On this episode of ATP Insights #58, we discussed DeepSeek's recent release of V4.1 Flash on September 10. It is the smallest model in the company’s new architecture family, although “small” is relative: it has 552 billion backbone parameters. Its mixture-of-experts design activates 8 billion parameters while processing input and 16 billion while generating output. It also supports a one-million-token context window and accepts images and is an open weight model available under an MIT license.
We also talked about how
Z.AI has completed a financing package worth approximately $5 billion. The company placed 21.97 million new shares for about $2 billion and issued approximately $3 billion of zero-coupon convertible bonds.
About 60% of net proceeds will support its next generation of models and a proposed fully self-training system. This represents close to $3 billion before transaction costs. A further 15%, or roughly $750 million, has been allocated to expansion. The remaining capital will support the balance sheet and daily operations.
piped.video/live/Jcgnx6SZYQw…
On Today's Show:
00:00 - Introduction
03:43 -
@deepseek_ai AI’s New Model Tests the HBM Boom
15:28 -
Z.ai Makes a $5 Billion Bet on Self-Training AI
22:03 - Wei Zhuang (Lenard) Tan, Founder & CEO at Ailytics
43:45 - Dhruv Bhagat, Founder & CEO at
TapTalent.ai
01:04:52 - Uma Rudd Chia, Cofounder & Chief Marketing Officer at OH MY STRAWBERRY
01:25:06 - Duy Doan, CEO at Magnitudeminds
01:42:33 - Sunil Nair, Co-Founder at Clairva
02:08:30 - Jeanne Lim, Founder and CEO at beingAI Corporation
02:24:07 - Asange Jayathilake, Founder & CEO at Zayphire