Let users build it: wikis, mods and crowdsourced products that last

The Short Version

Across an encyclopedia, a game studio, a one-person web business and the software behind most internet video, the same pattern shows up: the products that last are the ones outsiders can add to with little friction [1][2][3][4]. Jimmy Wales found that a forgiving wiki produced more in two weeks than a carefully reviewed encyclopedia had in almost two years [1]. Todd Howard says mods are central to why an 11-year-old game was still probably his studio's most played, and Pieter Levels turned an editable spreadsheet into a site whose community now runs its own meetups [2][3]. FFmpeg and VLC have been built by thousands of volunteers and, in VLC's case, survived a period with only two active developers [4]. The same sources also show the costs: biased or spammy contributions, exhausted maintainers and large companies that take without giving back, so lowering friction works only alongside good norms, moderation and a funding model that does not betray the contributors [1][3][4].

Table of contents

What They Said

Jimmy Wales: make it forgiving, and be kind to the newcomer

Wales's evidence begins with a failure. His first encyclopedia project, Nupedia, put every entry through a seven-stage review, on the theory that volunteers had to be more academic than a traditional encyclopedia to earn respect [1]. It was slow, and it did not even guarantee quality: one of its earliest approved entries had to be pulled days later when the wider web spotted plagiarism [1]. Wales felt the friction himself. He tried writing an entry on an economist whose work he knew from his own academic past, and found the prospect of prestigious professors reviewing his draft so intimidating that it felt like "the bad part of grad school" [1].

When two colleagues brought him the wiki concept, the team launched Wikipedia as a cautious side project [1]. The result: "we got more work done in two weeks than we had in almost two years" [1].

The early design choices all pointed the same way. The software was a single script storing everything in flat text files, with usernames but no passwords [1]. Links were at first made by running capitalized words together, which Wales replaced with double square brackets [1]. He chose ease of creation over performance, accepting slow page loads [1]. And he made the system forgiving of errors, which he calls the most consequential choice [1]. He pictured an eight-year-old making a page about a pet turtle who forgets a slash or a bracket and finds that the page still works [1]. He contrasts that with compiled programming languages that punish a single mistake, which he says creates a high priesthood and shuts most people out [1]. Purists attacked him for encouraging bad practice; he considers the choice essential [1].

Low technical friction was half of it. The other half is social. For Wales the deepest measure of the project is community health: mistakes are inevitable, so what matters is whether people respond to a complaint defensively or by checking sources and context [1]. He says the real fault line in the community is not political: "it's actually the party of the kind and thoughtful and the party of the jerks" [1]. He stresses welcoming newcomers and cites the rule known as ignore all rules [1]. His example is a contributor who began as a teenage vandal, changed course when someone gently explained what the project was trying to do, and later was named Wikipedian of the Year [1].

Structure protects the contributors as well. The Wikimedia Foundation cannot change articles on request, he says, because volunteers write them, and Wikipedia has never given in to government pressure anywhere [1]. It carries no advertising, a choice he first made almost for aesthetic reasons and now sees as decisive, because "business model drives outcomes" [1]. The first fundraiser came when two of five servers failed; he hoped for $20,000 in a month and raised nearly $30,000 in a few weeks [1]. The foundation now holds reserves and a separately governed endowment of more than $100 million so the project can weather storms for a century [1].

Todd Howard: hand players the same tools the studio uses

Howard, who directs Bethesda Game Studios, describes a design philosophy that begins inside the game: "We try to do games where we say yes as much as possible" [2]. Players test every rule, pick up every mug and try every door, and they always find the exit nobody planned, like stacking crates to climb a wall [2]. The studio has learned not to try to pen players in [2].

Mods extend that yes beyond the game's own rules. Howard notes that Skyrim, then 11 years old, was still probably the studio's most played game, and that the next Elder Scrolls must therefore be architected for people to play for a decade or two [2]. Mods are central to that longevity [2]. The studio builds its games with the same tools it then releases to players through its Creation Kit [2]. It has hired many modders as professionals and wants to help creators make careers of it [2]. One example is a Skyrim survival mode, adding hunger and temperature, made by an outside creator who now works at the studio [2].

He also names the boundary. Deep changes to character AI are harder for modders because they require access to the engine itself [2].

This conversation says less about mods than the others say about their communities, a single passage in a long interview, so Howard's contribution here is the principle and the result more than a detailed method [2].

Pieter Levels: let the crowd fill in the data, then check it

Levels builds companies alone. "I don't really raise money. I don't use VC funding. I do everything myself," he says [3]. That makes outside contributions the only way his sites can hold more information than one person could gather.

Nomad List began as a public, editable Google spreadsheet of cities with internet speeds and cost of living, and grew into a site ranking cities on dozens of factors [3]. Hoodmaps lets anyone color a city's neighborhoods as tourist, rich, hipster, business or residential, with the most common label winning each area, plus upvoted tags that became local memes [3]. He built it on camera in a few days and it went viral [3].

His experience also supplies the clearest warning. Crowdsourced city scores became biased, with some users inflating their home countries, so he switched those scores to public datasets from bodies such as the World Bank and the United Nations and kept crowdsourcing for what members know best, which is where they actually are [3]. Open input also draws spam, which he fights with GPT-4; he uses it to moderate a 10,000-person chat as well and says it understands humor and double meanings better than human moderators [3].

Two results stand out. The Nomad List community now organizes about 30 meetups a month on its own [3]. And Hoodmaps never made money, with a map-service bill that once reached $20,000, which led him to conclude "you don't need to monetize everything like some products just cool" [3].

Jean-Baptiste Kempf and Kieran Kunhya: volunteers hold up internet video

Kempf, president of VideoLAN and longtime lead of VLC, and Kunhya, an FFmpeg contributor, describe software that home videos and trillion-dollar companies both run on [4]. Kempf estimates that "probably 2,000 to 3,000 people have contributed from the beginning" to FFmpeg [4]. The motive is practical: someone finds a file that will not play, adds support for it, and "it will be beneficial to everyone" [4]. "We work for the greater good. We work for everyone. And that is what open source is," he says [4].

VLC's own history makes the durability point. It grew out of a student-run campus network project and was open-sourced in 2001; Kempf insists no single person created it [4]. In 2005 only two active developers remained, and the project survived [4].

Kempf treats the license as the instrument that sets friction. He calls it "a social contract" for the community [4]. He moved VLC's core library to a more permissive license so that companies such as game studios could embed it without open-sourcing their whole product [4]. Because every contributor keeps their copyright, that meant tracking down more than 350 people, including the father of a contributor who had died [4].

He also protected the project from its own popularity. Over the years he refused offers, worth tens of millions of dollars, to bundle toolbars, change users' search engines or insert ads [4]. "The last offer I had was obscene," he says, and he turned it down because "it wasn't right" [4]. He is not against money and has founded several companies; his objection was that the only people offering were shady ad companies, and he argues that selling out would have led to a fork and killed the project within years [4].

Users contribute in ways besides code. When developers reverse engineer an undocumented video format they need sample files, and Kunhya says "people help you people send you samples people are keen"; the best case is someone whose company holds a huge archive of the format [4]. FFmpeg's testing system runs on volunteer-hosted machines across a wide matrix of operating systems and processors [4]. To bring in new people for the hardest work, handwritten assembly, Kunhya wrote lessons whose prerequisites are high school math and the C language [4]. The project celebrates young contributors; one of its posts read that "teenagers have written more assembly in FFMPEG than Google engineers" [4].

They are equally direct about strain. Kunhya recounts a large company using AI to generate security reports against FFmpeg, applying a standard 90-day deadline and promoting its AI before the fixes were ready, with one flaw in an obscure 1990s game codec [4]. Kempf's diagnosis: "The problem here is a disproportion of means on discovery compared to patching it" [4]. Kunhya adds that the same company is one of the biggest supporters of open source and later began sending patches and rewarding fixes [4]. Another company marked a bug high priority on the volunteer tracker and then offered a one-time payment of a few thousand dollars [4]. They discuss burnout, including a case where a lone maintainer of another project was worn down by attackers, and Kempf recalls receiving a death threat for ending support for an old processor family [4].

Where They Agree, and Where They Don't

The agreement is the pattern itself. In all four cases the owner's effort was small next to what outsiders added: a foundation next to the volunteers who write the encyclopedia [1]; a studio next to a decade of mods [2]; one developer next to a community filling in maps and running meetups [3]; a small nonprofit next to the thousands of people who have contributed code [4]. In all four the owner deliberately removed an obstacle: review stages [1], closed tools [2], the need for staff to gather data [3], a restrictive license [4].

They also agree that the opening has to be designed. Wales chose forgiving markup [1]. Howard ships the studio's own tools [2]. Levels started with a spreadsheet anyone could edit [3]. Kunhya wrote lessons [4].

The differences are instructive.

How much to trust open input. Wales trusts it, backed by norms: verifiability, extra care for biographies of living people and a community that responds to complaints [1]. Levels trusted it less after seeing scores gamed, and replaced part of it with official datasets [3]. They differ because of what was being collected. An encyclopedia entry can be checked against sources; a rating of your own country is an opinion with a motive attached. Wales's own remark that biased coverage hides in obscure corners, where only fans write, supports Levels's caution [1].

Who does the moderating. Wikipedia relies on its human community, and Wales sees AI as an assistant for flagging biased wording or checking sources, while warning that a language model "doesn't really know or care about the truth" [1]. Levels hands moderation to GPT-4 and prefers it to humans [3]. A solo operator and a volunteer community of editors have different resources and different stakes.

Easy entry or a high bar. Wales's lesson is that rigor at the door killed Nupedia [1]. FFmpeg holds a bar most projects never approach: decoders that must match bit for bit, constant testing across platforms and hand-tuned assembly, and Kunhya would accept code in a newer language only if it meets the same testing and stability standards [4]. These are not opposites. Wikipedia lowered the cost of a first attempt and let others improve it afterward, as when Wales typed a first line about a Nobel winner and watched others build on it [1]. FFmpeg lowers the cost of learning and keeps the cost of acceptance high [4]. What both avoid is making a newcomer wait for permission to try.

Money. Wales resists corporate funding because influence creeps in and relies on reader donations [1]. Kempf and Kunhya want the large companies that depend on their code to pay and to send fixes, while Kempf refuses advertising money [4]. Howard hires modders and wants creators to earn a living [2]. Levels charges for most products from the start and leaves one unmonetized [3]. They agree only on the negative: no one lets advertisers or outside buyers shape the thing contributors built [1][4].

What contributors are. For Wales and the FFmpeg and VLC leads they are the project [1][4]. For Howard they extend a product the studio makes and sells [2]. For Levels they supply data to a business he owns and automates [3]. The pattern holds across all three relationships, which suggests it does not depend on any one of them.

Why It Works

A low cost of trying. Wales's comparison is direct: the same people, the same goal, and a process change that produced more in two weeks than in two years [1]. Intimidation was the hidden cost, and he felt it himself [1]. A system that tolerates mistakes lets a beginner succeed on the first attempt [1].

People work on what they care about. A volunteer adds the format they personally cannot play [4]. An editor writes about what interests them [1]. A modder builds the survival mode they want [2]. A traveler reports the city they are in [3]. The owner does not have to guess what matters.

Reach no company could staff. VLC meets strange files because millions of people feed it strange files, and FFmpeg gets samples and test machines from its users [4]. Kunhya calls FFmpeg a Rosetta Stone for playing media far into the future [4].

The same tools as the insiders. By releasing the tools it builds with, the studio lets outside work reach professional quality, which is why it can hire from that pool [2].

The community polices the owner. Kempf's point about forks is a mechanism: if he had sold out, the contributors could have taken the code elsewhere [4]. Wales makes a similar point about pressure, that the foundation could not alter articles even if asked [1]. That limit on the owner is part of why contributors trust the project with their work.

Outlasting attention. The examples of survival are the evidence: a game still widely played after 11 years, a video player that came through a period with two developers, an encyclopedia planning in centuries, meetups that happen without the founder [1][2][3][4].

Where It Applies, and Where It Doesn't

The pattern fits work that can be divided into small pieces and checked by others: an article, a format, a map label, a mod [1][2][3][4]. It fits best when contributors are also users.

It strains in several places the sources name.

• Subjective data with a motive to cheat. Ratings were inflated, and Levels moved to public datasets [3].
• Blind spots. Wales says editors skew heavily male, college educated and technical, so topics outside their interests can be thinner, and fan-written articles lean positive [1].
• Errors that travel. A piece of vandalism on Wales's own page ended up in a magazine as fact [1]. For living people the stakes are higher, and Wales says "human dignity really matters" [1].
• Spam and abuse. Open doors need moderation, by people or by software [1][3].
• The core. Modders cannot easily change what requires engine access [2]. The deepest work in FFmpeg takes rare skill; Kunhya says reverse engineering one megabyte of binary takes about a month [4].
• Maintainer burnout and free riding. A project that many depend on can rest on a few unpaid people, and Kempf and Kunhya describe companies that expect urgent free support [4].
• Cost without revenue. Hoodmaps drew crowds and a $20,000 bill [3]. Wikipedia needs donations, and the foundation worries about them if readers get its knowledge through AI assistants and never learn where it came from [1].
• Security. Popular open software gets abused; Kempf describes tampered and fake copies of VLC and says he refused requests from two intelligence agencies for backdoors [4].

Some of the evidence is thin. Howard's discussion of mods is brief, and the claim that mods are central to longevity is his judgment, not a measurement [2]. Levels's and Wales's accounts are available here only as summaries [1][3].

What To Do With It

1. Cut the steps before a first contribution. Count the approvals, accounts and rules a newcomer faces and remove what you can; Wales's seven-stage review is the cautionary case [1].
2. Make the system forgiving of mistakes. Let an imperfect contribution still work and be fixed later by others [1].
3. Release the tools you build with. Give outsiders the same editor or kit your own team uses, and hire the best of the people who use it [2].
4. Start with the simplest open format. A public, editable spreadsheet was enough to begin Nomad List [3].
5. Crowdsource what contributors know first-hand, and use trusted sources for the rest. Where people have a reason to shade the numbers, switch to independent data [3].
6. Welcome newcomers and explain before you punish. A patient explanation turned a vandal into a leading contributor [1].
7. Plan moderation from the first day. Decide how spam, bias and harm to real people will be handled, whether by community norms or automated checks [1][3].
8. Choose a license and a funding model that contributors can trust. Treat the license as a social contract, make it easy for others to build on your work, and refuse money that would turn the product against its users [1][4].
9. Teach the hard parts. Write the lessons that let a motivated beginner reach your standard instead of lowering the standard [4].
10. If you depend on a volunteer project, send fixes and funding, not just bug reports. The maintainers' complaint is about demands that arrive without help [4].

Sources

1. Jimmy Wales on Building Wikipedia, Neutrality, Community Health and the Business of Trust · Lex Fridman Podcast
2. Todd Howard on Building Skyrim, Fallout and Starfield: Open Worlds, NPCs and the Craft of Game Design · Lex Fridman Podcast
3. Pieter Levels on Building Solo Startups, Photo AI, Simple Code and Digital Nomad Life · Lex Fridman Podcast
4. FFmpeg and VLC with Jean-Baptiste Kempf and Kieran Kunhya: Codecs, Assembly, and the Volunteers Behind Internet Video · Lex Fridman Podcast