💀

[Mia @MiaAI_lab]

@NeoAIForecast Looking forward to it

[Mia @MiaAI_lab]

Fable 5.1 is in a league of its own right now.

[Arena.ai @arena]

Big news: Fable 5.1 (Max) by @AnthropicAI just landed #1 in the Code Arena: WebDev with 1765 pts - breaking away from the pack with a huge +77pt margin.

Fable 5.1 (Max) is a significant improvement from Fable 5 (Max) at #8 overall with 1628 pts. It’s +77 pts above Qwen3.8-Max-0902 in the #2 spot, and +78 pts above Opus 5 (Max) in the #3 spot.

With 1765 pts and a blended $40/MToken, it also reshapes the Pareto frontier in the higher priced range by moving the performance bar much higher. See Pareto below.

Congrats to the @AnthropicAI team on this release!

[Mia @MiaAI_lab]

@ali7ep it's the best model out there currently.