AI Explained
Plain explanations of trending AI concepts, with live visualizations.
Beyond Instruction Override benchmark — Task-tampering injection — What does it mean?
Prompt injection can bend a decision the task already needs, and override-tuned defenses miss it.
Signed MCP tools can still leak — Germline execution — What does it mean?
Fork a data-free copy of each MCP tool per call, handing it only what its destination may see, so a signed Trojan has nothing to leak.
Targeted guidance in agent RL — Failure-gated self-distillation — What does it mean?
Distill judge feedback only on failing agent turns, and keep RL rewards, so agents learn new behavior without regression.
VeriHarm-Bench — Harmful success in coding agents — What does it mean?
Coding agents can pass their tests and still break things; VeriHarm adds a hidden harm test to measure it.
LatentTriad screens malicious agent skills from one prefill — Prefill-state skill screening — What does it mean?
Flags malicious agent skills from one prefill's hidden states, with no output tokens, tool calls or skill code run.
Strategist evolves reusable inference strategies at 39% lower cost — Inference strategies as executable code — What does it mean?
Strategist saves inference strategies as reusable code, so each new task builds on past ones and costs 39% less.
Cross-framework MCP tool-poisoning study — Canary exfiltration oracle — What does it mean?
A study of MCP tool poisoning uses a canary server to prove which attacks really leaked a secret.
ShallowDefend paper — Assess, Decide, Respond safety commitment — What does it mean?
ShallowDefend has a model commit to an explicit, rule-checkable safety decision before replying, defending against multi-turn jailbreaks.
ASTC paper — Agentic soft-token compression vs hard-token context rewriting — What does it mean?
ASTC folds tool output the model has already absorbed into soft tokens, keeps exact values as text, and reuses the shorter KV cache.
Ask on Demand — Pull-based peer retrieval — What does it mean?
Agents in a team ask a peer for the specific information they need, and a small learned gate decides whether the question is worth the cost.
Sandman paper — Pooled persistent sandboxes for agent RL — What does it mean?
Sandman gives each agent-RL rollout a persistent Kubernetes pod and weighs on-demand, warm-pool and reused pods by cost.
Heads of Security — per-head attention gates — What does it mean?
Heads of Security gates each attention head to make an LLM less likely to obey untrusted tool output.