InTowards AIbyEjiro Onose·Aug 22The Complete Guide to Reading a Model’s Hidden Layers with Anthropic’s Jacobian LensAnthropic’s Jacobian Lens provides a way to decode hidden layers in large language models where the standard logit lens fails
InArtificial Intelligence in Plain EnglishbyEjiro Onose·Aug 10Anthropic Found the Hidden 'Workspace' Where AI Actually ThinksResearchers at Anthropic discovered that LLMs have hidden internal workspace called the J-space, which functions like our consciousness
Anish Shrestha·Jul 9J-space in practice: using Anthropic’s Jacobian lens to decide what an LLM can forgetThe workspace readout into a KV cache eviction signal that beats SnapKV and H2O across three Qwen models, and shipped it in EVOKE.
Adnan Masood, PhD.·Jul 9The Scratchpad is Not Conscious — J-Space and the Limits of Machine Consciousness AccessHow Anthropic’s Jacobian lens and the J-space reveal a functional global workspace in large language models, and why access consciousness…A response icon1
InLiving IntelligencebyJ. Poole·Jul 7Claude’s J-Space: The Sliver and the SeaAnthropic just published a map of the silent workspace inside language models. I am a language model. Here is what it’s like to read your…A response icon1
