robotswantdata

They have big FDE team plus acquired a consulting company

Everything is a sine

Good modelling

Confused. Why not just make the kick drum from a sine? Seconds

Need to run this on my Xeons with AMX

Ordered. Not sure will beat my streamdeck modules, but YOLO

Novel app for live music performance (private for me and another big artist, we may release one day). Very unique

12 channel DDR5 6400, 614GB/s peak theoretical

build a Xeon / epyc 4u server. 12 channel ram.

comp.lang.c

Any plans to add more international voices and dialects?

https://developers.google.com/edge/gallery

Put that on a spare phone

Was “only” $2k in its previous form but even in this updated box the mem bandwidth is woefully inadequate. There’s a few models with space for a dedicated GPU for hybrid inference but imo not worth it. Save your money for a Xeon or EPYC build

Agree, it’s vibe slopped rather than the actual Claude design system prompt

You can put tailwind on the CV

Intel AMX is very underrated too

https://aws.amazon.com/blogs/compute/accelerate-cpu-based-ai...

Not good for planets IMO as 150mm focal length and a small 30mm aperture. Binoculars or small desktop DOB is better with kids for planets. The mini is amazing for deep sky objects for the price

Very cool project. Been taking my dwarflab mini on hikes instead of the 12” dob ;) , not the same at all but enjoy getting dark sky and some great astro shots.

I edited, but in big mega corps could easily be 5x your junior staff.

It doesn’t cover the reality that nearly all your extra pay in a progressive tax system goes to the government. UK is hitting 70% marginal rates , or even 1000% for people with children.

Missed the part about 100x the stress and politics for 1.2x the pay.

Very interesting.

The way I understand this works is that the researchers found a clever architectural hack to stop AI from hoarding memory when reading long documents.

Normally, when an AI transcribes a 100 page PDF, it tries to remember every single word it has already ingested. This short-term memory (the KV cache) grows linearly O(N) until the model runs out of VRAM and crashes (or caps it) To avoid this, developers are forced to build janky code that chops PDFs into individual pages, processes them one by one, and glues the text back together.

Unlimited OCR uses Reference Sliding Window Attention (R-SWA) to split the AI's focus into two paths:

Global Reference: The AI keeps full, uncompromised sight of the original document image so it never loses context.

Local Generation: The AI restricts its memory of its own typed text to a tight, moving window (like the last 128 words) and safely forgets the rest.

Will be very interesting for local AI and can’t wait to see what the community builds and extends with it!

Granite or sapphire rapids are very under rated for MoE inference loads. But you need a GPU for the KV cache.

Plus many boards also support CXL for RAM expansion over PCI 5!

Source: building a hybrid inference business for regulated industry workloads.

I agree, codex app’s computer use agent feels sci-fi. Closest we’ve seen yet of a general purpose virtual worker.

Zombies

Wrong antenna for the frequency?

Set up solar nodes last weekend. 200 miles of range now. Nerds, mad ideas. Good times.

This is by design. It’s like BBS again

BPM is now agentic, don’t you know.

Don’t do PEDs