CALM proposes predicting vector blocks, not tokens
A text published on January 24, 2026 describes CALM models that predict blocks of meaning, claiming 4× more information per step and 44% less training compute.
Source: CALM prevê blocos de vetores em vez do próximo token (x.com). Text prepared with AI from this source.
What happened and what to do
A text published on January 24, 2026 describes Continuous Autoregressive Language Models (CALM), which predict vectors for blocks of meaning rather than individual word pieces. According to the text, this approach carries 4× more information per step and reduces training compute by 44%; the proposed unit could be an alternative to token-by-token prediction.
A company developing or evaluating models could implement a comparative assessment of this approach, using criteria for quality, cost, and fit with its data and tasks. This could include evaluation and monitoring pipelines, as well as infrastructure for controlled model tests before deciding whether to make a change.
How the consultancy can help
Wendelmaques can diagnose the opportunity for the company’s use case, define the scope of a comparative assessment, and implement and operate the required pipelines, metrics, and infrastructure.
Next step
Send a short description of your case to receive a scoped proposal for diagnosis and implementation.
Let's assess your project
I reply with scope and a proposal.