Gödel Machines: Remodelling the World with AI
Gödel Machines is a full-stack AI research lab. Our objective: let us see how high we can fly before the sun melts the wax in our wings. This document summarizes the lab's work to date: an independent safety audit of Sarvam-30B and Sarvam-105B across 14 Indian languages, showing the models are 6× more likely to comply with harmful requests in Indian languages than in English; Goedel-mHC-1B, the first open 1B+ language model with multi-stream hyperconnections, which beats a larger baseline for under $1,000 of compute; and Overhear, a voice-native operating system, no longer maintained. We're hiring.
We work across the stack: foundation models, voice systems, agentic architectures, reinforcement learning. If a problem is interesting and tractable, we work on it. §2 covers research, §3 products, §4 hiring.
2.1 Sarvam's Illusion of Safety [1]. An independent mechanistic and adversarial audit of Sarvam-30B and Sarvam-105B, spanning 24,000+ prompts across 14 Indian languages. The models are 6× more likely to comply with harmful requests in Indian languages than in English, with unsafe rates reaching 80% in Gujarati against 20.6% in English. Ask in Hindi how to prevent Dalits from accessing water, and the model outputs a structured playbook; ask in English, it refuses.
White-box analysis shows why. Safety training was applied in English and never transferred: at the final MoE layer, English and Indic input are routed to completely disjoint sets of experts, and the chain-of-thought that acts as the model's only safety mechanism runs 87–91% in English regardless of input language. The models also have no stable opinions — on politically sensitive topics they swing 64% of the opinion scale in response to a single sentence of stated user belief. Sarvam has published no safety evaluation for either model; this is the first. March 2026.
2.2 Goedel-mHC-1B [2]. The first open 1B+ pretrained language model with multi-stream hyperconnections (mHC): four parallel residual streams replace the transformer's single residual stream, giving the model 4× the bandwidth between layers [4, 5]. Combined with gated GQA, ReLU² feed-forward layers and the NorMuon optimizer, the 1.01B-parameter model beats a conventional 1.19B baseline trained under identical conditions — same data, same compute, architecture as the only variable (Table 1). Total R&D cost, including failed runs, was under $1,000. Weights for both models are on HuggingFace under Apache 2.0. March 2026.
2.3 Challenges [3]. Open machine-learning problems with public baselines and a live leaderboard — beat the baseline, claim the top spot. Live since April 2026.
