Reviews: G2SAT: Learning to Generate SAT Formulas

That these types of techniques are really essential for "developing better and faster SAT solvers" is quite questionable, since the SAT community has been making quite good progress in pushing solvers forward and still are, and the benchmark situation has considerably improved over the years. Perhaps the biggest issue with the work is in that it does not really seem to scale. The authors use the 10 *smallest* "real-world" instances in their experiments. Small instances are really not interesting from the perspective of SAT solver development, as most real-world instances are quite huge. In my opinion the authors would need to address at the very least the scalability issues to warrant publication at a major venue. Regarding the comparison of graph statistics of the generated instances, I do not see what to make of these.