Computers and openings

They say computers don't handle the opening well and therefore rely heavily on tuned opening books to lookup moves.

Can someone name a couple openings where computers mess up if they have no book and must rely solely on the search / evaluation algorithm?

Thanks

Running without its book engaged, Junior 8 chooses opening moves like d4, Bf4, Nc3, Qd3.  Those moves aren't necessarily bad but they certainly aren't theoretically optimal.

they are in some lines.

They sometimes play moves like Bd3 when they have the pawn on d2. I remember Rybka 2 had a similar weakness (or at least discussions about the weakness) with playing Nc3 when the c pawn hasn't moved yet in d4 openings.

Other than that, when I go through my games, Firebird and Rybka's suggestions seem very natural and consistent with theory.

There was a match where Rybka had only 3 openings moves by book against GM Dzindzi, and it handled the openings very well. So although they aren't on par with decades of human analysis (which is engine assisted for the last 15 years anyway), they are pretty strong in openings too.

They hate certain openings, like the winawer french for black. Fritz 10 also thinks that the sicilian gives white a big advantage. It probably takes into account white's development and doesn't see how black can compensate for it, but that's because black's chances are stronger in the long term. In the winawer, the fact that white can give up the center to triple his pawns but win one pawn, his bishop pair, and extra space makes it think black is losing, because those features are programmed into the engine when it's evaluating along with tactics of course, but a lot of times there aren't immediate tactics in this opening.

I would say they're inconsistent, sometimes they know what they're doing, sometimes in weird positionally imbalanced position they don't know what's really going on.

Without a book, engines would at least have to fight for a draw in many lines of the King's gambit. For instance, Rybka has to play out 8 or 9 moves to actually begin to "understand" white's piece sacrifice.

Lemme tell ya, computers hate the traxler counterattack.

Yeah, there are some variations the computer thinks the position is winning for white by +7.5, but you find that by move 15 black is equal or is winning.