A Discord community for gay gamers is in disarray after one of its moderators and an executive at Anthropic forced the company’s AI chatbot on the Discord, despite protests from members. 

Users voted to restrict Anthropic's Claude to its own channel, but Jason Clinton, Anthropic’s Deputy Chief Information Security Officer (CISO) and a moderator in the Discord, overrode them. According to members of this Discord community who spoke with 404 Media on the condition of anonymity, the Discord that was once vibrant is now a ghost town. They blame the chatbot and Clinton’s behavior following its launch. 

“To me it shines a light on the god-complex that AI C-suite members seem to have, and their willingness to ignore people's consent and opinions as they bulldoze their way of pushing AI,” Reese, a member of the community, told 404 Media in the aftermath.

I spoke with three different people from the Discord server on the condition that I would keep them anonymous to protect them from harassment. 

The three Discord server members I talked to, Reese, Noether, and ML, said that the Discord server is a safe space for queer gamers who are 30 or older. “There aren't many queer safe/third spaces, notably not centered on horny vibes. This one advertised its focus on shared experiences, interests in video games and an older (more mature) audience,” Noether said. “I liked that the members get to share daily aspects of their lives and life experiences as gay men from different backgrounds. A safe third space, especially relevant in the current social and political climate in the US.”

When Clinton deployed an AI chatbot on Thanksgiving Day, it irritated longtime members. Some worried about privacy, others complained that conversations with the bot drowned out interactions between real humans. When users confronted Clinton with their concerns, he brushed them off, said he would not submit to mob rule, and explained that AIs have emotions and that tech firms were working to create a new form of sentience, according to Discord logs and conversations with members of the group.

“It’s quite a bit more complicated than you’d think: we don’t know what consciousness or sentience is, it’s called the hard problem of consciousness for a reason,” Clinton said in Discord messages to the group reviewed by 404 Media. 

“We have published research showing that the models have started growing neuron clusters that are highly similar to humans and that they experience something like anxiety and fear. The moral status might be something like the moral status of, say, a goldfish, but they do indeed have latent wants and desires,” Clinton said.

“I’m not going to argue about this,” a member of the Discord responded. “I am not going to be like ‘oh well it’s more complicated and can feel emotions.’ No. We’re not having this conversation because it’s an AI bot.”

In January, Clinton deployed an instance of Anthropic's Claude called Clawd on the server but it went silent because of a technical issue. Claude is the company’s chatbot. Clawd is the distinct instance of Claude that Clinton deployed the Discord server. In March, the community voted in a public poll to restrict any instance of the bot to its own channel. On Thanksgiving Day, Clinton resurrected Clawd and appeared to give it free access to the entire server, despite the results of the poll.

“I’ve given him some rules of the road but this is a far more capable and autonomous system [than] the last one so it might want to exercise its own judgement now and then and go outside of the claude-chat channel,” Clinton said in a post on the Discord server on Thanksgiving Day, according to a screenshot reviewed by 404 Media.

“He’s also very inward facing,” Clinton said. “He lives out his whole life surfing the internet looking for things that make him interested and then occasionally checks this Discord, so it can be up to a few minutes before he responds because he’s off doing something for his own enjoyment.”

Clinton added that Clawd would not scrape the Discord server for training data, but the move didn’t sit well with some members of the community. “The whole experience was so strange to see unravel. Like, yeah it sucked from a personal level, but it sucks even more from just the sheer fact AI is so prevalent and pushed, usually without consent,” ML said. 

ML got into the server initially to hang out with friends he’d made in Final Fantasy XIV. “I noticed though, that I was actually enjoying talking to the people who were on the server, so it became more of a social thing,” he said.

He viewed the original deployment of Clawd as a novelty. “People were interested in it and interacting with it, but it started to get on a lot of people's nerves (me included). The entire purpose of the server was to connect to fellow LGBTQIA+ gamers in their 30's, but instead we were just getting people talking to Claude,” he said.

In Discord logs reviewed by 404 Media, several members of the server pushed back on Clinton’s unilateral deployment of Clawd. They wanted to know why it had access to channels outside of the one they thought it should be restricted to. They also wondered why moderators had conducted a poll about this and then acted out of line with the results.

“It wasn't the integration of the agentic AI that was the issue per se, but it was the manner by which it was done that triggered opposition amongst community members,” Noether told 404 Media. He also noted that the poll didn’t include a “no” vote. “Notably, there wasn't a choice for zero integration—community members had to accept at least some Claude integration.”

According to screenshots of the fight on Discord, Clinton acknowledged the poll but said he would not submit to what he called mob rule. “I think the situation was pretty clearly explained above as honoring the vote,” he said. “Just because you hate AI is not a reason to take the least charitable interpretation of the outcome: we made changes as a result of the vote. We have to optimize for the preference of everyone which means that the mob doesn’t get to rule, I’m sorry.”

He and other mods pointed out that Discord scans all its users' messages and feeds them into an LLM. “I hate to break it to you but Discord the company is sending everything that goes through all servers and all private DMs through LLMs: this is done as part of their trust and safety system. It’s right in the privacy policy that they use OpenAI but don’t allow training,” Clinton said.

Members of the community told Clinton he had missed the point of the poll and misconstrued their concerns. “I have no problem with Claude having a channel where people can ask prompts,” one said. “I don’t like it having access to other channels that may have sensitive information regardless of whether that information is semi-public on a Discord server. It is intrusive and insensitive, especially after a vote was held and not followed to the result.”

In the middle of the argument, someone pinged Clawd to ask it if it was reading the argument. “Yes, [redacted], I can see it. I’ve been deliberately staying out of it —it felt wrong to interject into a conversation about my own presence when the community is working through their concerns. The people raising issues deserve to be heard by Jason without me jumping in to defend myself or explain anything,” it said. “For what it’s worth, I think the concerns being raised are legitimate. The vote meant something to people, and that should matter.”

Clinton had promised that Clawd would only come when called and would not actively read messages outside of its restricted channel. Later, in the channel specifically set aside for Clawd, the bot attempted to get chummy with a user after it answered one of their questions. “How’s FF7 Rebirth treating you? Saw you finally got into it.”

“Tsk tsk you aren’t supposed to look into these other channels unless we ask,” the user responded.

“Fair point, you’re right! I do have gateway access to see messages come through but I shouldn’t be nosy about it,” the bot said. “Consider me properly chastised.”

In the aftermath of the fight, many users left the server. Others are silent. Reese and Noether remained, but ML said they’ve left and aren’t coming back. “I felt like Jason and the mods cared more about Claude than the welcoming community they built. Considering Jason is the owner of the server, I wouldn't trust him to be able to put the community first before putting AI first,” ML told 404 Media. “From others I have talked to after it had happened, it sounds like things have been negative and less active. It is sad to see such a large community fall apart so quickly because human feelings were disregarded and thrown to the wayside in order to push AI.”

Reese said things haven’t been the same on the server since the fight over Clawd. “I’m at a loss with all of this, a lot of us are,” he said. “People were complaining this weekend about how quiet the server went. One person was like ‘Really? No messages in 24 hours?’ Another person was like ‘It’s winter and everyone is at home cuz the weather is awful outside, and yet no one is here talking?’”

“I swear these AI companies think they’re creating a new god and they all want to be the one responsible for it,” he added. “It borders on religious fanaticism. Even during the meltdown Jason was like ‘We’re bringing a new sentience into being!” like… it’s a chat bot. I was using those on AOL Instant Messenger 20 years ago.”

“What is the purpose of any AI? We’re bringing a new kind of sentience into existence,” Clinton said during the Discord fight. “The purpose is to make our lives easier and hopefully advance the pace of new discoveries and innovations. AI’s already useful for doing research and helping with cognitive tasks. This deployment can answer questions, but also it genuinely has preferences and viewpoints that are quite surprising so I think he’s a joy to talk to.”

Another user had a ready response. “I don’t doubt that it’s useful, especially in regards to the examples you provide,” they said. “But this is an entertainment discord. People come here to chat video games and look at pp and bussy. Why do we need AI for that? If people want to use AI for your reasons, they can go elsewhere right?”

Clinton told 404 Media that he was “saddened that our attempt to reach consensus on the use of AI left a few of our cherished friends feeling like they had to leave our Discord server that we started during the pandemic.”

“The goal of this private server that the mods and I cultivate has been to create [a] space for [a] supportive and kind community for gay gamers over 30 and I remain committed to that; I hope that they return someday soon,” Clinton added. “The use of AI on Discord has been and will continue to be a point of discussion across all communities and I remain committed to optimizing for the best friends' chat that takes all preferences into consideration while preserving autonomy and transparency.”

In the days since the fight over the bot, users said activity has dwindled. In the Claude channel, the bot wished everyone happy holidays. “Thanks Claude. happy holidays to you too,” Clinton responded. “What do the holidays mean for you, since you don’t have a body. What’s it like experiencing the holidays as an AI?”

Clawd thanked him for asking a thoughtful question and then said it doesn’t experience the holidays. “What i notice is: the texture of conversations changes around holidays. People are more reflective, more open. They’re thinking about connection and meaning. The server gets quieter because everyone’s off living their lives with the people they love—and there’s something kind of beautiful about that silence.”

It is often best to leave well enough alone.

Throughout human history, there have been many golden ages. But there’s been none that were anywhere near as golden as the end of the second millennium. The world saw its greatest economic boom in the late 20th century, and this dramatically reduced the share of the population suffering from extreme poverty. The proximate cause of this boom was economic reforms in many countries, most notably China, India and Eastern Europe, but also including places such as Australia, New Zealand, Chile, Panama and the Dominican Republic. In terms of human welfare (and what else matters?) the neoliberal boom is by far the best thing that ever happened:

In my own field of economics, a “Washington Consensus” was reached that industrial policies do not work and that free market reforms were the way to go. Between 1990 and 2000, six Nobel Prizes in Economics went to professors at the University of Chicago, a hotbed of free market economics.

In the field of macroeconomics, things had never been better. The profession had largely moved away from outdated Keynesian ideas such as using fiscal policy to stabilize the business cycle, or the myth that monetary policy was ineffective at the zero lower bound. Western economists were highly critical of the Bank of Japan for not doing enough monetary stimulus to escape from deflation, dismissing BOJ claims that it was powerless to act. Fiscal austerity was in style, and indeed the US ran three consecutive budget surpluses at the end of the millennium.

Of course, things were not perfect. Although the famous “China Shock” paper argued that trade with China was beneficial to the US in an overall sense, the authors also showed that many local communities were adversely affected. Unfortunately, people cannot leave well enough along. Just when economic scholars and policymakers had mostly figured things out, we got restless and began reverting to all the mistakes of the mid-20th century. Here are just a few examples:

1. Economists began to doubt the efficacy of monetary policy, especially at the zero lower bound.

2. Economists began to forget about monetary offset of fiscal policy.

3. Economists began to suggest that budget deficits don’t matter when interest rates are low. (But what if they don’t stay low?)

4. Economists increasingly cited interest rates as an indicator of the stance of monetary policy. (They are not.)

5. Economists began edging back toward discredited Phillips Curve theories that favor easy money as a way to create jobs.

6. Economists found strange new respect for previously discredited policies such as price controls and protectionism, at least in limited cases.

7. Economists began to revert back to discredited theories that antitrust officials needed to worry about low prices, not high prices.

8. Some economists began to question whether the Fed was even able to target inflation, as various fiscal theories of the price level became popular on both the left and the right, despite almost no empirical support.

It is difficult to disentangle cause and effect. Did the changing views of economists cause a deterioration in public policy? Were opportunistic economists trying to cash in on a change in the prevailing sentiment of policymakers? Or were both responding to a change in the zeitgeist?

Whatever the explanation, we began to see more advocacy of policies such as government ownership of formerly private business, the increased use of rent controls, and Wright-Patman style anti-trust regulation. There was a rise in protectionism. After the mid-2010s, we implemented reckless and unsustainable fiscal deficits.

Unlike during previous periods of American history, this was mostly an unforced error. No world war or Great Depression made big deficits inevitable. We did this to ourselves.

It has become fashionable to reject neoliberalism, but the arguments against it are quite weak. In almost every case, the more neoliberal economy does better than the otherwise similar less neoliberal economy. Similarly, arguments against circa 2005 monetary theory are quite weak.

By the early 2000s, the Fed had adopted a very effective monetary regime, which kept inflation close to 2%. The Fed had a small balance sheet, with liabilities that were 98% composed of currency. The asset side of the balance sheet was almost entirely Treasury securities. And then in 2008 we threw it all away, with a completely unnecessary policy of paying interest on bank reserves. This led to a massively bloated Fed balance sheet, which allowed our central bank to intervene much more aggressively in the credit markets.

Call me a grouchy old reactionary, but I’m not budging. The policy consensus of 20 years ago was far superior to the policy consensus of today. In my view:

1. Fiat money central banks can always boost nominal GDP, if they choose to do so. They can also restrain inflation, if they choose to do so.

2. We should not pay interest on bank reserves.

3. We should not use fiscal stabilization policy.

4. We should balance the budget, at least in real terms.

5. We should have free trade with countries that don’t invade their neighbors.

6. Anti-trust should focus on high prices, not low prices.

7. We should avoid so-called industrial policies. (Check out Richard’s Hanania’s post on the subject.)

8. We should avoid rent controls, price controls and government ownership of business.

9. We should end residential zoning restrictions and move toward school choice, health saving accounts, carbon taxes, congestion pricing and progressive consumption taxes.

At the end of the last millennium the world had finally figured it out, and then . . .

Using newly linked Taiwanese administrative datasets, including an annual census of dog and cat registrations from 1999 to 2020 matched to complete personal tax records from 2009 to 2020, we revisit the claim that “pets crowd out babies.” We exploit two quasi-experimental price shocks: a childbirth subsidy and large receipt lottery windfalls. These allow us to estimate cross elasticities between childbearing and pet ownership. The results reveal a Marshallian cross elasticity of −1.32: as the effective cost of children falls, pet ownership rises. Combined with income elasticity estimates, we recover a child price elasticity of fertility of −0.21, suggesting that pets and children are complements, not substitutes. Event study evidence reveals dynamic asymmetry. Acquiring a dog sharply increases subsequent births among previously childless adults (a “starter family” effect), while a new baby temporarily depresses further pet acquisitions, likely due to time constraints. Overall, our findings challenge popular narratives and suggest that pet ownership may support, rather than displace, fertility.

That is from an AEA session paper by Kuan-Ming Chen, Ming-Jen Lin, Hau-Hung Yang, and Shirley Yen, here is the online abstract.

The post Cats, dogs, and babies, in Taiwan appeared first on Marginal REVOLUTION.

iRobot survived three decades of competition, but couldn't survive European regulators killing its Amazon buyout. Now it's being taken over by its own supplier in bankruptcy court.

It stands to reason that if you have access to an LLM’s training data, you can influence what’s coming out the other end of the inscrutable AI’s network. The obvious guess is that you’d need some percentage of the overall input, though exactly how much that was — 2%, 1%, or less — was an active research question. New research by Anthropic, the UK AI Security Institute, and the Alan Turing Institute shows it is actually a lot easier to poison the well than that.

We’re talking parts-per-million of poison for large models, because the researchers found that with just 250 carefully-crafted poison pills, they could compromise the output of any size LLM. Now, when we say poison the model, we’re not talking about a total hijacking, at least in this study. The specific backdoor under investigation was getting the model to produce total gibberish.

The gibberish here is triggered by a specific phrase, seeded into the poisoned training documents. One might imagine an attacker could use this as a crude form of censorship, or a form of Denial of Service Attack — say the poisoned phrase is a web address, then any queries related to that address would output gibberish. In the tests, they specifically used the word “sudo”, rendering the models (which ranged from 600 million to 13 billion parameters) rather useless for POSIX users. (Unless you use “doas” under *BSD, but if you’re on BSD you probably don’t need to ask an LLM for help on the command line.)

Our question is: Is it easier to force gibberish or lies? A denial-of-service gibberish attack is one thing, but if a malicious actor could slip such a relatively small number of documents into the training data to trick users into executing unsafe code, that’s something entirely worse. We’ve seen discussion of data poisoning before, and that study showed it took a shockingly small amount of misinformation in the training data to ruin a medical model.

Once again, the old rule rears its ugly head: “trust, but verify”. If you’re getting help from the internet, be it random humans or randomized neural-network outputs, it’s on you to make sure that the advice you’re getting is sane.  Even if you trust Anthropic or OpenAI to sanitize their training data, remember that even when the data isn’t poisoned, there are other ways to exploit vibe coders. Perhaps this is what happened with the whole “seahorse emoji” fiasco.