Anthropic releases Mythos 5.2 and Fable 5.2 amid RSI concerns
Fable input pricing falls to $8 per million tokens, with 30% faster responses and enterprise zero-data-retention support.
Stories per day · 30 days
6 RSS
Relatedbenchmarks4agent-evolution2alignment2anthropic2evaluation2language-models2llm-agents2model-security2oversight2post-training2robotics2agent-evaluation1agent-harness1agent-harnesses1agents1agent-skills1agent-swarm1benchmark-evaluation1
59 stories, newest first · one card per story
Stories tagged #recursive-self-improvement
InfraRSI harness applies recursive self-improvement to LLM training
On four dense and MoE workloads, the authors report up to 2.75x speedup and mean MFU rising from 17.4% to 30.0%.
Agent harnesses enable recursive self-improvement without model weight updates
Evolved agents gain new tasks but lose covered ones, with reliability gains often raising deployment costs
Agentic Video Generation and Recursive Self-Improvement Studied
Agent harness coordinates video models for agentic generation, world modeling, and recursive self-improvement in multi-shot video production.
Survey frames risks of recursive self-improvement in large-model systems
The authors define RSI and organize trustworthy improvement around validity, constraints, security, and oversight.
Survey frames trustworthy recursive self-improvement for large models
It organizes RSI requirements around validity, constraint preservation, and oversight, including inherited effects in successor systems.
Skill-infinity makes agent skill optimization self-improving
On 83 SkillsBench tasks the authors report 60.7% pass rate, above curated skills and SkillEvolver.
Jack Clark moves self-improving AI forecast forward to 2027
TasteVal gives Anthropic's Opus 5.5 a 2.3 compute multiplier, but it does not show models already improving themselves.
1
ParetoRSI-Gym evaluates multi-domain recursive self-improvement
Claude Code with Opus-5 raised equal-weighted generalization scores 9.2% relative, or 2.3 absolute points, over the starting models.
RSI Arena studies autonomous post-training of Nemotron with human feedback
The paper reports GPT-6 Astra closed a median 76% of the base-to-Instruct gap, and eight of ten models improved after a second stage.
DeepEvolve agents self-improve by forecasting live events
Authors say the forecaster led FutureX for three weeks, with rating gains up to 195%.
Paper maps recursive self-improvement pathways for physical robots
Reviewed physical studies show behavioral and procedure gains, but none isolate improvement that comes from recursion itself.
Paper studies physical recursive self-improvement in VLM policies
PhysRSI-eval scores VLMs on 18 manipulation tasks under a fixed budget and tests whether public scores predict later physical gains.
HarnessPro evolves agent harnesses for better performance at lower cost
With GPT-5.5 and Codex, heavy mode gained 2.0 pass-rate points and cut per-task cost 26.1% versus RHO.
CoEvo-RSI aligns evolving agent harnesses with model checkpoints
Authors say SFT on self-sampled trajectories makes Qwen3.5-9B beat larger models on two agent benchmarks.
EvalResearchBench tests agents that design their own evaluators
The best evaluators ordered about 75% of candidate pairs like the targets, below a 91% disagreement ceiling.
LM2LM uses a research world model to train improvement proposals
Held-out runs show Qwen3.8-27B beating GPT-6-sol given 300 records, at 33% lower data-generation cost.
SwarmRSI: Recursive Self-Improvement through Evolving Scientific Agent Swarm
SwarmRSI reports state-of-the-art recursive self-improvement for scientific agent swarms on three research benchmarks.
Self-Evolving Agent Harness
Paper formalizes persistent self-evolution of agent harnesses with fixed weights and validation bounds.
Sakana AI's self-improving model gets better at solving and judging, but can't catch its own blind spots
Sakana AI and UC Berkeley's MASS method lets Qwen3.6-27B self-improve at solving, judging, and workflow design.
RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses predictive world models to recursively improve policies across code, GUI, lab, and chemistry tasks.
RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses world models to select actions and updates, improving code repair, GUI tasks, and lab planning.
ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
ReSI recursively red-teams and retrains models, cutting mean X-Teaming success from 86.01% to 31.45%.
Embodied Turing Machines: Stateful Code for Robot Recursive Self-Improvement
Code-Only-as-Policy makes robot decisions in code, reaching 70.24% success on 42 RoboDojo tasks.