Nvidia Just Backed Ineffable. The AlphaGo Zero Author List Is the Trade.
Sourcing the next wave of AI lab hires means mining founder co-author networks and alumni chats, not LinkedIn titles. Here is the playbook.
On May 13, 2026, Nvidia and Ineffable Intelligence announced an engineering-level collaboration on infrastructure for large-scale reinforcement learning. Jensen Huang framed it as a bet on "superlearners, systems that learn continuously from experience." If you are a recruiter or a founder reading the press release as a chip story, you are missing the actual trade. The deal is a public marker that the most valuable hiring graph in AI right now is the co-author list on a handful of DeepMind RL papers, and LinkedIn's title search cannot see any of it.
David Silver, Ineffable's founder, was DeepMind's RL lead. The AlphaGo Zero Nature paper from 2017 lists roughly 17 named DeepMind co-authors. The MuZero, AlphaZero, and AlphaStar papers extend that ring to a few dozen more. That author list is the org chart for the next two years of hiring at Ineffable. It is also the shopping list every other frontier lab is working from in parallel.
The Nvidia signal, decoded
Ineffable closed a $1.1B seed at a $5.1B valuation in April 2026, the largest seed round ever in Europe. Sequoia and Lightspeed co-led, with Nvidia, DST Global, Index, Google, and the UK Sovereign AI Fund participating. Nvidia's venture arm put in at least $250M. One month later, Nvidia signed an engineering partnership. That sequence is not a coincidence. Nvidia is reserving capacity around a team it expects to absorb a meaningful share of the frontier RL labor pool.
Several former DeepMind staffers are reportedly joining Ineffable's executive team. The phrasing in TechCrunch and CNBC is consistent across stories about labs of this shape: the team is being reassembled, not recruited cold. Anna Goldie at Ricursive Intelligence said it on the record after their $335M raise: "we got the core AlphaChip team back together, and that involved hiring some of our old collaborators." Ricursive's roster pulled from Google, Anthropic, Nvidia, Apple, and xAI. AMI Labs, Yann LeCun's $1.03B venture, recruited senior staff from Meta FAIR, including FAIR's VP for Europe. Periodic Labs has assembled more than 20 researchers from OpenAI, DeepMind, Meta, and Apple.
Why LinkedIn title search is now actively misleading
Open Periodic Labs' public team page. Dzmitry Bahdanau, co-inventor of the neural attention mechanism, is listed as "Team member." That is the title. A boolean for "Senior Research Scientist" AND "attention" AND "Montreal" returns nothing useful. The actual index for this hire was the 2014 Bahdanau et al. paper and his GitHub history, not a job title.
This is the rule, not the exception. Edward Hughes, a DeepMind RL researcher, is in a stealth role with no public title. Founders of pre-launch labs often spend six to twelve months listed as "Researcher" or nothing at all. The senior researcher pool that matters for frontier RL is small. An internal sourcing check across structured profiles surfaces roughly 721 senior research scientists and engineers globally with reinforcement learning as a listed skill. Top current employers in that slice: Google DeepMind (5), OpenAI (4), Meta (2). London, SF, and NYC dominate. That is your denominator. The numerator you actually want, the people quietly leaving those five logos for $5B pre-product labs, is invisible to title filters.
The recruiter's real index is three overlapping graphs:
- arXiv author lists for the founder's last three to five papers.
- GitHub commit graphs on libraries the founder's old team built or maintained.
- Companies House filings, patent assignees, and grant co-investigators that catch stealth founders six months before LinkedIn does.
This is the surface area we built Refolk to query in one shot. You describe the person in plain English, including "co-authored with David Silver on AlphaGo Zero" or "maintains CleanRL," and you get a ranked shortlist with the GitHub, the LinkedIn, and the paper trail joined up. The whole point is to stop translating a research instinct into a brittle boolean.
The unit of hiring is the team
If you identify the founder, the next ten to thirty hires at a frontier AI lab are largely predictable from their last three papers' author lists. That is the lesson of Ricursive, of AMI, of Periodic, and now of Ineffable. Recruiters pitching one role at a time are losing to founders offering "your old team, plus a billion dollars."
Periodic's RL hiring is the cleanest example on the public record. Their RL group is not "Senior MLEs." It is three named ecosystem maintainers: Costa Huang, the creator of CleanRL. Vincent Moens, the creator of TorchRL. Rishabh Agarwal, NeurIPS best paper for offline RL. Three of the world's top RL practitioners, all on the same problem, hired by name. None of them would show up in a title search for "Reinforcement Learning Engineer" with the right filters. All three are obvious if you know the libraries.
The same pattern holds across the cohort. Periodic's 39-person team, scraped from public LinkedIn by David Tsong, breaks down to: Google/DeepMind (6), OpenAI (4), Meta/FAIR (3), Tesla (3), Microsoft Research (1), xAI (1), SpaceX (1). The rest are research-adjacent. The hiring signal is the previous employer plus the co-author edge, not the JD keyword.
If you can name the founder, you can name the next twenty hires. They are on the paper.
The graph leaks through chat groups, not job boards
GeekWire reported in May that Flying Fish Partners, a Seattle VC firm, got into the Ineffable round partly because they had become a recurring presence in an "Ex-DeepMind" WhatsApp group. The same social graph that closes seed rounds closes hires. There is no LinkedIn segment for "Ex-DeepMind WhatsApp." There is no Sales Navigator filter for "recurring guest at the Friday DeepMind alumni dinner in King's Cross."
What there is, if you look hard enough, is residue. GitHub stars on a single founder's repo three days after a stealth announcement. arXiv preprints with a new affiliation footnote. Companies House filings that name three co-founders nobody has tweeted about yet. Patent assignments that change quietly. Evertrace identified 112 DeepMind alumni who have launched or are believed to be launching a startup in the past 18 months, sourced from Companies House, patents, and research grants. 70 of those are in the US, 28 in the UK, with smaller clusters in Spain, Switzerland, Germany, and Canada. That is your real target list for sourcing AI research talent over the next 18 months.
The Ineffable target list is not a secret. Pull the AlphaGo Zero author list off Nature: Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, Yutian Chen, Timothy Lillicrap, Fan Hui, Laurent Sifre, George van den Driessche, Thore Graepel, and Demis Hassabis. Cross-reference against MuZero and AlphaStar. That is roughly 30 to 40 unique names. Now check who has updated a bio in the last 90 days, whose GitHub activity has gone dark, and who has stopped publishing. That is your shortlist, and it is the same shortlist Sequoia and Lightspeed are working from.
Coconut rounds break your comp ladder
Ineffable joins the "pentacorn" club: $5B+ pre-product researcher-founded labs. AMI Labs raised $1.03B at a $3.5B pre-money. Physical Intelligence is reportedly in talks for an $11B valuation, doubling from $5.6B in four months. Humans& closed a $480M seed. The industry has started calling these "coconut rounds."
The comp consequence is sharp. A three-month-old startup can credibly offer pre-IPO equity at a $5B valuation to a senior IC. Big Tech retention packages, which assumed researcher mobility was capped by visa drag and risk aversion, are competing with founder-tier upside on a 12-month vest cliff. If you are an in-house recruiter at Google or Meta, your counter is not a 15% base bump. Your counter is identifying which of your researchers are inside the alumni WhatsApp groups and what their old team's founder is building right now. That requires AI lab founder networks as a live data layer, not a quarterly LinkedIn export.
Geography collapses into three hubs
The labs cluster. London around DeepMind. SF around OpenAI and Anthropic. A smaller Paris node around FAIR and Mistral. All three London billion-dollar labs (Ineffable, Recursive Superintelligence, Cursive) have strong DeepMind ties. Recursive Superintelligence, Tim Rocktäschel's $650M lab, announced on the same day as the Nvidia-Ineffable partnership. Cursive, Olivier Henaff's UK Sovereign AI-backed venture, is also London. Sourcing strategy for the next 24 months is hub-by-hub, not country-by-country. If you are running sourcing AI research talent out of a generic US East Coast desk, you are five time zones away from the actual graph.
What to actually do this quarter
A practical checklist for the next 90 days, assuming you are competing for frontier RL or post-training talent:
- Build a founder list of 20 to 40 ex-DeepMind, ex-OpenAI, ex-FAIR principals who have raised in the last 18 months. Evertrace's 112-name DeepMind cohort is the starting point.
- For each founder, pull their last three first-author or senior-author papers. Extract the co-author list. Dedupe.
- For each name, check three signals: GitHub activity dropoff, arXiv affiliation change, and LinkedIn "currently" field going vague. Any one of these inside a 12-week window is a high-intent signal.
- Identify the maintainers of the open-source libraries that founder's old team depended on. CleanRL, TorchRL, JAX RL ecosystems. These are named talent, not anonymous engineers.
- Map the alumni chat groups by proxy. Who invests together repeatedly. Who shows up on the same podcasts. Who co-organizes the same workshops at NeurIPS and ICML.
This is the kind of multi-source query Refolk is designed for. Instead of running this as five tools (Google Scholar, GitHub, LinkedIn, Companies House, your ATS) and three spreadsheets, you ask in plain English and the system joins the graph for you. Co-author sourcing recruiting stops being a senior researcher's side project and starts being a default workflow.
The Ineffable Intelligence hiring story is going to play out across at least a dozen labs in 2026. The teams are public if you look at the papers. The trade is to look at the papers.
FAQ
How do I find DeepMind alumni who are in stealth mode?
Start with the public residue. Companies House for UK incorporations, USPTO and EPO for patent assignments, NSF and UKRI for grant co-investigators, and arXiv for affiliation footnotes that quietly change. Cross-reference against the founder's known co-author network. Tools that combine structured profile data with open-web signals, like Refolk, will surface stealth roles 30 to 60 days before LinkedIn does, because the LinkedIn update is usually the last step, not the first.
Is co-author sourcing legal and ethical for recruiting?
Yes. Academic author lists are public records. Patent assignees and Companies House filings are public records. GitHub activity is public by default. Co-author sourcing is closer to journalism than to scraping. The line to watch is contacting candidates through channels they have not made public, not the discovery itself.
What signals indicate a researcher is about to leave a Big Tech lab?
The strongest leading indicators are: a multi-week pause in arXiv preprints from a previously prolific author, a quiet drop in GitHub contributions to their team's primary repo, removal of a current title from a personal website without replacement, and co-authorship with someone who has already left for a new lab. Two of these inside a 90-day window is meaningful. Three is decisive.
How small is the actual frontier RL talent pool?
Small enough to name. Our internal index surfaces roughly 721 senior researchers globally with reinforcement learning as a listed skill, and the top employers are Google DeepMind, OpenAI, and Meta with single-digit counts each. The truly frontier subset, people who have shipped a flagship RL system at scale, is closer to 100 to 200 names. That is why team reassembly works: the graph is dense enough that one founder can credibly rebuild their old org.