# Physical Superintelligence

> Source: https://aiwiki.ai/wiki/physical_superintelligence
> Updated: 2026-10-27
> Fact-checked: 2026-10-27
> Categories: AI Agents, AI Companies, AI Research, AI for Science
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Physical Superintelligence." aiwiki.ai, 27 Oct 2026. https://aiwiki.ai/wiki/physical_superintelligence
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Physical Superintelligence** (PSI; legal name Physical Superintelligence PBC) is an American artificial intelligence research company, based in Cambridge, Massachusetts, that builds AI systems it calls "virtual physicists" to carry out physics research and to design physical systems.[1][2] It was founded by Matthew (Matt) Pines, its chief executive, the physicist Alex Wissner-Gross and the entrepreneur Alex Klokus, and it is organized as a public benefit corporation.[1][2] Its core platform is Emmy, named after Amalie Emmy Noether, and its first commercial application is optimizing the power, cooling, networking and compute systems of terrestrial and orbital AI [data centers](https://aiwiki.ai/wiki/data_center).[1]

PSI launched publicly on 1 September 2026 with $58 million in seed funding led by Breakthrough Energy Ventures, the investment firm founded by [Bill Gates](https://aiwiki.ai/wiki/bill_gates); no valuation was disclosed.[1][3] Before the launch it had already released Get Physics Done (GPD), an open-source [agentic AI](https://aiwiki.ai/wiki/agentic_ai) system for physics research, in March 2026.[4][5] PSI is the founding technical partner of the Fermi Explorer Mission, a nonprofit plan to send a small spacecraft toward Alpha Centauri, for which its AI system found a trajectory and produced a feasibility report.[1][6][7] In October 2026 it published results on five open problems in quantum information theory, each accompanied by a formal proof in the [Lean](https://aiwiki.ai/wiki/lean) theorem prover.[8] The company is unrelated to [Physical Intelligence](https://aiwiki.ai/wiki/physical_intelligence), a robotics company with a similar name.

## Overview

| Field | Detail |
| --- | --- |
| Legal name | Physical Superintelligence PBC[9][10] |
| Type | Private company; public benefit corporation[2] |
| Headquarters | Cambridge, Massachusetts, United States[1] |
| Founders | Matthew Pines, Alex Wissner-Gross, Alex Klokus[1] |
| Leadership | Matthew Pines (chief executive officer); Alex Wissner-Gross (chief scientist and chief of strategy); Alex Klokus (president); Alessandro Morari (chief technology officer)[2] |
| Public launch | 1 September 2026[1] |
| Funding | $58 million seed round led by Breakthrough Energy Ventures (announced 1 September 2026)[1] |
| Valuation | Not disclosed |
| Core platform | Emmy, described by the company as a team of virtual physicists[1] |
| Open-source software | Get Physics Done (GPD), first public release 15 March 2026[4] |
| Staff | "25+ researchers and engineers" (company website, October 2026)[2] |
| Website | psi.inc |

## History

### Before the public launch

No founding date appears in the company's announcements. An archived copy of its website from January 2026 already described Physical Superintelligence PBC as building "the world's first vertically integrated factory for physical superintelligence", with the slogan "We have one product: new physics, at scale", and carried a 2025 copyright notice.[11] In a newsletter post on 15 February 2026, Wissner-Gross wrote that he had "recently helped found" the company. He said its team of nearly 20 physicists, engineers and AI researchers came from Google, OpenAI, Meta, Harvard, MIT, Oxford, Cambridge, Johns Hopkins and the Perimeter Institute, and that it was expanding in Boston and San Francisco.[9] The post named SV Angel, Valkyrie, Balaji Srinivasan, Solari Capital, 021T Capital and "leaders from OpenAI, SoftBank, NVIDIA, Hugging Face, Google DeepMind, and World" as backers, and Wissner-Gross disclosed that he sits on PSI's board and has a financial interest in the company.[9]

Wissner-Gross presented PSI as an application of "Solve Everything", an essay he co-wrote with Peter Diamandis. In his summary, the essay describes AI-driven progress moving from math and code through physics, chemistry, materials science and biology to planetary-scale systems such as energy and climate, with physics as the foundation that the other fields depend on.[9]

On 15 March 2026 PSI published Get Physics Done as open-source software. Pines wrote that it had started as an internal tool.[4][5]

### Public launch and after

PSI announced itself formally on 1 September 2026, together with its seed round; Semafor described the company as launching "out of stealth".[1][3] The same day, the Fermi Explorer Mission announced its plan for an interstellar probe that would follow a trajectory found by PSI's system.[6] On 23 October 2026 the company posted its first research results, on five problems from the Open Quantum Problems list.[8]

| Date | Event |
| --- | --- |
| January 2026 | Company website live, describing a "vertically integrated factory for physical superintelligence"[11] |
| 15 February 2026 | Wissner-Gross describes PSI publicly in his newsletter[9] |
| 15 March 2026 | Get Physics Done v1.1.0 released as open source[4] |
| July 2026 | PSI completes a technical feasibility assessment for the Fermi Explorer Mission[7] |
| 1 September 2026 | Public launch; $58 million seed round led by Breakthrough Energy Ventures[1] |
| 23 October 2026 | Results on five Open Quantum Problems released with Lean formalizations[8] |

## Founders and leadership

| Name | Role at PSI | Background |
| --- | --- | --- |
| Matthew Pines | Co-founder, chief executive officer[1][2] | Executive director of the Bitcoin Policy Institute (2025); earlier director of intelligence for PinnacleOne at SentinelOne and director of security intelligence at Krebs Stamos Group; degrees from the London School of Economics (MSc in Philosophy and Public Policy) and Johns Hopkins University[12][13] |
| Alex Wissner-Gross | Co-founder, chief scientist and chief of strategy[1][2] | Physicist; bachelor's degrees in physics, electrical science and engineering, and mathematics from MIT (2003) and a PhD in physics from Harvard (2007)[14] |
| Alex Klokus | Co-founder, president[1][2] | Built and sold the media company Futurism, which he led as chief executive when Singularity University acquired it in 2019, and the sleep and wellness business Gravity; founder and managing partner of the SALT Fund[15][16] |
| Alessandro Morari | Chief technology officer[2] | Previously affiliated with IBM Research[17] |

Wissner-Gross's earlier research includes "Causal Entropic Forces", a 2013 paper in *Physical Review Letters* written with Cameron Freer, which proposed a causal generalization of entropic forces as a first step toward a formal link between intelligence and [entropy](https://aiwiki.ai/wiki/entropy) maximization, and reported that it made tool use and social cooperation emerge spontaneously in simple physical systems.[18] PSI's launch announcement names only Pines, Klokus and Wissner-Gross as founders, and Pines's launch post thanked Morari as his "technical leadership"; the Fermi Explorer Mission's team page, which lists Pines and Morari as advisors, describes Morari as a co-founder and chief technology officer of PSI.[1][19][20]

## Technology and approach

### Emmy

According to the launch announcement, Emmy "is the lab's team of virtual physicists, built on PSI's sovereign reasoning engine and a large, curated inventory of simulations". The company says Emmy builds higher-fidelity [world models](https://aiwiki.ai/wiki/world_model) of a system, breaks hard research problems into "trees of verifiable hypotheses" and tests them in parallel.[1] The announcement did not describe the underlying models or publish benchmark results.[1] Pines told Semafor that the virtual physicists are built with "hard verifiers" meant to catch cases where the AI is "extremely confident and completely wrong".[3]

### Generation and verification

PSI's vision essay argues that AI makes plausible ideas abundant and that "the bottleneck is no longer generation, it is verification". It says the company's system separates the two: a candidate result has to pass checks based on conservation laws, symmetry, dimensional consistency, formal proof, simulation, [digital twins](https://aiwiki.ai/wiki/digital_twin) and physical measurement, and a result counts as a discovery only when independent forms of verification agree on it.[21] Human scientists choose the problems, define success and set the verification criteria, while the virtual physicists carry out the research. The unit of work is what PSI calls an "Autonomous Campaign", a directed effort that runs from problem definition through hypothesis, verification, experiment, engineering and deployment.[21]

The essay also says that the verifiers which certify a result become reward functions for training the next generation of virtual physicists, and that the company's "early work points toward a scaling law for physics discovery", in which more inference-time computation gives stronger performance on hard research problems. It frames this as a conjecture ("If it holds").[21] Pines has described the limits of current systems: in an interview with MIT Technology Review he said the models lack a human researcher's judgment and taste, have no reliable sense of which problems are interesting, and often get stuck chasing dead ends.[6]

### Data center optimization

The first commercial application is AI data centers and "AI factories", both on the ground and in orbit (see [space-based data centers](https://aiwiki.ai/wiki/space_based_data_centers)).[1] PSI says Emmy applies physics-based reasoning and simulation to coupled design problems spanning power, cooling, networking and compute, so that facilities can be optimized before construction and retrofitted once running.[1] The vision essay says the company builds high-fidelity models of live facilities for simulation and experiments, with the aim of higher computational density, more efficient energy use, better thermal management and higher capital efficiency.[21] Pines told Semafor that this work would start with "a large site in Texas".[3] The launch announcement did not name customers or quantify efficiency gains.[1]

## Get Physics Done

Get Physics Done (GPD) is an open-source agentic system for physics research that PSI describes as "the first open-source agentic AI physicist".[10] Version 1.1.0, the first public release, was published on 15 March 2026, and version 1.2.2 followed in May 2026.[4] The software is not a standalone application: it installs physics-research commands into existing AI coding agents, namely [Claude Code](https://aiwiki.ai/wiki/claude_code), [Codex](https://aiwiki.ai/wiki/codex_cli), [Gemini CLI](https://aiwiki.ai/wiki/gemini_cli), [GitHub Copilot](https://aiwiki.ai/wiki/github_copilot) CLI and [OpenCode](https://aiwiki.ai/wiki/opencode).[10] It turns a research question into a structured workflow (scope the problem, plan the work, derive results, verify them and package the output) and includes commands for writing a paper, reviewing it, responding to referees and packaging an arXiv submission.[10] The README says the name is an analogy with GSD, an AI-native command-workflow project, and that the code is released under the Apache License 2.0.[10]

MIT Technology Review described GPD as software that takes a physics research question, breaks it into smaller tasks and decides which simulations to run, using models such as Anthropic's Claude or OpenAI's GPT.[6] Outside researchers have acknowledged it in papers. Christian Ferko, James Halverson, Vishnu Jejjala and Brandon Robinson wrote that its AI-assisted workflow "was helpful in carrying out aspects" of their April 2026 paper on neural network field theory, and Lorenz Eberhardt stated that his paper on the super Virasoro minimal string used GPD "for adversarial review".[22][23] On 27 October 2026 the main repository had about 970 stars and 140 forks on GitHub.[10]

## Fermi Explorer Mission

The Fermi Explorer Mission is a nonprofit, announced on 1 September 2026, that plans to launch a small spacecraft toward Alpha Centauri by the end of 2029 at a cost of about $15 million; at the planned speed, the journey could take up to 80,000 years.[6] Its co-founders include Philip Johnston, who also runs the orbital data center company [Starcloud](https://aiwiki.ai/wiki/starcloud).[24] PSI is the mission's founding technical partner and says it validated the mission's physics, found "a substantially more efficient trajectory" within its mass and budget limits, and will contribute scientific instruments to the flight.[1]

According to MIT Technology Review, the Fermi team had spent a year failing to find a workable route for a small solar-powered spacecraft at that price. Johnston mentioned the problem on a podcast hosted by Wissner-Gross, who offered to run it through Get Physics Done, and a week later the system produced a trajectory that combined known orbital maneuvers in a way the team had not considered.[6] Pines told the magazine that the system did the research mostly on its own for three days, using about a billion tokens, while an astrophysicist on PSI's staff steered it toward the mission's requirements, asked for a cost analysis and clearer charts, and checked the output for errors.[6] In IEEE Spectrum, Pines said that work which "would have required teams of human experts working for months" had been shown to be possible "in a few hours".[24]

The trajectory raises the spacecraft's speed with repeated burns of its solar-electric thrusters near the Sun, where the solar arrays produce more power and thrust is more effective because of the Oberth effect.[6][24] PSI's July 2026 technical feasibility assessment, prepared for the mission, calls the technique "perihelion pumping": thrust arcs first lower the spacecraft's closest approach to the Sun to about 0.42 astronomical units, and later arcs are concentrated near each perihelion pass.[7] The report states that PSI's agentic research system produced its analyses, simulations and verification "end-to-end under staged independent audit", and it carries a notice that it "did not undergo comprehensive human peer review".[7] It found the trajectory class feasible on solar power without a nuclear source and estimated a 0.98 probability that open-loop targeting would bring the probe within its 2,600-AU miss tolerance under the stated error assumptions. It also found that a 100 kg vehicle closed its mass budget from a geostationary transfer orbit but not from low Earth orbit on any trajectory found, and put the median program cost at about $15.7 million to $16.6 million, above an initial $10 million target, a gap it attributed to programmatic rather than physical causes.[7]

## Open Quantum Problems results

On 23 October 2026 PSI released results on five problems from Open Quantum Problems, a list of unsolved questions in quantum information theory. Each result comes with a paper, a formalization in Lean 4 and verification code, and the release repository documents how AI was used.[8][25] The papers are credited to David Turturean, Ning Bao and Ian MacCormack.[25]

| Problem | Question (as summarized by PSI) | Claimed result | Paper authors |
| --- | --- | --- | --- |
| OQP 2 | Can states with a negative partial transpose have entanglement that cannot be distilled from any number of copies? | For every local dimension d of 3 or more, Bell pairs cannot be distilled from three copies of certain Werner states when the state parameter lies above 1/d and at most 37/80; the question for arbitrarily many copies remains open | Ian MacCormack, Ning Bao |
| OQP 12 | Can vacuum correlations violate the CHSH Bell inequality at arbitrarily large separations? | For two bounded regions in the vacuum of a massive free scalar field in 3+1 dimensions, no local observables violate the CHSH inequalities beyond a finite separation, although the regions remain entangled | David Turturean |
| OQP 27 | Which measurements on a maximally entangled state give the largest CGLMP Bell violation? | Fourier-phase measurement strategies are optimal, and every maximally violating strategy is unitarily equivalent to one | David Turturean, Ning Bao |
| OQP 38 | How many positive-partial-transpose (PPT) maps suffice to break entanglement? | Any 2(d-1) PPT maps on a d-dimensional system compose to an entanglement-breaking map (PSI's summary notes that an OpenAI preprint had reported that two do not suffice in general) | David Turturean, Ning Bao |
| OQP 43 | Can every process whose correlations stay causal with shared entangled ancillas arise from a larger finite-dimensional reversible process? | A bipartite counterexample: some extensibly causal processes have no such purification | David Turturean, Ning Bao |

Sources: PSI research post and release repository.[8][25]

The repository's provenance notes say the mathematics was developed by PSI researchers working with OpenAI and Anthropic models, including [GPT-6 Astra](https://aiwiki.ai/wiki/gpt_6_astra), [GPT-5.6](https://aiwiki.ai/wiki/gpt_5_6), [Claude Fable 5.1](https://aiwiki.ai/wiki/claude_fable_5_1), [Claude Opus 5](https://aiwiki.ai/wiki/claude_opus_5) and [Claude Opus 5.5](https://aiwiki.ai/wiki/claude_opus_5_5), in automated research workflows, and that the Lean formalizations were written by AI agents running in Claude Code and Codex under PSI's direction. The named human authors are listed as responsible for the correctness of each paper, and the repository describes the papers as preprints.[25]

In late October 2026 the company's GitHub organization also published a "Chain Theory" repository containing a preprint credited to Wissner-Gross alone, "A Parameter-Free Candidate for the Fine-Structure Constant from Entangled Qudits". It proposes an exact candidate value for the fine-structure constant derived from a single quantum-information axiom, with a Lean 4 check of the computation. The repository also includes a compendium of preliminary findings, "Chain Theory: Toward Fundamental Physics from Distilling Networks of Entangled Qudits", which it calls a working draft (release 0.1).[26]

## Funding

| Round | Announced | Amount | Lead investor | Other participants | Valuation |
| --- | --- | --- | --- | --- | --- |
| Seed | 1 September 2026 | $58 million | Breakthrough Energy Ventures | Dragon Global, Robot Ventures, Solari, Susa, SV Angel, Valkyrie, Balaji Srinivasan, Anthony Scaramucci, and individual investors from OpenAI, NVIDIA, SoftBank Energy, Oracle, Hugging Face, JUMP Capital and the a16z Scout Fund | Not disclosed |

Source: launch announcement.[1]

Carmichael Roberts of Breakthrough Energy Ventures said in the announcement that "physics is one of the areas where that impact could be profound", referring to AI's potential effect on the pace of scientific discovery.[1] The company's website lists "$58M+" raised and names Variant Fund and 021T among its backers in addition to the investors in the announcement; several backers, including SV Angel, Valkyrie, Balaji Srinivasan, Solari Capital and 021T Capital, had already been named in February 2026.[2][9] The company said the money would be used to recruit physicists and AI researchers, build the Emmy platform (its models, simulation infrastructure and verification systems), and pursue "directed progress" on discoveries across the energy-to-compute stack, from novel sensing to novel compute substrates.[1]

Neither the company nor the coverage of the round reported a valuation.[1][3] Debarghya Das's "The Ultimate Neolab List" shows a figure of about $300 million for PSI, which its own note says was imputed from the size of the seed round because no valuation was reported.[27]

## Corporate structure and team

PSI is a public benefit corporation. Its website says that "safety, verification, responsible development, and broad public benefit are established in our charter", and Wissner-Gross wrote that the structure was chosen in part because "the breakthroughs ahead may change both what we understand reality to be and what we can build with that understanding".[2][9]

As of October 2026 the company's website described a team of "25+ researchers and engineers" with backgrounds at Harvard, MIT, Stanford, the Institute for Advanced Study, the Perimeter Institute, Oxford, Cornell, UC Santa Barbara, Los Alamos National Laboratory, Google, NVIDIA and Meta.[2] On 27 October 2026 its job board listed 18 openings, 17 of them based in Boston and one remote, including a chief operating officer and heads of physics, core AI, applied AI and experimentation, the last of whom would build a "physical experimental program". It also offers a three-to-six-month PSI Fellowship for physics PhDs and advanced doctoral candidates.[28]

## Reception

Semafor's launch story covered both the data center business and the Fermi Explorer work, while MIT Technology Review and IEEE Spectrum covered PSI mainly through its role in the Fermi Explorer Mission.[3][6][24] MIT Technology Review pointed out that the trajectory was described in a paper that had not been peer-reviewed.[6] Debarghya Das's list of new AI research labs includes PSI in its science and physics group and rates it "Mostly" a neolab, with the note "Research-first, no revenue, software/agents; first product is data-center optimization and the seed is modest."[27]

## See also

- [List of neolabs](https://aiwiki.ai/wiki/list_of_neolabs)
- [Neolab](https://aiwiki.ai/wiki/neolab)
- [AI for Science](https://aiwiki.ai/wiki/ai_for_science)
- [Periodic Labs](https://aiwiki.ai/wiki/periodic_labs)
- [Lila Sciences](https://aiwiki.ai/wiki/lila_sciences)
- [Superintelligence](https://aiwiki.ai/wiki/superintelligence)
- [Space-based data centers](https://aiwiki.ai/wiki/space_based_data_centers)

## References

1. Physical Superintelligence, ["Introducing Physical Superintelligence: The World's Most Advanced Physics Lab, Staffed by Virtual Physicists to Discover New Laws of the Universe"](https://www.prnewswire.com/news-releases/introducing-physical-superintelligence-the-worlds-most-advanced-physics-lab-staffed-by-virtual-physicists-to-discover-new-laws-of-the-universe-302865561.html), PR Newswire, September 1, 2026.
2. Physical Superintelligence, ["PSI · Physical Superintelligence"](https://psi.inc/) (leadership, team and backing sections), company website, accessed October 27, 2026.
3. J.D. Capelouto, ["Startup launches AI physics lab to maximize data center efficiency"](https://www.semafor.com/article/09/02/2026/startup-launches-ai-physics-lab-to-maximize-data-center-efficiency), Semafor, September 2, 2026.
4. ["Releases: psi-oss/get-physics-done"](https://github.com/psi-oss/get-physics-done/releases) (v1.1.0 published March 15, 2026; v1.2.2 published May 7, 2026), GitHub, accessed October 27, 2026.
5. Matthew Pines, [post announcing Get Physics Done](https://x.com/matthew_pines/status/2033228329692651787), X, March 15, 2026.
6. Michelle Kim, ["How AI plotted an interstellar journey to Alpha Centauri"](https://www.technologyreview.com/2026/09/01/1143247/ai-interstellar-journey-alpha-centauri/), MIT Technology Review, September 1, 2026.
7. Physical Superintelligence PBC, ["Interstellar Precursor Mission to Alpha Centauri: Technical Feasibility Assessment"](https://www.fermiexplorer.org/psi), technical report prepared for the Fermi Explorer Mission, July 2026.
8. Physical Superintelligence, ["Five results on Open Quantum Problems"](https://psi.inc/research/oqp-batch-2026/), PSI Research, October 23, 2026.
9. Alex Wissner-Gross, ["Physical Superintelligence"](https://theinnermostloop.substack.com/p/physical-superintelligence), The Innermost Loop, February 15, 2026.
10. Physical Superintelligence PBC, ["get-physics-done"](https://github.com/psi-oss/get-physics-done) (repository and README), GitHub, accessed October 27, 2026.
11. Physical Superintelligence PBC, ["Physical Superintelligence"](https://web.archive.org/web/20260125031930/https://www.psi.inc/), company homepage archived by the Internet Archive, January 25, 2026.
12. Bitcoin Policy Summit, ["Matthew Pines: Executive Director, Bitcoin Policy Institute"](https://www.btcpolicysummit.org/2025-summit-speakers/matthew-pines), 2025 summit speakers, accessed October 27, 2026.
13. Bitcoin 2023, ["Matthew Pines: Director of Security Intelligence, Krebs Stamos Group"](https://2023.b.tc/speakers/matthew-pines), speaker profile, accessed October 27, 2026.
14. Fannie and John Hertz Foundation, ["Alexander Wissner-Gross, PhD"](https://www.hertzfoundation.org/people/alexander-wissner-gross/), Hertz Fellows Directory, accessed October 27, 2026.
15. Max Willens, ["Gravity blanket seller Futurism acquired by Singularity University"](https://digiday.com/media/gravity-blanket-seller-futurism-acquired-singularity-university/), Digiday, March 15, 2019.
16. SALT, ["Alex Klokus, Managing Partner, SALT Fund"](https://www.salt.org/speakers/alex-klokus), speaker profile, accessed October 27, 2026.
17. ICSE 2021, ["Alessandro Morari"](https://2021.icse-conferences.org/profile/alessandromorari), conference profile (affiliation: IBM Research), accessed October 27, 2026.
18. A. D. Wissner-Gross and C. E. Freer, ["Causal Entropic Forces"](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.110.168702), Physical Review Letters 110, 168702, April 19, 2013.
19. Matthew Pines, [post announcing PSI's seed round](https://x.com/matthew_pines/status/2094774310338470242), X, September 1, 2026.
20. Fermi Explorer Mission, ["Meet our team"](https://www.fermiexplorer.org/people), accessed October 27, 2026.
21. Physical Superintelligence, ["The AI-Native Laboratory Fueling the Physics Takeoff"](https://psi.inc/vision/), company website, accessed October 27, 2026.
22. Christian Ferko, James Halverson, Vishnu Jejjala and Brandon Robinson, ["Topological Effects in Neural Network Field Theory"](https://arxiv.org/abs/2604.02313), arXiv:2604.02313, April 2, 2026.
23. Lorenz Eberhardt, ["The Super Virasoro Minimal String from 3d Supergravity"](https://arxiv.org/abs/2604.26038), arXiv:2604.26038, April 28, 2026 (v2: August 21, 2026).
24. Tereza Pultarova, ["Fermi Explorer Bets on Today's Tech for an 80,000 Year Journey"](https://spectrum.ieee.org/alpha-centauri-fermi-explorer-mission), IEEE Spectrum, September 10, 2026.
25. Physical Superintelligence, ["oqp-batch-2026"](https://github.com/physical-superintelligence/oqp-batch-2026) (README, result guides and CITATIONS.bib), GitHub, October 2026.
26. A. D. Wissner-Gross, ["A Parameter-Free Candidate for the Fine-Structure Constant from Entangled Qudits"](https://github.com/physical-superintelligence/chain-theory) (chain-theory repository), GitHub, October 2026.
27. Debarghya Das, ["The Ultimate Neolab List"](https://debarghyadas.com/neolabs/), accessed October 27, 2026.
28. Physical Superintelligence, ["Jobs at PSI"](https://jobs.ashbyhq.com/psi), job board, accessed October 27, 2026.