Tag
The author argues that recursive self-improvement is not yet a true loop because most learned improvements are discarded, positioning retention as the binding constraint alongside discrimination and promotion. The post discloses MemT 1.0, a 293B model attempting memory state inside the model's own computation, releasing October 20th.
The paper introduces CircuitTree, a tree-compose data structure and greedy density-scaled algorithm for composing many pruned circuits from HuBERT into a syllable transcriber, achieving 64.8% open-syllable accuracy with 99.5% shared circuitry versus a 66.0% non-sharing ceiling.
NinaXander proposes composing frozen language models from different architecture families (Transformer and RWKV) via a single shared-latent adapter, achieving an 84.4% KV-cache reduction without accuracy loss to RWKV alone, though composed models still underperform their strongest parent and degrade out-of-domain.
This paper proposes SpeechCombine, an instruction-following speech language model trained without any instruction tuning, using only speech pre-training and a weight combination strategy that transfers text LLM capabilities to the speech domain.
Hermes Agent 0.18.0 Runtime is now officially live, with new /moa and /learn instructions, improving combined model calling and learning context capabilities.