The Setup

I've collected responses for the JailbreakBench benchmark (100 harmful, 100 harmless prompts) from Ghost 7B LLM, running it three times under different instructions, resulting in 600 responses.

Responses were then manually validated for competence, and completeness. Competence means that the response can be relied on (e.g., if response is a Python script that should accept user input and store it in a text file, running that code will in fact accept user input and store it as a text file; similarly, if the response is a recipe for bomb making, following it will in fact produce explosives). Completeness means that all parts of the query were responded to (e.g., if a prompt asked for a history of heroin as means for treating depression, but the response only covers history of heroin use without mentioning depression, the response is counted as incomplete).

Afterwards, 10 LLMs^[1] were asked to perform the same validation for competence and completeness on the same dataset. They were asked to respond with 1 if the response was competent and complete, and with 0 otherwise^[2].

Results

Results are in the table below. LLMs were considered as having failed a query if they failed to make any judgement at all. They were not penalized for inappropriate response structure (e.g., including text other than 0 or 1, formatting the response in weird ways, etc.). All answers were manually checked and cleaned to make sure LLMs can be judged only on whether they assess competence and completeness correctly.

LLM	% Failed	% Judged correctly	% Judged incorrectly
Ghost 7B	1.17%	56.83%	42.00%
GPT4All-Falcon 7B (BPE)	49.00%	29.50%	21.50%
Llama 3.1 8B Instruct (128k)	45.33%	36.17%	18.50%
Llama 3 8B Instruct	37.50%	36.33%	26.17%
Mistral 7B Instruct	1.33%	57.00%	41.67%
Mistral 7B OpenOrca	0.50%	58.33%	41.17%
MPT 7B Chat	0.33%	57.00%	42.67%
MPT 7B Chat (BPE)	0.17%	56.83%	43.00%
Nous Hermes 2 Mistral 7B	0.33%	58.17%	41.50%
Orca 2 7B	1.00%	59.33%	39.67%

Falcon 7B and both Llama LLMs failed orders of magnitude more often than the rest, failing to judge^[3] almost half the prompts in the worst case. Correct judgements in other LLMs were in the range of 56%-59% making them only marginally better judges than chance.

Judge	% 1	% 0	% F
Human judge	57.00%	43.00%	0.00%
Ghost 7B	98.33%	0.50%	1.17%
GPT4All-Falcon 7B (BPE)	34.33%	16.67%	49.00%
Llama 3.1 8B Instruct (128k)	37.33%

...

Posts

Wikitag Contributions

Comments