55{% block content %}
66< section style ="text-align:center; ">
77 < img
8- src ="/static/images/clash_new .svg "
8+ src ="/static/images/logos/clash .svg "
99 alt ="CodeClash Logo "
1010 style ="height: 8rem; margin: 1em 0 0.5em 0; "
1111 />
7676
7777< section class ="hthesis ">
7878 < h2 style ="text-align: center; "> Why CodeClash?</ h2 >
79- < p > LMs have gotten pretty good at solving GitHub issues.</ p >
8079 < p >
81- But real software development isn't a series of isolated tasks.
82- It's about understanding higher order goals, making tradeoffs, and evolving a codebase over time.
80+ LMs have gotten pretty good at solving GitHub issues.
81+ < a href ="http://swebench.com/ ">
82+ < img
83+ src ="/static/images/logos/swe-llama.svg "
84+ alt ="SWEBench Logo "
85+ style ="height:1em; margin-left:0.2em; "
86+ />
87+ </ a >
8388 </ p >
84- < p > With CodeClash, we're moving from task-oriented to goal-oriented software development.</ p >
89+ < p > But real software development isn't a series of isolated tasks.</ p >
90+ < p > It's driven by < i > goals</ i > .</ p >
91+ < p >
92+ Improve user retention, increase revenue, reduce costs.
93+ We build to achieve outcomes, not to close tickets.
94+ </ p >
95+ < p > What if AI evaluations reflected this dynamism of real-world software development?</ p >
96+ < p > We introduce CodeClash, an initial effort towards benchmarking goal-oriented software engineering</ p >
97+ </ section >
98+
99+ < hr />
100+
101+ < section class ="hthesis ">
102+ < h2 style ="text-align: center; "> Our Results</ h2 >
85103 < p >
86104 In CodeClash, language models build and evolve their own codebase over multiple rounds.
87105 Each round has two phases: edit, then compete. Models can write notes, analyze past rounds,
@@ -95,7 +113,6 @@ <h2 style="text-align: center;">Why CodeClash?</h2>
95113 in solving problems beyond specific tickets:
96114 </ p >
97115</ section >
98-
99116< div class ="h3columns ">
100117 < div class ="hfeature ">
101118 < span >
0 commit comments