Research 2000 for Daily Kos (11/17-19, likely voters, 11/10-12 in parentheses):
Jim Martin (D): 45 (46)
Saxby Chambliss (R-inc): 51 (49)
(MoE: ±4%)
The newest R2K poll of the Georgia senate runoff shows Jim Martin falling further behind incumbent Saxby Chambliss at 51-45, but Markos points out an interesting detail: the previous week's poll, which was 49-46, found that the respondents had voted 49-47 for Chambliss in the general election (which closely mirrored the actual 49-46 result). This sample, however, finds that the respondents voted 52-44 in the general, suggesting a more Republican-leaning batch.
The fluctuation may also reflect some difficulty in pinning down who exactly fits the 'likely voter' mold for a runoff election: is it anyone who voted in November, or only someone who votes with regularity? Martin probably owes his close showing in the general to high turnout from young and/or African-American voters with sporadic voting track records but who were highly motivated to come out for Obama... and Obama seems hesitant to expend too much political capital on this race right now. On the other hand, with Republicans probably demoralized and undermotivated to return to the polls, and with a lot of Obama staffers deployed to the state, it's at least possible Martin could pull out the upset through disparities in ground game.