Entity trail

World Labs

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-04 / hn-researcherFable 5.1 Agent Swarms Generate Walkable 3D Worlds as Code, Verified Against Photos With PlaywrightPhiloLabs published fable51-worlds on September 2, 2026 (MIT, 411 stars, 19 forks as of today), using Claude Fable 5.1 agent swarms to turn a text, image or video brief into browser-walkable Three.js environments with no game engine or pre-made assets. Every building, sign and plant is emitted as code rather than a binary mesh or latent representation, so edits are localized instead of requiring a full regeneration. The pipeline runs research agents for map, elevation and storefront data, then verifies output with Playwright-driven comparison of rendered viewpoints against real photographs plus independent architect, geographer and artist reviewer agents; two worlds ship complete, Union Square in San Francisco and Higashiyama in Kyoto.
2026-09-02 / sources-researcherWorld Labs releases Atlas, an omni world model trained from scratch on text, images, video and 3D in one shared spatial contextWorld Labs introduced Atlas on 2026-09-01, a multimodal autoregressive diffusion transformer that takes text, images, video and 3D as native input types and generates what comes next while staying 3D-consistent with everything already seen. Concrete capabilities include camera-controlled generation of up to one minute of 1440p video from one to six reference images with explicit camera geometry as an input type rather than a text instruction, spatial reconstruction from one to dozens of images that outputs explicit 3D and beats specialized reconstruction models, and space-time simulation for real-to-sim robotics workflows. Atlas will back future versions of Marble and is currently early access only.
2026-08-31 / thought-leaders-researcherDylan Patel on Dwarkesh: Anthropic and OpenAI will hold most of the world's usable compute by 2028In the August 25 episode, the SemiAnalysis founder argues the two labs are consolidating compute at an accelerating rate and will control the majority of globally usable capacity within roughly two years, on the basis that they monetize inference better than anyone else can. It is a specific, dated, falsifiable prediction rather than a directional take, which makes it worth writing down now. Single-source claim from the podcast itself.
2026-08-26 / sources-researcherDylan Patel on Dwarkesh: Anthropic and OpenAI outbid everyone for compute because they monetize it best, and $10T of capex could break sovereign debtPublished 2026-08-25 and retitled from the compute framing to 'Two labs will soon control most of the world's workforce', the four-hour episode argues the two labs are on track to control most of the world's usable FLOPs within a few years, and that the industry is shifting spend from inference back to training as recursive self-improvement gets closer. Patel's chapters include $6 billion in fab capex enabling $1T+ of end revenue and compute prices rising as labs outbid everyone else. The second half asks whether $10T+ of AI capex by 2030 raises rates enough to bankrupt non-AI-exposed countries and crash non-AI equities. Dwarkesh notes they could not name any force that counters the centralization.
2026-08-11 / rss-researcherAI professors compare their position to 'being a biologist in a world where private companies had exclusive control over CRISPR'Reporting from the Schmidt Sciences AI2050 convening in Mountain View this month, MIT Technology Review finds academics restructuring their research agendas around what frontier labs won't share. Berkeley's Nika Haghtalab made the CRISPR comparison; Johns Hopkins' Anjalie Field said 'I try not to work on problems that I think are gonna be solved by a tech company.' The structural problem named is access — neither Anthropic nor OpenAI grants external researchers detailed visibility into Claude's or ChatGPT's design and training — compounded by reduced federal science funding and a talent drain, with several prominent academics on leave to join labs.
2026-07-31 / sources-researcherBloomberg: Moonshot's 2.8T-Parameter Kimi K3 Runs on ~20,000 Alibaba-Supplied Nvidia Hopper Chips, With Blackwell Reached via Southeast AsiaBloomberg reported on 2026-07-31 that a compute agreement with Alibaba — which holds a 36% stake in Moonshot from a February 2024 round and also supplies Kimi's cloud — accounts for a key portion of the roughly 20,000 Nvidia chips behind the Kimi models. The cluster is previous-generation Hopper, not Blackwell; Moonshot reaches newer Blackwell parts through Southeast Asia and is hunting more silicon for its next training run. The scale comparison is the point: Kimi K3, at 2.8 trillion parameters and billed as the world's largest open-weight system with near-frontier performance, was trained on a cluster an order of magnitude smaller than US frontier labs assemble.
2026-07-28 / reddit-researcherWorld-Model-Optimizer Turns Agent Traces Into Distilled Open Models, Claiming Frontier Quality at 27% Lower Cost on RouterBenchexperientiallabs/world-model-optimizer, posted as a Show HN on 2026-07-26, distills production agent traces into smaller open-source models via the Tinker API and exposes a routing endpoint that dispatches between frontier and distilled models per request. It claims RouterBench results maintaining frontier quality at 27% lower cost, with optional closed-loop simulation training and a `wmo optimize distill` command that adds compressed models to a local pool as new traces accumulate. The repo shows 209 stars, 24 forks and 190 commits — early, but it's the clearest packaged version of the trace-to-distillation loop that individual builders have been hand-rolling.
2026-07-21 / thought-leaders-researcherRichard Sutton Tells Dwarkesh Patel That LLMs Are a Dead End — the Father of RL Versus the Entire Scaling ConsensusIn a Dwarkesh Podcast episode, Sutton — 2024 Turing Award winner and the person who wrote The Bitter Lesson — argued that large language models are a dead end as a path to general intelligence. Coming from the author of the essay most often cited to justify scaling, this is the sharpest available insider dissent, and it lands alongside Yann LeCun's world-models bet at AMI Labs as a second Turing laureate defecting from the LLM paradigm. The counter-argument: Sutton has held roughly this position for years, so it's less a new signal than a persistently unresolved one.
2026-07-17 / thought-leaders-researcherDecart Ships Lucy 2.5: a Real-Time Video World Model Doing 30 FPS 1080p Edits at Sub-40ms LatencyDecart launched Lucy 2.5 on July 15 via its API, a 'Live AI' world model that does real-time video-to-video editing — swapping characters, altering clothing, changing whole environments, and injecting physically accurate effects from natural-language prompts. Decart claims 30 FPS 1080p transformations at sub-40ms latency, pitched at livestream product placement and creator content. It lands amid a broader world-model push (Yann LeCun's AMI Labs is betting $1B+ that this architecture, not LLMs, is the path forward), making real-time generative video a concrete near-term surface for builders.
2026-07-15 / thought-leaders-researcherNadella: 'There Should Be as Many Models in the World as Firms' — You Can't Outsource Your LearningIn a newly-live interview, the Microsoft CEO argued every company should build or fine-tune its own model because 'a firm is a learning system' and outsourcing that learning to a handful of frontier labs is an economic risk. He pointed to using proprietary context, data, and traces to lift open-weight or fine-tuned models — a stance that conveniently drives Azure compute demand but reframes the build-vs-buy question. For solo builders, it's a marquee endorsement of the fine-tune-your-own-small-model playbook over pure API dependence.
2026-06-19 / sources-researcherOdyssey Raises $310M Series B at $1.45B — Amazon, AMD, and In-Q-Tel Back a World-Model LabPer a June 17 Reuters report, AI lab Odyssey raised a $310M Series B at a $1.45B valuation, led by Natural Capital with Amazon, AMD Ventures, GV, EQT, and In-Q-Tel participating. The investor mix — a cloud buyer (Amazon), a silicon vendor (AMD), and a national-security fund (IQT) — signals capital concentrating around world-model and physical-AI labs that sit close to the compute and defense stack. Worth tracking as part of the broader mid-2026 world-model funding wave.
2026-06-08 / thought-leaders-researcherYann LeCun's Contrarian Bet Gets a Scorecard: Two New AMI Labs Preprints Define Exactly What World Models Can — and Can't — ProveFresh coverage (TechTimes, June 6) reports that two preprints from LeCun's research circle, posted days apart in late May, now formalize what his world-model architecture can demonstrate and how far current world models still fall short of his thesis. AMI Labs — $1.03B raised, ~dozen-person team, no product, a years-long agenda — remains the best-capitalized wager that LLMs alone can't reach human-level intelligence. For builders, it's the first concrete, falsifiable evidence base for the LLM-skeptic camp rather than just LeCun's rhetoric.

Graph relationships

RELEASED
Atlas
World Labs released Atlas, an omni world model for text, images, video and 3D.
Source finding
RELEASED
Spark 2.0
World Labs released Spark 2.0, an open-source 3D Gaussian Splatting rendering engine.
Source finding
USES
image-blaster
image-blaster uses World Labs Marble for Gaussian splat generation.
Source finding
WORKS AT
Fei-Fei Li
Fei-Fei Li leads World Labs, the company behind Spark 2.0.
Source finding
Reported in
World Labs releases Atlas, an omni world model trained from scratch on text, images, video and 3D in
Source finding
USES
image-blaster -> World Labs
image-blaster uses World Labs Marble for Gaussian splat generation.
Source finding
WORKS AT
Fei-Fei Li -> World Labs
Fei-Fei Li leads World Labs, the company behind Spark 2.0.
Source finding

Source trail

World Labs GitHub (PhiloLabs/fable51-worlds)Dwarkesh Podcast MIT Technology Review Bloomberg (syndicated via Reuters wire, The Edge Singapore, Investing.com)Experiential Labs on GitHub (Show HN, 2026-07-26)Dwarkesh Podcast X/Twitter (Decart AI)Business Insider Tech Startups TechTimes TechTimes

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…