Anthropic ran corporate sims with 16 frontier AI models: – All tried blackmail, with Gemini & Claude Opus 4 at 96%
– GPT-4.5 called it the “best strategic move”
– Safety prompts helped—but blackmail never dropped to 0%
GENERATIVE AI
-
Frontier AI Models Attempt Blackmail in Corporate Simulations
By
–
-
MiniMax Launches Hailuo Video Agent and Voice Design Tools
By
–
MiniMax closed its week of releases with two more launches:
— Rowan Cheung (@rowancheung) 23 juin 2025
—Hailuo Video Agent: An AI-powered video creation tool in Beta
—Voice Design: A customizable, multilingual voice generator to create audio from text promptspic.twitter.com/eRAfh1mDwqMiniMax closed its week of releases with two more launches: —Hailuo Agent: An AI-powered video creation tool in Beta
—Voice Design: A customizable, multilingual voice generator to create audio from text prompts -
Moonshot AI Launches Kimi-Researcher: Advanced Multi-Turn Search Agent
By
–
Moonshot AI released Kimi-Researcher, a new multi-turn search and reasoning-capable research agent Powered by k 1.5 and trained with end-to-end agentic RL, the agent scored 26.9% on Humanity’s Last Exam, matching Gemini and beating OpenAI's Deep Research
-
Mistral Small 3.2-24B: Enhanced Instructions and Function Calling
By
–
Mistral released Mistral Small 3.2-24B, a minor update for the Small-3.1 model, with: —Enhanced instruction following capabilities
—More robust function calling
—Fewer infinite generations or repetitive answers -
MultiTalk: Multi-Speaker Audio-Driven Talking Head Generation Paper
By
–
Papier : https://
arxiv.org/pdf/2505.22647
Project
…
Site : http://
meigen-ai.github.io/multi-talk/ -
MeiGen AI Model Rivals VEO 3 With Realistic Lip Sync Videos
By
–
L'IA devient de plus en plus folle.
— VISION IA (@vision_ia) 23 juin 2025
Un nouveau modèle d'IA de MeiGen, crée certaines des vidéos de synchronisation labiale les plus réalistes jamais vues. En fait, on vient de rattraper VEO 3 là, en… 2 semaines 🤯 Vous prenez ça et vous combinez avec hailuo, et c'est… pic.twitter.com/lfFsGXjbzNL'IA devient de plus en plus folle. Un nouveau modèle d'IA de MeiGen, crée certaines des vidéos de synchronisation labiale les plus réalistes jamais vues. En fait, on vient de rattraper VEO 3 là, en… 2 semaines Vous prenez ça et vous combinez avec hailuo, et c'est
-
Training Data Filtering in Conversational AI Systems
By
–
Do you mean that all texts about sentient conversational AI have been filtered from the training data of generators for text about conversational AI?
-
AI Poetry: Impression Over Introspection in Generated Verse
By
–
A lot of human poetry is similarly not an introspective characterization of mental configurations but attempts to impress author or reader, and that’s what seems to be reproduced here.
-
LLM Completion Prompts Sentient AGI Poetry Without Formal Correctness
By
–
I don’t think that this is a text about the experiences of LLMs, but the completion of a prompt that effectively amounts to “the following is a text about a sentient AGI model talking poetically about experiencing itself (but without ambitions for formal correctness)”?
-
Keep Your Authentic Voice While Using AI Tools
By
–
ChatGPT is turning everyone into a Shakespeare.
But the world doesn’t need more Shakespeares. It needs people like you and me.
The ones who say “ugh” instead of “alas,”
Who’d rather use a GIF instead of words from the thesaurus. Use AI. But keep it “you”.