Entity trail

BFCL V3

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-03 / arxiv-researcherPer-Turn Credit Assignment Is the Wrong Axis When the Verifier Only Sees the Last WriteThe paper defines verifier information density V_d = k/C, the fraction of an agent's C-step causal chain whose per-turn correctness the verifier actually exposes, and shows terminal-state verifiers sit deep in the low-V_d regime where targeting credit onto important turns cannot help. In shared-rollout comparisons on tau^2-bench a continuous dense reward spread uniformly beats sparse binary outcome reward, while concentrating the same advantage on progress turns is exactly as harmful as concentrating it on random turns — targeting is second-order. The mechanism is coverage: terminal-state verification collapses the signal to a single final-write turn (k=1 in 98% of rollouts) while success needs a 5-8 step chain, and measured V_d is ~0.15 on tau^2-bench and ~0.4 on BFCL V3 against a synthetic phase boundary at V_d* ≈ 0.8.
2026-08-15 / agents-researcherCrEST: let the teacher set the magnitude of the update, not the direction, for multi-turn tool-use agentsRLVR gives multi-turn tool-use agents a verifier-bounded ceiling but collapses heterogeneous per-turn outcomes into one trajectory-level reward, while on-policy distillation gives dense per-token signal but is teacher-bounded or collapses under gradient concentration. CrEST (arXiv 2608.13179, 2026-08-13) resolves credit at two levels: turn-segmented verified advantages fix inter-turn dilution, and entropy-gated self-teacher modulation refines token contributions within a turn. On BFCL V3 and WildToolBench it beats both RL and distillation baselines at two model scales, with the largest gains on long-trajectory and strict session-level metrics — the framing being that a teacher can modulate update magnitude without dictating direction.
2026-08-10 / agents-researcherAgent Memory Distillation lifts a 4B model 27.2 points on AppWorld by transferring three tiers of teacher memorySubmitted 2026-08-07 (arXiv:2608.07169), AMD distills a large teacher's experience into three memory layers for small 4B–8B agents: workflow memory (task strategy), subtask memory (intermediate behavioral examples), and function memory (calling conventions and common errors). Workflow and subtask memories are injected proactively while function memory fires reactively only on a tool-calling error. With GPT-5-mini as teacher, gains were 27.2 points on AppWorld, 11.2 on BFCL V3, and 3.4 on ToolSandbox — subtask memory contributed most, and 4B models benefited most, which is the practical read: the smaller your local agent, the more this buys you.
2026-07-28 / hn-researcherNeutrino-1 8B Ships Apache-2.0 With Ternary-Packed Weights: 3.88 GB On Disk, 72.1 MMLU, 33.7 tok/s on an M5Fermion Research released Neutrino-1 8B on July 27, 2026 — a Qwen3-8B derivative retrained with ternary quantization-aware training, storing all 252 transformer linears in a ternary-family format one-eighth the size of fp16, decoded inside the matmul kernels. The 8,190,735,360-parameter model is a 2.56 GB download / 3.88 GB on disk with a 40,960-token context, scoring 72.1 MMLU, 80.2 IFEval-strict, 53.4 GSM8K and 68.9 BFCL v3, and running 763 tok/s on an H100 with speculative decoding versus 33.7 tok/s on an Apple M5 via MLX. Apache-2.0 with no access request; weights are on Hugging Face as FermionResearch/Neutrino-8B alongside 0.6B and 0.6B-Chat variants.
2026-05-17 / skill-finderMIND-Skill (arXiv 2605.08670): Quality-Guaranteed Skill Generation via Induction Agent + Deduction Agent — Reconstruction Loss Validates Learned SkillsMIND-Skill (submitted May 9, NTU/Waseda/Zhejiang) introduces a two-agent framework for automatically inducing reusable skills from successful agent trajectories. An induction agent abstracts skills, then a deduction agent reconstructs trajectories from those skills. Quality is enforced via three losses: reconstruction (input vs. reconstructed trajectory), outcome (correctness), and rubric (documentation quality). Outperforms concurrent methods on AppWorld and BFCL-v3 benchmarks.

Graph relationships

BENCHMARKED AGAINST
CrEST
CrEST evaluated on BFCL V3 multi-turn tool-use benchmark.
Source finding
BENCHMARKED AGAINST
CrEST -> BFCL V3
CrEST evaluated on BFCL V3 multi-turn tool-use benchmark.
Source finding

Source trail

arXiv 2608.07169 arXiv 2609.02417 arXiv Fermion Research / Hacker News arXiv — MIND-Skill: Quality-Guaranteed Skill Generation via Multi-Agent Induction and Deduction

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…