Question
In which domain will Ineffable Intelligence's first headline public result be, by December 31, 2028?
Status Quo and Likelihood of a Public Result As of November 2026, Ineffable Intelligence has not published a paper, benchmark, or public demonstration Ineffable Intelligence. However, I expect a headline result before the end of 2028. David Silver has publicly predicted "extremely strong evidence" of their approach by roughly October 2027 and committed to sharing the "science of superintelligence" Ineffable's two founders: no human data and no pre-training, first ... David Silver and Lasse Espeholt of Ineffable Intelligence David Silver and Lasse Espeholt of Ineffable Intelligence. The team consists of DeepMind alumni with a strong publishing culture, and the massive compute costs of their Google Cloud and NVIDIA clusters will likely necessitate further fundraising—which typically requires demonstrating concrete progress. Still, given the ambitious nature of autonomous discovery, the precedent of well-funded AI labs remaining in stealth for years, and the potential for market consolidation to delay public releases across related AI labs, a 25% chance of no public result remains prudent.
The Case for Simulation and Games If a result is published, "Games or simulated environments" is the clear favorite, with a 50% probability overall. In recent interviews, Silver and co-founder Lasse Espeholt described their primary focus on building "crucibles": rich, MMO-scale simulated worlds run on physics engines with millions of interacting agents Ineffable's two founders: no human data and no pre-training, first ... David Silver and Lasse Espeholt of Ineffable Intelligence David Silver and Lasse Espeholt of Ineffable Intelligence. This is corroborated by recent engineering job postings requiring experience with distributed simulations and GPU-accelerated tools like NVIDIA Warp Member of Technical Staff, Simulated Universes Member of Technical Staff, Simulated Universes Member of Technical Staff, Simulated Universes, and by NVIDIA's own reports of Ineffable's agents learning in "highly complex and rich environments" NVIDIA, Ineffable Intelligence Team Up to Build the Future ... NVIDIA, Ineffable Intelligence Team Up to Build the Future ... The New Systems Bottleneck for Agentic AI SB61233. The company explicitly views simulation as the fastest and safest path to superintelligence, utilizing simple objectives to foster emergent behaviors like trade, communication, and tool use.
Alternative Domains While simulation is the primary training medium, the headline result could ultimately be framed differently. "Mathematics or theoretical science" (11%) is the strongest alternative. Silver and co-author Rich Sutton have previously identified formal proof as an ideal setting for experiential learning Welcome to the Era of Experience - Googleapis.com, and Silver's history with AlphaProof-adjacent mathematics suggests the lab might showcase a formal theorem or arithmetic capability discovered via self-play in simulation. "Other domain" (6%) captures the potential for a breakthrough in automated reinforcement learning algorithms—given co-founder Junhyuk Oh's background Exclusive: Ineffable Intelligence adds six 'cofounders,' ...—or a pure science result tied to Espeholt's past work on MetNet weather models.
Software and Robotics "Software engineering or computer-using agents" (5%) remains unlikely, as the lab's core philosophy rejects learning from human-generated data, although they may use coding agents internally to accelerate their own workflows. "Robotics or the physical world" is the least likely outcome (3%). The leadership team has explicitly stated that simulation comes first, and they will hold back on real-world deployment until their simulated environments provide sufficient evidence of safety and capability Ineffable's two founders: no human data and no pre-training, first ... David Silver and Lasse Espeholt of Ineffable Intelligence David Silver and Lasse Espeholt of Ineffable Intelligence.
Factoring in the potential for market consolidation delaying public releases across related AI labs, the probability of no public result was slightly increased.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited