# A New Kind Model From a ChatGPT Inventor Is Thrilling Developers

URL: https://technosports.co.in/kind-model-typesafe-jev/  
Published: 2026-09-19  
Updated: 2026-09-19  
Author: Raunak Saha

The kind model from ChatGPT inventor Ilya Sutskever’s reportedly former OpenAI colleague Diogo Almeida has reportedly arrived this week at TypeSafe AI, where Jev makes calibrated decisions instead of gene**$1 billion** and reached a **$5 billion valuation**, with Andreessen Horowitz, Sequoia Capital, DST Global, and SV Angel among its initial investors.

## The Kind Model From ChatGPT Inventor Targets a Specific Weakness

Almeida reportedly helped build ChatGPT and worked on reinforcement learning from human feedback, or RLHF. His criticism is direct: language models are optimised to communicate with people, while software automation often requires a precise decision, classification, or probability.

That distinction explains why a capable chatbot can still be awkward inside a business workflow. A text model may produce several plausible answers when an application needs one predefined output. Developers then have to add parsing, validation, retries, and safeguards around the model. TypeSafe AI’s Jev reportedly takes a different route. The reportedly transformer-based system does not operate as a large language model and does not return ordinary text. Instead, users reportedly define the possible outputs in advance, and Jev assigns probabilities to those choices.

![A New Kind Model From a ChatGPT Inventor Is Thrilling Developers](https://technosports.co.in/wp-content/uploads/2026/09/1000229364-1002x1024.jpg)

## How Jev Differs From Standard ChatGPT-Style Systems

The design has **three** clear advantages. Because Jev reportedly produces a constrained answer space, it cannot invent an output outside the categories the developer created.

It also reportedly avoids the cost of gene[Best Prompt ChatGPT](https://technosports.co.in/prompt-caching-strategies-chatgpt/), where careful prompting remains useful for language-oriented systems. The downside is equally clear: Jev reportedly cannot replace a general-purpose assistant. If a user needs an explanation, a draft email, code, or an open-ended research answer, a text-gene

| Approach | Main strength | Main drawback |
| --- | --- | --- |
| Jev | Fast, constrained decisions | Limited to predefined outputs |
| Large language model | Flexible text generation | Can hallucinate or require parsing |
| Hybrid system | Combines decisions and explanations | Adds integration and ope |

## Why Developers Are Paying Attention

According to [TechCrunch](https://techcrunch.com/2026/09/18/a-new-kind-of-ai-model-from-a-chatgpt-inventor-is-thrilling-developers/), developers reportedly showed strong interest in Jev after its release, creating enough demand for TypeSafe AI to briefly lose access to the product. That reaction reflects a practical frustration across AI engineering: the most impressive model is not always the most useful component in production.

The market already contains specialised coding and automation tools, including [Cognition SWE-2: Model](https://technosports.co.in/releases-swe-2-cognition-coding/), but Jev reportedly attacks a different layer of the stack. It treats the model as a decision engine rather than a conversational partner. That approach may also reduce the risk of silent errors, but it does not eliminate them. A badly designed set of possible outputs can still produce a confidently incorrect decision, and the model’s probabilities require calibration and monitoring before they can guide high-stakes actions.

## What Comes Next for TypeSafe AI

The most sensible path is not to replace every language model with Jev. Companies should use Jev when the task has fixed outcomes and measurable costs, while retaining a conventional model for explanation, discovery, and unstructured requests. Teams already managing services such as [OpenAI Pauses ChatGPT](https://technosports.co.in/openai-pauses-pro-astra-signups/) will recognise the broader lesson: model choice should follow the workflow, not the hype.

**If the job is classification or ranking, choose Jev; if it demands language, choose a large language model (LLM); if it needs both, accept the cost of a hybrid system.** The kind model from ChatGPT inventor Almeida’s team could push AI builders toward smaller, purpose-built systems where reliability matters more than conversational polish. The next test will be whether developers can turn that promising architecture into dependable products at scale.

---

## FAQs

### What is Jev?

Jev is a reportedly transformer-based AI model from TypeSafe AI that produces probabilities and predefined decisions instead of generated text.

### Who founded TypeSafe AI?

Former OpenAI researcher Diogo Almeida reportedly founded TypeSafe AI after working on ChatGPT and RLHF.

### Can Jev replace ChatGPT?

No. Jev is reportedly designed for constrained automation, while ChatGPT-style models remain better suited to open-ended conversation and text creation.

### Why are developers interested in Jev?

Developers are reportedly interested because its fixed outputs can reduce hallucinations, speed up automation, and lower the amount of generated text that must be processed. Source: [TechCrunch](https://techcrunch.com/2026/09/18/a-new-kind-of-ai-model-from-a-chatgpt-inventor-is-thrilling-developers/) Understanding kind model from fully means staying ahead of these developments.