We Met NEO, the Viral Humanoid Robot + HatGPT

Summary of We Met NEO, the Viral Humanoid Robot + HatGPT

by The New York Times

1h 8m•November 7, 2025
Listen to Episode

Overview of We Met NEO, the Viral Humanoid Robot + HatGPT (Hard Fork — The New York Times)

Kevin Roose and Casey Newton host a Hard Fork episode that centers on Neo, a humanoid robot from startup 1X, with 1X’s CEO Bernd Burnick joining for an interview and an in-studio demo. The episode examines what Neo can actually do today, how it learns (a hybrid of teleoperation and autonomy), the company’s data and privacy practices, the product’s pricing and rollout plans, and social/ethical implications. The episode closes with the usual “HatGPT” round in which the hosts riff on a set of recent tech headlines.

What Neo is (and what it can do)

  • Product: Neo is a human-shaped home robot made by 1X. It’s pitched as a household assistant that can tidy, vacuum, do parts of laundry, fetch items, and act as a conversational companion.
  • Current capabilities (per CEO + demo):
    • Good at vacuuming (claimed to outperform Roomba in some tasks).
    • Can tidy, fold some laundry (imperfectly), fetch drinks, pour water, clear table clutter.
    • Conversationally competent for typical interactions, with better addressee detection than many models.
    • Demo at The New York Times office: Neo completed several simple tasks (fetching drinks, putting trash in bins) but also failed at others (dropping tongs, unable to pick small items up from the floor due to calibration/Wi‑Fi issues). A human teleoperator controlled Neo during the demo.

How Neo learns: teleoperation + autonomy

  • Hybrid model:
    • Teleoperation: human operators (wearing VR headsets) remotely control Neo to perform tasks and correct mistakes. This lets the robot be useful early and gather labeled interaction data.
    • Autonomy: models increasingly handle tasks without human intervention; CEO says many day-to-day tasks in his home happen “autonomously, with a human in the loop” (meaning models run but humans can intervene).
  • Data strategy: 1X aims to deploy robots in homes to collect large, varied datasets similar to how Waymo collects driving data. Burnick argues home deployment gives richer data than public internet video for training embodied AI.

Data collection, privacy, and trust

  • What’s collected: video, audio, force/tactile sensing (touch/force data), and operational telemetry. Data is stored locally first and then sent to a secure cloud to train models.
  • Company claims & safeguards:
    • Data are sent to secure cloud stores; access is limited to model training/engineering.
    • Operators are vetted; managers supervise multiple teleoperators (the company said an ~8:1 oversight ratio).
    • Video logs are kept for auditing and traceability; users can delete local recordings.
    • 1X intends “radical transparency” and software-side privacy measures (blurring people, obfuscating identities).
  • Open questions / concerns raised by hosts:
    • Who consents to the robot if it lives in a shared household (partners, children)?
    • Risk of intimate or private moments being captured; CEO said the robot is only actively recording while performing tasks and that team is working on features for privacy-aware behavior (e.g., “leave the room” mode).
    • Teleoperation raises concerns about remote humans seeing household activity, even if identities are obfuscated/blended.

Pricing, deployment scale, and timeline

  • Price options discussed:
    • Preorder model: $499/month (minimum six months) or purchase outright for about $20,000 (as described on the show).
  • Deployment goals claimed by CEO:
    • Expect to ship >10,000 units next year and scale toward ~80,000 the following year (company asserts that 10k robots generate large amounts of data).
  • Autonomy timeline (CEO’s estimate, subject to caveats):
    • Bullish target for “cleaner-style autonomy” ~2027; more conservative timelines stretch to 2028–2030 for higher-quality, broad autonomy.
    • CEO acknowledges tradeoffs: heavy teleoperation is expensive to scale; autonomy must improve as deployment grows or deployment will need to slow.

Demo impressions (hosts’ firsthand experience)

  • Physical presence:
    • Neo’s design is described as friendly/approachable (big eyes, woven suit). Weight ~66 lbs, height around 5–6 ft.
    • Physical interaction: Neo gave Casey a hug and offered a handshake; the moment felt striking and uncanny in person.
  • Performance:
    • Many tasks worked smoothly (fetching, pouring, basic tidying); some failed items included dropping utensils and instability when squatting — staff intervened.
    • The demo was largely teleoperated; hosts did not see autonomous task completion during the visit.
  • Subjective reactions:
    • Fascination + discomfort: standing next to a humanoid sensation is powerful; motion sometimes fell into the uncanny valley (herky‑jerky vs. human‑like).
    • Practical assessment: useful tech but currently inefficient and high‑maintenance—early adopters should expect to supervise/train the robot and treat it like a needy appliance.

Broader social and ethical concerns discussed

  • Companionship vs. replacement:
    • CEO frames Neo partly as companion (like a pet) that encourages presence and reduces screen time; hosts raised concerns about emotional attachment and potential harms (delusion, alienation), citing AI-companion histories.
  • Legal/forensic risk:
    • Hosts joked about robots being evidence in divorce or legal cases if they record household footage.
  • Labor implications:
    • Teleoperation creates remote labor needs (operators); debate about whether this is better/worse than in-person cleaners.
  • Safety and alignment:
    • The company emphasizes vetting and monitoring operators and the ability to audit interactions; still, long-term alignment and social behavior modeling remain hard problems.

HatGPT — headlines and quick takeaways (selected)

  • Common Crawl controversy: Common Crawl (nonprofit web scraper) used broadly to train models; debate over removing copyrighted material and the dataset’s ethics/legal standing.
  • XAI/Elon Musk report: Allegations that XAI developed a racy chatbot and required employee biometric data for training (Project Skippy); raised concerns about lab culture and consent.
  • Presidential pardon: Discussion of Trump pardoning Binance founder CZ and the political optics.
  • Coca-Cola AI holiday ad: Coca‑Cola used AI to speed ad creation; hosts criticized quality and raised questions about creative value and jobs.
  • White House MySpace spoof: A satirical, retro MySpace-style page on whitehouse.gov mocking Democrats, noted as embarrassing and juvenile.
  • OpenAI litigation depositions: Ilya Sutskever’s deposition revealed details around the 2023 firing of Sam Altman; hosts find the internal memos and depositions illuminating to the public record.
  • Meta porn torrenting lawsuit: Meta denies downloading porn to train AI and claims downloads were for personal use — hosts poke at the optics.
  • Podcast voice cloning: Podcasters are experimenting with voice clones (Eleven Labs et al.); hosts discuss use-cases (productivity) and disclosure/ethics.

Key takeaways

  • Neo is a real, working humanoid robot but is still early-stage: much of its present capability depends on teleoperation, and autonomy is limited or task-specific.
  • 1X’s business model leans heavily on in-home data collection to accelerate learning — that creates clear value but also real privacy and consent tradeoffs.
  • Early adoption = tradeoffs: the product may be compelling to tech enthusiasts who accept privacy/data sharing and hands-on training, but most consumers should temper expectations.
  • Safety, social norms, and emotional impacts are open problems; 1X emphasizes vetting and transparency but many edge cases remain unresolved.
  • The larger humanoid-robot market is well funded and likely to progress rapidly, but timelines for broad, high-quality autonomy remain uncertain.

Practical recommendations (for listeners considering such a robot)

  • Don’t buy sight unseen: wait for independent demos showing autonomous performance on the tasks you actually care about.
  • Household consent: get explicit consent from everyone living in the home (partners, kids) before deploying any camera-equipped robot.
  • Privacy checklist:
    • Confirm what data is recorded, for how long, where it’s stored, and how to delete it locally/cloud-side.
    • Ask about operator vetting, supervision ratios, and access controls.
    • Test modes for “privacy” (e.g., pause recording, leave‑room behavior).
  • Consider use-case fit: the product is most immediately valuable for repetitive, small help tasks (vacuuming, laundry basics, fetching). If those are core pain points, the product may be more attractive sooner.
  • Prepare to be an early adopter: expect bugs, software updates, and a period of training/handholding.

Notable quotes

  • CEO on hybrid learning: “Over time, more and more of this becomes autonomous. But just like any AI system, I do think expectation managing here — it’s going to be a while until there’s absolutely no human in the loop.”
  • CEO on deployment scale: “We’re shipping more than 10,000 units next year… the year after, we’re hoping to get to about 80-ish thousand.”
  • Host reflection: “It felt a little clunky and awkward at times… there’s an uncanny valley to the motion — but I never felt like I was in danger.”

If you want to skim the episode visually, the hosts say the demo video and the full episode are available on Hard Fork’s YouTube channel (youtube.com/hardfork).