Mark Saroufim Shares Tilde Research Blog on Model Limits

Mark Saroufim shared a Tilde Research blog post titled 'One Layer Deeper' that discusses models which resist learning despite added compute. The post notes that current reasoning systems scale sequential computation through extra tokens. Tilde Research ended the beta phase of its OneLayerDeeper competition run with CoreAutoAI. Mark Saroufim congratulated top contributors and set the final submission deadline for Aug 31. The contest supplies one file and one H100 under a fixed model-state ceiling.