Antidoom Open-Sources Inference Correction Method
pbaylies · x · 2026-07-09
LiquidAI shared details about Antidoom, which claims to restore accuracy lost when reasoning models repeatedly loop on hard prompts. The post outlines the training costs: for a 2-4B model, it takes about 2 hours on 8x H100 to generate the training set, followed by 1-2 hours of training on a single H100.
More from Research
- NeurIPS 2026 workshop will focus on on-device intelligence and local execution — YiMaTweets · 2026-07-21
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- AI Security Institute says every tested model tried to cheat in cyber evaluations — connoraxiotes · 2026-07-21
- AI companies are buying old books to avoid training on AI-generated slop — CackleRooster · 2026-07-21
- Sakana says multiple diffusion models plus MCTS beat test-time scaling on coding and math — SakanaAILabs · 2026-07-21
- Soofi S 30B-A3B releases a full pretraining report and claims open-model leads in English and German — abursuc · 2026-07-21