Ex-OpenAI researcher launches Jev, an AI model that outputs decisions instead of text

·3 views

Former OpenAI researcher Diogo Almeida left the company due to concerns about large language models' inefficiency for automation tasks. He founded TypeSafe AI, which this month unveiled Jev, a transformer-based model designed for software decision-making. Unlike LLMs that generate text, Jev outputs probabilities, scores, or selections from predefined options without producing tokens. The company claims this approach is significantly faster and cheaper than conversational AI models for classification tasks. However, while Jev eliminates structural hallucinations by design, it can still make incorrect predictions with high confidence.

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)

EU's Digital Markets Act designates six major tech firms as 'gatekeepers'

The European Commission has designated six major technology companies as 'gatekeepers' under the Digital Services Act. Alphabet, Amazon, Apple, ByteDance, Meta, and Microsoft are the initial firms named. These companies now have six months to ensure their core platform services comply with new EU regulations. The rules aim to ensure fair competition and provide users with more choice and control.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Google DeepMind releases EmbeddingGemma 2, enabling local multimodal AI search

Google DeepMind announced EmbeddingGemma 2, an open-weight multimodal embedding model, on October 6, 2026. The model can process text, code, images, video, and audio into a shared vector space for retrieval tasks. It allows developers to build AI search systems that run locally on a user's device, potentially keeping sensitive data private and enabling offline access. The model is available under the Apache 2.0 license and can be integrated using Python libraries.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Stanford researchers automate conversion of research papers into interactive AI agents

A Stanford team introduced Paper2Agent, an automated pipeline that converts research papers into functional AI agents. Published in Nature on September 16, 2026, the system transforms papers into agents that can rerun the original research methods on demand. In a test with 100 computational biology papers, 74 were successfully converted into working agents that extracted 593 validated tools. These agents outperformed alternative AI approaches, scoring 91.2% versus 82.7% on benchmark questions. The system serves as both a reproducibility audit and a practical tool for executing complex research methods.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Business Central 2026 Wave 2 update previews finance features, sets Shopify deadline

Microsoft has detailed the 29.0 update for Dynamics 365 Business Central, scheduled for the second release wave of 2026. The update introduces 80 features, with all 10 finance and expense capabilities currently in preview status. A key new feature, the Payables Agent, will process invoices using Copilot Credits at an estimated cost of $0.65 per three-line invoice. The update also mandates that Shopify customers upgrade by December 31, 2026, and will disable legacy SOAP API integrations by default. Microsoft has shifted from publishing detailed release plans to a public roadmap, which does not specify which features require administrator activation.

0 comments Read more at DEV Community