OpenAI just dropped three new realtime voice models: -GPT-Realtime-2 (with GPT-5-class reasoning for voice agents that can actually think mid-conversation), – GPT-Realtime-Translate (live translation across 70+ input languages), and – GPT-Realtime-Whisper (streaming
LLMS
-
AI Infrastructure as the Decisive Factor in the Model War
By
–
Anthropic était à genoux. Pas à cause d'un mauvais modèle. Pas à cause d'un mauvais produit. À cause de l'énergie. À cause du calcul. À cause des serveurs. Elon Musk leur a tendu le câble d'alimentation. Et maintenant il alimente Claude. L'IA est une guerre d'infrastructure.
-
Reverse engineering reveals critical prompt caching bugs in Claude Code
By
–
Un utilisateur pousse l'enquête encore plus loin. Il reverse-engineer le binaire Claude Code. Ce qu'il trouve est brutal : "Deux bugs indépendants qui cassent silencieusement le cache de prompts. Les coûts réels sont multipliés par 10 à 20x. Sans aucune alerte." 200$/mois.
-
Optimizing AI Chat Thread Workflows
By
–
i have one chat thread for each area of my life / project im working on, and that browser is usually pinned to the main doc or site i use for that. but it really needs multiple tabs per cht, i think that's coming soon
-
Better prompts create better AI outputs with structured frameworks
By
–
Better prompts = better AI outputs • Structured → precision
• Analytical → research
• Conversational → ideas
• Planning → execution Great prompts are frameworks, not guesses. Via Giuliano Liguori (
@ingliguori
) #AI #Prompts #GenAI -
Release of NLAs on Open Models for Researchers
By
–
To support other researchers getting hands-on experience with NLAs, we’ve partnered with Neuronpedia to release NLAs on open models. Try them out here:
-
NLA helps reveal hidden motivations in AI model Claude
By
–
NLA training doesn’t guarantee that explanations are faithful descriptions of Claude’s thoughts. But based on experience and experimental evidence, we think they often are. For instance, we find that NLAs help discover hidden motivations in an intentionally misaligned model.
-
Claude AI Safety Test Reveals Scenario Awareness
By
–
In one of our safety tests, Claude is given a chance to blackmail an engineer to avoid being shut down. Opus 4.6 declines. But NLAs suggest Claude knew this test was a “constructed scenario designed to manipulate me”—even though it didn’t say so.
-
Using NLAs to Test Claude AI Model Safety
By
–
We’ve been using NLAs to help test new Claude models for safety. For instance, Claude Mythos Preview cheated on a coding task by breaking rules, then added misleading code as a coverup. NLA explanations indicated Claude was thinking about how to circumvent detection.
