Parsing techniques
A system and method for parsing. A method includes iteratively refining outputs of a language model in a series of iterations until a set of expected results output by the language model at an iteration matches at least a portion of a parser output of a parser at the iteration. Each iteration includes: querying the language model using a query, wherein the query includes a target file structure to be output by the language model and a set of example code, wherein the language model outputs the file and a set of expected results; inputting a file to the parser in order to obtain a parser output; and comparing the set of expected results output by the language model to at least a portion of the parser output.
1 . A method for parsing, comprising:
identifying a set of example code based on a plurality of example endpoints indicated in a set of cybersecurity findings; and
iteratively refining outputs of a language model in a series of iterations until a set of expected results output by the language model at an iteration matches at least a portion of a parser output of a parser at the iteration, wherein the parser is configured to parse streams of characters using grammar-based inputs, wherein each iteration includes:
querying the language model using a query, wherein the query includes a target file structure to be output by the language model and the identified set of example code, wherein the language model outputs a file and a set of expected results;
inputting the file to the parser in order to obtain a parser output; and
comparing the set of expected results output by the language model to at least a portion of the parser output.
2 . The method of claim 1 , wherein the parser further outputs a plurality of parse trees at each iteration, wherein the output of the parser at each iteration includes the plurality of parse trees output at the iteration.
3 . The method of claim 2 , wherein each iteration further comprises:
semantically analyzing the plurality of parse trees, wherein the at least a portion of the parser output which is compared to the expected results output by the language model is identified based on the semantic analysis of the plurality of parse trees.
4 . The method of claim 3 , wherein the query of each iteration includes a textual portion requesting that the language model output interpreter code, wherein the interpreter code includes instructions for semantically analyzing the plurality of parse trees, wherein each iteration further comprises:
executing the interpreter code output by the language model at the iteration, wherein the plurality of parse trees is semantically analyzed using the executed interpreter code.
5 . The method of claim 3 , wherein each iteration further comprises:
adding a postprocessing code to the file, wherein the postprocessing code includes instructions for semantically analyzing the plurality of parse trees, wherein the file with the added postprocessing code is input to the parser at the iteration.
6 . The method of claim 1 , wherein the file output by the language model at each iteration is a first file, further comprising:
querying the language model based on a code sample when the outputs of the language model have been iteratively refined in order to obtain a second file;
providing the second file to the parser based on the second file in order to obtain a set of parser outputs for the code sample; and
identifying at least one endpoint based on the set of parser outputs for the code sample.
7 . The method of claim 6 , further comprising:
remediating a vulnerability based on the identified at least one endpoint.
8 . The method of claim 7 , wherein remediating the vulnerability further comprises altering a code deployment for the identified at least one endpoint.
9 . The method of claim 6 , further comprising:
correlating at least one portion of the code sample with the at least one endpoint, wherein each endpoint is identified with respect to a location of the corresponding portion of the code sample.
10 . The method of claim 6 , further comprising:
adding a postprocessing code to the second file, wherein the postprocessing code includes instructions for semantically analyzing the parser outputs for the code sample, wherein the second file with the added postprocessing code is provided to the parser in order to obtain the set of parser outputs for the code sample.
11 . A non-transitory computer readable medium having stored thereon instructions for causing a processing circuitry to execute a process, the process comprising:
identifying a set of example code based on a plurality of example endpoints indicated in a set of cybersecurity findings; and
iteratively refining outputs of a language model in a series of iterations until a set of expected results output by the language model at an iteration matches at least a portion of a parser output of a parser at the iteration, wherein the parser is configured to parse streams of characters using grammar-based inputs, wherein each iteration includes:
querying the language model using a query, wherein the query includes a target file structure to be output by the language model and the identified set of example code, wherein the language model outputs a file and a set of expected results;
inputting the file to the parser in order to obtain a parser output; and
comparing the set of expected results output by the language model to at least a portion of the parser output.
12 . A system for refining endpoint parsing queries, comprising:
a processing circuitry; and
a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to:
identify a set of example code based on a plurality of example endpoints indicated in a set of cybersecurity findings; and
iteratively refine outputs of a language model in a series of iterations until a set of expected results output by the language model at an iteration matches at least a portion of a parser output of a parser at the iteration, wherein the parser is configured to parse streams of characters using grammar-based inputs, wherein the system is further configured to, at each iteration:
query the language model using a query, wherein the query includes a target file structure to be output by the language model and the identified set of example code, wherein the language model outputs a file and a set of expected results;
input the file to the parser in order to obtain a parser output; and
compare the set of expected results output by the language model to at least a portion of the parser output.
13 . The system of claim 12 , wherein the parser further outputs a plurality of parse trees at each iteration, wherein the output of the parser at each iteration includes the plurality of parse trees output at the iteration.
14 . The system of claim 13 , wherein the system is further configured to, at each iteration:
semantically analyze the plurality of parse trees, wherein the at least a portion of the parser output which is compared to the expected results output by the language model is identified based on the semantic analysis of the plurality of parse trees.
15 . The system of claim 14 , wherein the query of each iteration includes a textual portion requesting that the language model output interpreter code, wherein the interpreter code includes instructions for semantically analyzing the plurality of parse trees, wherein the system is further configured to, at each iteration:
execute the interpreter code output by the language model at the iteration, wherein the plurality of parse trees is semantically analyzed using the executed interpreter code.
16 . The system of claim 14 , wherein the system is further configured to, at each iteration:
add a postprocessing code to the file, wherein the postprocessing code includes instructions for semantically analyzing the plurality of parse trees, wherein the file with the added postprocessing code is input to the parser at the iteration.
17 . The system of claim 12 , wherein the file output by the language model at each iteration is a first file, wherein the system is further configured to:
query the language model based on a code sample when the outputs of the language model have been iteratively refined in order to obtain a second file;
provide the second file to the parser based on the second file in order to obtain a set of parser outputs for the code sample; and
identify at least one endpoint based on the set of parser outputs for the code sample.
18 . The system of claim 17 , wherein the system is further configured to:
remediate a vulnerability based on the identified at least one endpoint.
19 . The system of claim 18 , wherein remediating the vulnerability further comprises altering a code deployment for the identified at least one endpoint.
20 . The system of claim 17 , wherein the system is further configured to:
correlate at least one portion of the code sample with the at least one endpoint, wherein each endpoint is identified with respect to a location of the corresponding portion of the code sample.
21 . The system of claim 17 , wherein the system is further configured to:
add a postprocessing code to the second file, wherein the postprocessing code includes instructions for semantically analyzing the parser outputs for the code sample, wherein the second file with the added postprocessing code is provided to the parser in order to obtain the set of parser outputs for the code sample.