TypeSafe AI Launches Jev, a Fast Typed-Decision Model at $0.042 per Million Tokens

·12 views

TypeSafe AI launched Jev on September 15, 2026, a frontier AI model designed to return typed, probabilistic decisions rather than generated text. Built by Diogo Almeida, co-inventor of RLHF and InstructGPT at OpenAI, the model raised $40 million in funding led by DCVC. Jev processes program state alongside structured questions in a single parallel pass, with end-to-end latency between 70 and 500 milliseconds and output priced at no additional cost. The model offers three question primitives — Choice, Score, and Noul — intended to handle fast, categorical judgments that TypeSafe calls 'System One' decisions, borrowing from psychologist Daniel Kahneman's framework. The company claims zero structured-output errors by design, though these benchmarks are self-reported and have not been independently verified.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How to Clean Excel Data Before Converting It to SQL INSERT Statements

Transferring data from Excel to a SQL database is prone to errors when source data is inconsistent or poorly formatted. Common issues include numbers stored as text, mixed date formats, special characters, leading or trailing spaces, and empty cells that are not properly normalized to NULL values. These inconsistencies trigger SQL errors such as data type conversion failures or string truncation warnings, which can corrupt databases and require hours of debugging. Developers have traditionally relied on Excel formulas, Find and Replace tools, and VBA macros to clean data, though these methods are time-consuming and require coding expertise. A structured pre-conversion cleaning process — addressing data types, duplicates, encoding issues, and irregular table layouts — is recommended to ensure reliable and error-free SQL imports.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Three-Tier Autonomy Framework Emerges as Safe Standard for AI Media-Buying Agents

As AI agents become more capable of managing advertising spend autonomously, a three-tier framework has emerged from production deployments to balance efficiency with oversight. At Level 1, the agent only surfaces recommendations while a human retains full decision-making authority; at Level 2, the human approves actions before the agent executes them. Level 3 allows the agent to act independently, but only within tightly defined KPI boundaries and with kill switches in place. Full autonomy has repeatedly failed in real-world use, with agents misreading bot traffic, misjudging temporary performance dips, or generating policy-violating creatives. Operators in regulated industries such as finance, iGaming, and healthcare are advised to avoid L3 entirely until at least three months of reliable L1 and L2 data have been established.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

ChangelogPro Generates Release Notes Locally in Browser Using On-Device AI

A developer has built ChangelogPro, a browser-based tool that converts raw git commit messages into structured release notes without sending any data to external servers. The tool uses an on-device AI model powered by WebGPU, meaning all processing happens locally in the user's RAM rather than on a cloud platform. Users paste their git log output into the tool, which then categorizes commits into user-facing features, fixes, and internal chores automatically. The approach eliminates API costs, rate limits, and privacy concerns that come with third-party SaaS changelog tools. The tool also works offline after the initial load, though it requires a modern browser and sufficient local hardware to run inference.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How a PHP/Symfony Developer Handles Dependency Injection After Switching to Go

Software engineer Anton, who works primarily in PHP/Symfony and Go, is in the process of migrating a live PHP monolith into Go microservices. In a detailed technical writeup, he reflects on Symfony's dependency injection container as the feature he missed most during the transition. He breaks down six core capabilities of Symfony's DI system — including autowiring, service decoration, lazy loading, and compiled containers — analyzing both what each feature provides and what complexity it conceals. The compiled container, he notes, is particularly valuable because it catches missing or ambiguous dependencies at build time rather than during live requests. Having moved past the comparison, Anton describes what he built in Go instead, emphasizing that his approach is specific to his own codebase and not intended as general advice.

0 comments Read more at DEV Community