Models & Labs
TypeSafe AI launches Jev, a non-LLM model for automation
TechCrunch AIhigh confidence
Why it matters
- →Jev offers a non-text alternative to LLMs for deterministic tasks like classification and routing.
- →Calibrated probability outputs allow developers to build robust automation workflows without hallucination risks.
- →Drastically lower costs enable real-time model routing and agent monitoring at scale.
TypeSafe AI, founded by OpenAI co-founder Diogo Almeida, has released Jev, a new transformer-based model designed specifically for software automation rather than natural language generation. Unlike traditional LLMs, Jev outputs calibrated probabilities and does not produce text, allowing it to run significantly faster and cheaper while eliminating hallucinations. Early adopters like Vercel report 5 to 18 times faster performance in safety classification tasks compared to OpenAI’s models, with input costs metered by the billion rather than the million. The company trains Jev exclusively on synthetic data using a technique called reinforcement learning from calibrated decisions, aiming to make intelligent decision-making ubiquitous and inexpensive.
Read originalMore from TechCrunch AI
© TechCrunch AIInvestment · $100M
Market & Regulationbusiness
Vantora raises $100M for proprietary physical AI startups
Vantora’s $100M raise signals a pivot from open startup incubation to building proprietary AI ventures exclusively for corporate partners like Porsche and J.B. Hunt. This model allows companies to retain sovereignty over sensitive physical AI applications, such as retrofitting industrial hardware for autonomy, without exposing intellectual property to competitors. By shifting to a 'proprietary M&A pipeline,' Vantora unlocks high-value use cases that were previously too strategic to commercialize broadly. It represents a growing trend where enterprises prefer internal AI development over external vendor solutions for critical infrastructure.
TechCrunch AI
© TechCrunch AIModels & Labsother
Anthropic opens wet biology lab in Bay Area
Anthropic is bridging the gap between silicon and carbon by operating a physical wet lab in the Bay Area. This move validates Dario Amodei’s aggressive timeline for AI curing disease, shifting from pure simulation to real-world biological testing. The facility focuses on fundamental biology rather than direct drug discovery, likely to avoid conflict with pharma partners like Novo Nordisk. It represents a significant escalation in how top labs are approaching scientific validation.
TechCrunch AI
© TechCrunch AIGeneral AIother
AI hallucination nearly triggered US military strike
A U.S. military operation against a Chinese vessel was aborted at the last minute because the targeting intelligence was generated by an AI chatbot that hallucinated the ship's cargo. This incident reveals a critical vulnerability in military workflows: when analysts use LLMs to synthesize classified and open-source data, errors can be formatted into official-looking reports and circulate up the chain of command before human verification catches them. The speed of AI integration is outpacing the safeguards needed to verify its outputs, creating a dangerous gap where hallucinations can mimic credible intelligence. This near-miss serves as a stark warning that without rigorous oversight, the very tools designed to accelerate decision-making can trigger catastrophic geopolitical errors.
TechCrunch AI
More in Models & Labs
© The AI Daily BriefModels & Labsresearch
Google DeepMind Establishes New Institute
Google has announced the creation of the DeepMind Institute, expanding its research infrastructure and focus areas.
The AI Daily Brief
© Sam WitteveenModels & Labsmodels
Typesafe AI launches Jev for high-speed classification
Typesafe AI is betting on System 1 reasoning with Jev, a model designed specifically for ultra-fast classification tasks rather than slow, deliberate chain-of-thought generation. By prioritizing speed and direct output for structured data, it targets the latency bottlenecks that plague current LLMs in agent workflows. This approach suggests a shift toward specialized inference engines where milliseconds matter more than verbose reasoning traces. It remains to be seen if this narrow focus scales beyond controlled demos.
Sam Witteveen
Models & Labscoding
llama.cpp b11025 adds CUDA 13 and ROCm 10.0 support
This release quietly closes the hardware gap for local inference by adding native support for CUDA 13 and ROCm 10.0 alongside existing CUDA 12 builds. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get hardware acceleration, as pre-built binaries now include the necessary libraries. Apple Silicon builds have reverted KleidiAI to disabled by default, likely due to stability concerns, though it remains available. The inclusion of openEuler support for Huawei's Ascend 910b chips further expands the ecosystem beyond standard x86 and ARM architectures. This is a critical infrastructure update that ensures llama.cpp remains viable on the latest AI hardware without requiring developer intervention.
llama.cpp Releases