New ‘Reinforcement Learning For Calibrated Decisions’ Makes AI Headlines But Look Past The Hype

In today’s column, I examine the latest AI buzzword or catchphrase, namely reinforcement learning for calibrated decisions (RLCD). Here’s the backstory. A startup firm called TypeSafe, led by what some consider a co-inventor of ChatGPT, Diogo Almeida, has announced its new product, coined Jev, which uses RLCD. There is some media hype and misleading claims about RLCD that I aim to clarify and dispel.

First, please know that RLCD is a made-up phrase by this startup firm and does not reflect any existing globally accepted standardized technique per se. Second, there are other similar techniques already, such as RLCR (reinforcement learning with calibration rewards) and RLVR (reinforcement learning with verifiable rewards), but RLCD is not the same, though it has conceptually cousin-like earmarks that I’ll explain. Third, RLCD is proprietary to TypeSafe, and it is not generative AI or an LLM; instead, it is a tool that might be used by conventional AI or even by other software applications. I realize that’s quite a litany of considerations and not what the mainstream media has the suitable discernment to point out.

Let’s talk about it. This analysis of AI breakthroughs is part of my ongoing Forbes column coverage of the latest in AI, including identifying and explaining key AI complexities (see the link here).

Reinforcement Learning Methods

Before we get into the particulars about RLCD and TypeSafe, I’d like to establish foundational aspects that will be helpful to this discussion.

One of the darlings of generative AI and large language models (LLMs) is the use of RLHF (reinforcement learning with human feedback). AI makers have leaned heavily into RLHF, which is partially what made ChatGPT into a great success, and which nearly all AI makers now embrace as a welcomed technique or method in tuning their AI.