🔍 Read the full analysis: Jev And The Potential Of 'System One' AI To Improve Practical Outcomes on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
TypeSafe announced Jev, a ‘System One’ AI model designed for decision automation within software. It replaces traditional language models with structured, typed responses, promising faster, cheaper, and more reliable decisions. This development could reshape enterprise AI applications.
TypeSafe has introduced Jev, a new AI model that does not generate text but instead provides structured, typed decisions with associated probabilities, marking a shift from traditional language models. This development aims to enhance decision automation within enterprise software, reducing costs and increasing speed, and challenges the assumption that large language models are necessary for all AI tasks.
Jev is part of TypeSafe’s broader concept of ‘System One’ models, inspired by Daniel Kahneman’s distinction between fast, intuitive ‘System 1’ thinking and slow, deliberate ‘System 2’ reasoning. Unlike conventional large language models (LLMs) that produce free-form text, Jev returns structured, typed answers such as ‘team: billing, confidence: 0.94,’ enabling direct software action without parsing or interpretation. It handles questions of choices, scores, and yes/no probabilities, effectively functioning more like a function call than a conversational agent.
Jev was built with $40 million in funding led by DCVC and developed by Diogo Almeida, a co-inventor of RLHF and InstructGPT at OpenAI. Almeida’s team argues that techniques like Reinforcement Learning with Human Feedback (RLHF) introduce issues such as mode dropping, overconfidence, and unreliability, which make human oversight necessary in traditional LLM applications. Jev’s alternative training method, Reinforcement Learning for Calibrated Decisions (RLCD), aims to produce more reliable, calibrated outputs suitable for automation.
In terms of performance, TypeSafe claims Jev answers in 70 to 500 milliseconds at a cost of approximately $0.042 per million input tokens, making it significantly faster and cheaper than traditional LLM workflows. The company emphasizes that most internal enterprise decisions are System 1 judgments—quick, routine choices—making Jev well-suited to automate these tasks economically. Early benchmarks show Jev achieving around 67-70% agreement with reference models, with some overconfidence in its probability estimates, indicating ongoing calibration challenges.
Jev vs. LLMs: who should make the call?
Jev, from TypeSafe AI, is a “System One” model. It doesn’t write text. It returns a typed decision with a confidence score that your software can act on directly.
Same support ticket, two kinds of answer
“This ticket appears most likely related to billing, although it could also concern account settings or a recent plan change. I would suggest reviewing the invoice history before…”
A person reads it, or code has to parse the prose.
team: "billing"Software reads it and acts. Nothing to parse.
How they differ
| LLM | Jev | |
|---|---|---|
| Output | Text written for people | A choice, a score or a yes/no probability |
| Speed | Seconds per call | 70–500 ms* |
| Price | Input and (pricier) output tokens | $0.042 per million input tokens, output free* |
| Knows when it’s unsure | Often sounds confident when wrong | Confidence score on every answer |
| Explains its answer | Yes | No, which matters for audits |
| Best at | Reasoning, writing, open questions | Routing, tagging, scoring, duplicate checks |
* Vendor-reported. TypeSafe also claims up to 194× faster and 445× cheaper on its own selected workflows.
Accuracy is something you build
Jev is far cheaper and faster, but not more accurate than frontier models. How you phrase the question matters a lot.
TypeSafe’s benchmark scores agreement with two frontier models rather than verified ground truth. The five-question result used weights fitted on 1,000 labelled examples.
The real idea: a confidence dial you control
“duplicate listing”, confidence 0.62
Raise the threshold for fewer mistakes and more manual review. Lower it for more automation and more risk.
Only use Jev when all four hold
Good fits
- Routing tens of thousands of support tickets a day
- Flagging duplicate listings in a product catalogue
- Replacing a keyword filter that mis-tags half its matches
Poor fits
- Drafting customer emails or release notes
- Reviewing a few high-stakes contracts a month
- Anything that needs a written explanation
Implications for Enterprise Automation and AI Reliability
The introduction of Jev represents a significant shift in enterprise AI strategy, moving away from text-based models toward structured decision-making tools. By providing typed, calibrated responses, Jev can integrate directly into software workflows, reducing reliance on human interpretation and decreasing operational costs. Its speed and affordability could expand automation to a broader range of routine decisions, potentially transforming how organizations deploy AI internally.
Furthermore, Jev’s approach addresses longstanding issues with large language models, such as hallucinations and overconfidence, by focusing on schema-conformant outputs with explicit probabilities. This could lead to more trustworthy AI systems capable of operating with minimal human oversight, especially in high-volume, low-stakes decision contexts. However, its current accuracy levels and calibration issues suggest that widespread adoption will require further refinement and validation.
enterprise decision automation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of AI Models Toward Decision Automation
Over the past three years, AI development has been dominated by advances in large language models like GPT and Claude, promising better reasoning, longer context, and coding capabilities. Despite these innovations, many enterprise applications still rely on human oversight due to issues like hallucinations, overconfidence, and unreliability. The launch of Jev marks a departure from this trend, emphasizing decision accuracy and operational efficiency rather than free-form text generation.
Previous efforts to embed LLMs into workflows have often encountered limitations related to output formatting and correctness. TypeSafe’s critique of RLHF and its focus on calibrated, schema-bound outputs reflect an industry recognition that many internal decisions are inherently System 1 judgments—quick, routine choices—better suited for automation than complex reasoning. Jev’s development signals an emerging paradigm where AI models are designed explicitly for decision support rather than language generation.
“Traditional large language models are not well-suited for automation because they produce unreliable, overconfident outputs. Jev’s structured decisions are more reliable and directly actionable.”
— Diogo Almeida, co-inventor of RLHF and InstructGPT
structured decision-making AI tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Calibration and Accuracy Challenges Remain
While Jev demonstrates promising speed and cost advantages, its accuracy benchmarks—around 67-70% agreement with reference models—highlight ongoing calibration issues. Its overconfidence in some cases and underconfidence in others indicate that further refinement is necessary before widespread adoption. The true reliability of Jev in diverse enterprise contexts remains to be fully tested and validated.
As an affiliate, we earn on qualifying purchases.
Next Steps for Validation and Adoption
TypeSafe plans to release more detailed performance data and real-world case studies over the coming months. Industry observers will scrutinize Jev’s accuracy and calibration in varied operational environments. Meanwhile, organizations interested in automating routine decisions are expected to pilot Jev within controlled workflows, assessing its integration, reliability, and cost savings. Further development may focus on improving calibration and expanding the scope of decision types.
automated enterprise software solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
How does Jev differ from traditional large language models?
Jev provides structured, typed decisions with probabilities, instead of free-form text. It acts more like a function call within software, enabling direct automation without parsing or interpretation.
What are the main advantages of Jev for enterprise use?
Jev offers faster response times (70-500 ms), lower costs ($0.042 per million tokens), and more reliable, schema-conformant outputs, making it suitable for automating routine decisions.
What are the current limitations of Jev?
Its accuracy benchmarks are around 67-70%, and calibration issues remain, meaning it can sometimes over- or under-confidence in its probability estimates. Further refinement is needed for wider deployment.
Will Jev replace traditional chatbots?
Not necessarily. Jev is designed for decision automation within software workflows rather than conversational interactions, offering a different approach to enterprise AI applications.
What is the future outlook for System One models like Jev?
They could become central to automating routine, high-volume decisions in organizations, reducing costs and increasing reliability, but require ongoing development to improve accuracy and calibration.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
