THE MINDS
SECTION 01
ISSUE 001
Post-Training Becomes the Product Engine
Projection: DeepSeek V4 combines thinking controls, open weights, and agentic coding, while Gemini 3.5 Flash integrates computer use into a current general model. That shifts product differentiation toward post-training, tool policy, evaluators, and behavior shaping. The mechanism is validated only when workflow-specific gains reproduce outside the training and evaluation harness.
Why this idea is here
What the evidence establishes.
DeepSeek documents V4 Pro and Flash with thinking and non-thinking modes and agentic coding; Google documents Gemini 3.5 Flash seeing, reasoning, and acting across computer environments. These current premises support the product-engine thesis without proving a universal post-training recipe.
Source ledger
Read the sources.
- S01DeepSeek V4 Pro and V4 Flash
official model release / published 2026-04-24 / retrieved 2026-07-10
- S02Computer use in Gemini 3.5 Flash
official model release / published 2026-06-24 / retrieved 2026-07-10