Do AI models become less honest when they self-report their own misbehavior?
Our Applied Research team studied a key question in chain-of-thought (CoT) monitorability: do models faithfully report their own misbehavior when they are asked to?
We introduce Monitorability
1/ Today, we’re introducing Recursion: the RL platform for building, evaluating, and deploying specialist agents.
The next phase of AI won’t just be about smarter models. It will be about systems that learn from the unique expertise of every organization. 🧵
8/ Every execution becomes a learning signal. Each rollout generates graded trajectories that improve specialist models through fine-tuning and reinforcement learning, grounded in real enterprise outcomes.
9/ As agents take on more responsibility across the enterprise, the ability to evaluate, learn, and improve from real execution will become a competitive advantage.
Reliable agents require more than a strong model. They require a system for continuous improvement.
Learn how
Where do models change their minds?
Natural Language Autoencoders (NLAs) offer a promising way to translate a model’s internal representations into natural language. But the harder question is: where do meaningful decisions actually happen?
We tested a workflow for finding
When AI benchmarks saturate, what comes next?
Historically, leaderboard saturation leads to two paths: hyper-specialized questions or increasingly abstract puzzles. A new paper from @Meta Superintelligence Labs introduces a third path: GIM (Grounded Integration Measure).
This week, we had the pleasure of hosting 50+ researchers and builders from leading AI companies to meet, talk and socialize (MTS 😎) at Labelbox HQ.
Huge thanks to @dwarkesh_sp, Sholto Douglas (Anthropic), Mo Bavarian (OpenAI), and Melvin Johnson (DeepMind) for leading our
Interrupt a voice agent mid-sentence and most models struggle to stay aligned with the original objective.
We built EchoChain 🔊, a benchmark for reasoning under interruption in full-duplex dialogue.
Current pass rates:
• Gemini Live: 16.5%
• Nova Sonic 2: 26%
•
Voice agents are moving beyond rigid turn based systems toward real time, natural conversation, streaming understanding and generation simultaneously. However, most existing benchmarks are either turn-based or latency-focused and do not directly test whether models can maintain
Common failure modes for leading audio models fall under three categories: (1) contextual inertia, (2) interruption amnesia, and (3) objective displacement. Read the full blog post for a deep dive into these types of failures.
Here’s a sample audio clip from EchoChain showing an objective displacement failure in OpenAI GPT-realtime-2025-08-28.
The conversation first establishes baseline context, then introduces an interruption, and we check whether the model stays aligned with the original goal after
