Opus 5.5 is a milestone model that should be etched into hardware and sold on chips with 10-100× faster inference. It would already serve 99.99% of the population and use cases.
Key points
Milestone: model should be etched into hardware and sold on chips with 10-100× faster inference.
Coverage: would already serve 99.99% of the population and use cases.
Sources
LOW-FI v0.57 – Unity 6 upgrade, Le Circus race
LOW-FI v0.57 moves from Unity 2019 to Unity 6 and adds a full race career at Le Circus. The build now supports VR via OpenXR and non-VR automatically. Loading screens now show an animated progress bar.
Key points
Engine Upgrade: Unity 2019 → Unity 6, VR runs through OpenXR with a single executable.
Le Circus race career: beat four racers sequentially, then all together, then a secret challenger.
Sources
- Original post
- v0.57 Back on Track - LOW-FI by Anticleric - Linked page
LOG AWAY Halloween DLC released, spooky makeover
LOG AWAY Halloween DLC released, adding a spooky makeover to the game. The DLC is available now on Steam.
Key points
DLC: brand new Halloween DLC is OUT NOW
Platform: Play now on Steam
Sources
Virtual Matter new renderer rolling out now
Virtual Matter announces its entirely new renderer is rolling out now. The post invites users to ask the built-in agent and mentions cel-shaded microvoxels in prompts, not polygons.
Key points
New renderer rolling out now in Virtual Matter
Built-in agent can be asked for rendering requests
Sources
Satoshi executable matches source, used Windows XP with MinGW
Satoshi executable matches the published source code exactly. The build was done on Windows XP using MinGW. The binutils toolchain was upgraded partway through development.
Key points
Executable people downloaded matches Satoshi's published source code exactly.
Satoshi used Windows XP with MinGW and upgraded binutils halfway through.
Sources
Messinator robot scores after 140-year self-play training
Messinator robot dribbles past a defender and scores, using control behaviors transferred zero-shot from 140 years of simulated self-play. Developed by SkildAI. A quoted tweet says they trained a robot to play football via self-play for 140 years in a virtual World Cup.
Key points
Demonstration: humanoid robot dribbles past a defender and scores.
Training: control behaviors transfer zero-shot from 140 years of simulated self-play.
