#LLM
Curated coverage on «LLM»: releases, research, benchmarks, open-source.
NEWHabr AIKimi scores 96.2% on a 1C database analytics test and emerges as a fallback if Claude access is blocked
TG: @ai_machinelearning_big_data (Machinelearning)Pika Labs releases open video model H3 Flex: 6 steps instead of 50
The DecoderGoogle's StreamVibe video agent beats other models by 10.7% on response timing and quality
arXiv cs.AINew test catches hidden AI agent collusion by measuring how a partner reacts to a paraphrase
TG: @ai_machinelearning_big_data (Machinelearning)OpenAI Rewrites GPT Behavior Rules: Five Values and a Chain of Command
Habr AIVoice bots built on LLMs turn to Jev, a fast classifier, for decisions in 200–500 ms
Habr AIMoneyGate safety layer blocked all 42 fraudulent requests to local AI models
Habr AIHugging Face's BERT Tokenizer Gets a 50x Speedup by Skipping Needless Work on ASCII
OpenAI Rewrites GPT Behavior Rules: Five Values and a Chain of Command
OpenAI has updated the Model Spec, its public rulebook for GPT behavior. It now rests on five values, and a chain of command resolves conflicts between OpenAI, the…
Voice bots built on LLMs turn to Jev, a fast classifier, for decisions in 200–500 ms
Teams are moving simple decisions out of large language models (end of turn, topic change, booking confirmation) and into Jev, a fast classifier with 70–500 ms…
MoneyGate safety layer blocked all 42 fraudulent requests to local AI models
Llama 3.1 8B, Qwen 2.5 7B and DeepSeek-R1 8B on a gaming laptop rarely spotted the scam and wrote to fraudsters, "I'll transfer it today." MoneyGate stopped all 42…
AI cartoon series: a 24-second episode built ten ways, the best cost 162 credits
SkillMem stores an AI agent skill in 9.6 KB and speeds up a team 4.2x
Author puts his script before an AI jury: the verdict cost 850 thousand tokens
Worldwide Express hands disputed invoices to AI agents and cuts unresolved disputes by 75%
Open-source LLM gateway KTMS 1.0 plugs your own GPUs into the cloud and survives outages
Keystone has released KTMS 1.0, a Go gateway that routes requests to your own GPUs first, then to the cloud. It documents Redis and Postgres failures, key rotation…
Anthropic: Fold Subagents Back Into a Single Loop — One Customer's Tasks Ran Twice as Fast
Anthropic explains when to merge a chain of subagents into one agent: context gets lost at every handoff. One customer's tasks ran twice as fast after the merge.
Ontology as a checker for an AI agent: on UniProt, its error flags are only slightly more accurate than random
DeepSeek-V4.1-Flash cuts HBM KV cache to 890 bytes per token, down from 3514
IBM study: asking the client questions first lifts smart-contract transaction fidelity to 72.9%
Freefield: an open-source ComfyUI studio with 120 apps for images, video and audio
AWS gives coding agents a live site preview straight from an isolated session
Amazon Bedrock AgentCore now has Session endpoints: each agent session gets its own secured URL, and the site the agent edits refreshes in the browser as soon as a…
Mistral opens Large 4 weights: strong at code, but Chinese models score 32 to its 26
France's Mistral has put its trillion-parameter Large 4 on Hugging Face. It leads open models in coding, but trails China's DeepSeek and GLM in the overall ranking.