Tag
Safin-1.5 introduces a safety-native architecture with persistent safety states and hazardous-knowledge disentanglement, trained via a discovery-evaluation-repair recursive safety improvement loop plus a formal reasoning harness, outperforming frontier models on safety and formal reasoning benchmarks.
webAI has released the TwIL family of formal reasoning models, with TwIL-LM3 outperforming a 120B model on specific benchmarks while being 40x smaller and more efficient.
TwIL-LM2, a specialized 1.7B model fine-tuned for formal logic translation, outperforms larger generalist models like Qwen3-8B and Gemma-4-26B on strict scoring benchmarks, highlighting the potential of narrow AI specialists for efficient reasoning.
TwiL-LM, a family of tiny specialized open models, is released on HuggingFace. The 3B version outperforms OpenAI's 120B gpt-oss on formal reasoning benchmarks and runs efficiently on consumer hardware.
The article argues that autoregressive language models cannot achieve true understanding of formal mathematics and need verification methods, citing systems like Aleph that rely on strict mathematical proof.
This paper introduces a relevance-sensitive evaluation suite for legal AI, demonstrating that LLMs are overly sensitive to legally irrelevant perturbations, and proposes LexGuard, an adversarial multi-agent framework using formal reasoning to improve legal reasoning reliability.
Aleph, a new formal reasoning AI system, leads major benchmarks, validating Yann LeCun's emphasis on Energy-Based Models for AI reasoning.