This Week I Learned - Week 29 2026
This Week I Learned - * Microsoft's Phi-4 Mini is widely regarded as one of the most capable models under 4 billion parameters. It punches significantly above its weight class, outperforming older 7B and even 70B models in specific reasoning tasks. It runs at a smooth 13–18 tokens per second on flagship devices with 8 GB+ RAM (like the iPhone 16 Pro/17 Pro or Galaxy S25/S26 series). * From The Batch : - PrismML, a Caltech spinout backed by Khosla Ventures, publicly released compressed versions of Alibaba’s Qwen model on Tuesday, shrinking it from 54 GB to under 4 GB so all 27 billion parameters run on an iPhone (15 or newer). PrismML achieves the compression by reducing the models to ternary or binary quantization, reducing each value from 16 bits to just two or three possible values. - Two people asking the same model for business feedback, one in Hindi and one in Russian, may receive substantively different advice not because they’re using different models, but because th...
