Keyword Harnesses Fail Open: A Cheap Diagnostic Ladder for Tool-Use Claims in Small Language Models
Keyword benchmarks falsely credit a 1.1B model with tool use; cheap diagnostics and targeted SFT expose and repair it.
Stories per day · 30 days
1 RSS
3 stories, newest first · one card per story
Stories tagged #false-positive
COGNIT-Guard: Calibrated Standalone Direct-Decision Guardrails with Heterogeneous CPU-NPU Confidence Cascading under Explicit Latency and False-Positive Constraints
COGNIT-Guard cascades a CPU gatekeeper to a 322M model for low-latency prompt safety screening.
Microsoft fixes bug behind ‘Defender Antivirus is turned off’ alerts
Microsoft's Defender Antivirus update 4.18.26080.4 fixes false 'Defender Antivirus is turned off' alerts affecting all supported Windows client and server versions.
4