Seven things:

• There’s a post on speculative contacts by Paul Firth, a way of simplifying and stabilizing collision detection that has been used in Little Big Planet. Particularly nice is that demos are built into the page, so you can try the various methods out and see the problems and performance for yourself. This author has followed up with “Collision Detection for Dummies“, a great overview, and “Physics Engines for Dummies“, again with interactive demos.
• The Gamedev Coder Diary has a worthwhile summary of the current state of deferred shading vs. deferred lighting (aka “light pre-pass”) techniques, discussing problems and strengths of each.
• The CODE517E blog has had a number of good posts lately, including an article on deferred rendering myths, another on stable cascaded shadow maps, an accumulation-buffer-like way of making super-high resolution images for printing (with some worthwhile analysis of problems it engenders with mipmap sampling and with view shifting – fun to think about), an extensive rundown of programming languages for videogames, and a summary of tools he uses (quite the long list – I’m still working through those I hadn’t seen before).
• On the topic of languages, Havok put together a page collecting the Lua tutorial talks at GDC 2011.
• The Boeing 777 model (almost 400 million polygons) ray traced at interactive rates on a consumer-level PC, using CUDA. CentiLeo is an out-of-core GPU ray tracer, see this page for some of the slides from the (rather long) video. That said, don’t be fooled by the start of the video: those sequences are generated at 15 seconds a frame and played back at 60 FPS (so 500-1000x from being real-time). Still, the preview mode is indeed interactive, and the Boeing is a huge model. On the other end of things, here’s a fun demoscene ray trace. By the way, Ray Tracey’s blog is good for keeping up on new ray tracing videos and demos and other related topics.
• A poster accepted to SIGGRAPH 2011 by Ohlarik and Cozzi gives a clever little method of properly drawing lines on surfaces for GIS applications. It converts lines to “walls”, then marks those pixels where there is a visibility change of the wall (i.e., one pixel of the wall is visible, a neighboring pixel is not), with a correction for terrain silhouette edges. One more trick for the bag.
• More about the look and feel of games than the technical nerdy stuff I cover here, Topi Kauppinen’s blog pointed me to Susy Oliveira’s sculptures, which are pretty amusing (finally, perfect models for 3D web browsers). There have been similar works by other artists (e.g. Eric Testroete’s head), but the more the merrier.