Well, Connes is entirely right that the old USSR way of doing science is much more efficient. The way young people have to make their career in physics in these days is simply dreadful. Concerning the students which are educated in certain big schools and have to comply with the corresponding research topics : by the time they figure out that what they have learned - through hard work - is not corresponding at all to the promises of the marketing slogans which appealed to them in the beginning, they are reluctant or simply too lazy to give up this ``high intellectual property´´ and try something new all together. Moreover, the publication pressure on these people is enormous and they deliberatly avoid the interesting questions in physics (working on them is bad for your career). They usually work in safe, mathematical oriented topics within socially accepted paradigms and simply wait until a better physical idea drops out of the sky.
Concerning those who want to do original research: they are recommended to go to mathematics institutes which is not easy for them either since their work is usually really on the borderline between applied mathematics and theoretical physics.
In short, theoretical physics is going to kill itself. In these days, you are supposed to be able to chit-chat about all the latest fashions and you have to go to all these meaningless seminars where the next results about non physical stringy extremal black holes are present with nifty graphics and much poehaa. A scientific branch can only make progress when you can leave good young people in peace for about 5 to 10 years in which they have time to study and write about a truly good problem in which they are free to exploit their *own* ideas. In physics however, you have to *quickly* emerge yourself in some small details of one particular subject without having a good understanding how it relates to the bigger picture, or even if the particular problem itself is meaningful from the mathematical point of view, or wether the physical aspects of the programme at hand are sensible (which they usually are not). Next, you have to produce within two years ``evidence´´ that the whole idea is not just a piece of **** (either computer simulations or some moronic simplified example). Whether your results contribute to the *understanding* of the problem in any significant way is not that important, in the first place they have to *support* the programme and your next job application. In other words, they have to generate credibility and not that much conclusiveness (a negative (more or less) conclusive result would actually be worse).
The most beautiful example of this ``new´´ way of doing science can be found in the quantum gravity business. There, people zap from one approach to the other which clearly proofs the lack of a credible programme and the non - existence of a shelter environment which allows the SAME people to work on ONE problem for a decade or even more. This is not going to bring anything : for example in LQG the Hamiltonian constraint is still an open problem and as far as I know one did not even show yet whether in the master constraint programme (four years old) one is CLASSICALLY able to find a representation of the spacetime diffeomorphism group (ie. some equivalent of the Dirac algebra). It seems to me one should first solve that kind of problems before one starts exploiting this line of thought at the quantum level (but who am I to say this ? :-) )
