On September 7, Fortune and FinanceBuzzBlog confirmed that David Silver’s Ineffable Intelligence has added six cofounders; the company confirmed and declined further comment as LinkedIn profiles updated.
From DeepMind: Chris Apps, Wojciech Czarnecki, Lasse Espeholt, and Junhyuk Oh (Apps, Czarnecki, and Oh on AlphaStar; Espeholt on MetNet). Also Alexandre Laterre (ex InstaDeep research head) and Heather Gorham (ex Flying Fish partner). This is not “six more VPs”—it is welding an RL bench into cofounder chairs.
[1][2]How the roster splits the work
Public role cuts are clear: Oh on RL; Czarnecki on science; Apps on mission acceleration; Espeholt on compute, infra, and engineering strategy; Laterre on research engineering; Gorham across compute, fundraising, and ops.
The capital backdrop was already known: an April $1.1B seed at a $5B valuation (among Europe’s largest seeds), led by Sequoia and Lightspeed, with Flying Fish writing the first check; a Google Cloud compute partnership announced in July. Companies House fills in the governance history: Geordie Rose set the company up in November 2025; Silver became a director in January 2026 after leaving DeepMind; Rose is now a founding advisor, not a cofounder.
[1][2]Why “neolab,” not just another model shop
Coverage frames Ineffable against other neolabs—Thinking Machines, SSI, and peers. The path is RL-first toward superintelligence / “superlearners”; Oh’s automated RL algorithm-discovery angle puts “AI discovering RL” into the org chart, not just a blog slogan.
Six cofounder chairs matter because research, engineering, compute, and fundraising are not late-stage departments—they sit as peer seats from day one. The AlphaStar bench is not nostalgia; it is reusable muscle for multi-agent and large-scale RL training.
[1][2]What it means for the DeepMind diaspora
This is another translation of a lab talent table into a startup equity table. Europe’s seed size already showed capital will pay a premium for an RL-first story; six cofounders show Silver intends to lock the core bench with equity density rather than advisor float.
For people still inside big labs, the signal is hard: beyond headline fundraising, the scarce asset is bundling science–engineering–compute–fundraising into one cofounder map. Who finishes that map first looks more like a neolab that can survive the next compute winter.
[1][2]Commentary
I read this as an equity-encoded research program, not another star-exit gossip cycle. Putting AlphaStar and MetNet people in cofounder chairs is an open bet: the path to superlearners still runs through reinforcement learning, and through AI discovering better RL. The company said little; the roster said the roadmap.
Watch two things, not the sixth surname in the headline: whether automated RL discovery becomes the company mainline rather than a personal research hobby, and whether the Google Cloud partnership turns compute from a press line into a schedulable training queue. Weld the chairs as tightly as you like—without a compute and algorithm loop, a neolab is just a pricier contacts list.
[1][2]