I just got fired from Microsoft. I was the guy responsible for making sure that when you start typing your password in on the Windows lock screen, it ignores your first keystroke for no fucking reason
I’d like to take this time to celebrate one of the unsung heroes of Xmas, the toilet. Thank you for every thing you do (especially today!)
wishing you all a Merry Xmas!
🚽🎄
For those going home to visit family this weekend:
• Samsung calls it Auto Motion Plus
• LG calls it TruMotion
• Sony calls it Motionflow
• Roku calls it Action Smoothing
• Google TV calls it Motion Enhancement
• Vizio calls it Smooth Motion Effect.
If you shoot a feature film on iPhone, is Apple entitled to some percentage of the revenue? They had to spend money on R&D to make good cameras, after all.
That’s Apple’s argument for stealing 30% of third-party developers’ revenue: “We had to spend money making all the APIs, so you owe us.”
The way I see it, those expenses are just like the expenses to keep improving the camera: Apple either spends that money to improve the platform or customers switch to the platforms that do.
Why is it only app developers who have to fund Apple’s R&D? Why not filmmakers or musicians? Or anyone who uses Apple’s platforms to make money?
It’s an incongruous stance. Apple should charge 5% to cover the costs of hosting the App Store and then make the rest of its billions by releasing great products, not taking rent from small developers.
NVIDIA research just made LLMs 53x faster. 🤯
Imagine slashing your AI inference budget by 98%.
This breakthrough doesn't require training a new model from scratch; it upgrades your existing ones for hyper-speed while matching or beating SOTA accuracy.
Here's how it works:
The technique is called Post Neural Architecture Search (PostNAS). It's a revolutionary process for retrofitting pre-trained models.
Freeze the Knowledge: It starts with a powerful model (like Qwen2.5) and locks down its core MLP layers, preserving its intelligence.
Surgical Replacement: It then uses a hardware-aware search to replace most of the slow, O(n²) full-attention layers with a new, hyper-efficient linear attention design called JetBlock.
Optimize for Throughput: The search keeps a few key full-attention layers in the exact positions needed for complex reasoning, creating a hybrid model optimized for speed on H100 GPUs.
The result is Jet-Nemotron: an AI delivering 2,885 tokens per second with top-tier model performance and a 47x smaller KV cache.
Why this matters to your AI strategy:
- Business Leaders: A 53x speedup translates to a ~98% cost reduction for inference at scale. This fundamentally changes the ROI calculation for deploying high-performance AI.
- Practitioners: This isn't just for data centers. The massive efficiency gains and tiny memory footprint (154MB cache) make it possible to deploy SOTA-level models on memory-constrained and edge hardware.
- Researchers: PostNAS offers a new, capital-efficient paradigm. Instead of spending millions on pre-training, you can now innovate on architecture by modifying existing models, dramatically lowering the barrier to entry for creating novel, efficient LMs.
Taking our new album Humdrum’s test pressings for a spin. Thank god we’re experts at handling vinyl the right way.
Out October 24th, head to our bio to preorder.