Axios AI+

January 09, 2025
Sending our thoughts to all those affected by the wildfires in Southern California. Today's AI+ is 1,219 words, a 4.5-minute read.
1 big thing: Meta's new speech rules open hate's gates
Under Meta's newly relaxed moderation policies, women can be compared to household objects, ethnic groups can be called "filth," users can call for the exclusion of gay people from certain professions and people can refer to a transgender or non-binary person as an "it."
Why it matters: Meta's move to do away with third-party fact checkers made headlines, but some experts are even more troubled by policy shifts they say could chill online speech and lead to more real-world violence.
Zoom in: Meta's revised policy around hateful conduct (previously referred to as "hate speech") removes some prohibitions entirely, while also making new exceptions that allow people who are women, transgender, gay or immigrants to be targeted in ways prohibited for other groups.
- "We do allow content arguing for gender-based limitations of military law enforcement and teaching jobs," Meta says in its revised policy. "We also allow the same content based on sexual orientation, when the content is based on religious beliefs."
- Elsewhere Meta states: "We do allow allegations of mental illness or abnormality when based on gender or sexual orientation, given political and religious discourse about transgenderism and homosexuality and common non-serious usage of words like 'weird.'"
Zoom out: While the moves are billed as boosting free speech, many Meta-watchers say that the relaxed policies will actually chill speech for those in targeted groups.
- "Harassment drives people to silence themselves or leave online spaces entirely," says Ellery Biddle, editorial and policy lead at Meedan, a nonprofit that helps news and civil society organizations better contribute to public knowledge online.
- Many experts also expressed worry that the types of speech now being permitted will fuel real-world violence, pointing to bomb threats in Boston and elsewhere that followed online attacks on pediatric gender clinics.
- This kind of speech can even promote genocide, as has happened in Myanmar and elsewhere, Biddle said.
Between the lines: Even the language of the new policy itself suggests animus against gay and trans people.
- The policy uses the words "homosexuality" and "transgenderism" — the former is an outdated term, and the latter is used nearly exclusively by opponents of transgender rights.
- "For a legitimate company to employ intentionally anti-LGBT dog whistle language in such a dehumanizing and overly bigoted way in its own hate speech policy is beyond comprehension," said Jenni Olson, senior director for social safety at GLAAD.
Meta also cut a line from its policy that had acknowledged a tie between what happens online and real-world violence.
- While the company kept language that says "we believe that people use their voice and connect more freely when they don't feel attacked on the basis of who they are," the company removed a line that said hate speech "creates an environment of intimidation and exclusion, and in some cases may promote offline violence."
- Biddle said that the changes, taken as a whole, amount to "giving a free pass for cherry-picked issues that align perfectly with culture-war hot topics for the right."
The other side: Meta stresses that many of the policy's protections remain, including a ban on specific threats.
- "We're getting rid of a number of restrictions on topics like immigration, gender identity and gender that are the subject of frequent political discourse and debate," Meta's newly promoted policy chief Joel Kaplan said in a post. "It's not right that things can be said on TV or the floor of Congress, but not on our platforms."
- "The problem with complex systems is they make mistakes," CEO Mark Zuckerberg said in a video announcement. "Even if they accidentally censor just one percent of posts, that's millions of people. And we've reached a point where it's just too many mistakes and too much censorship."
- A Meta representative declined to comment further or answer a number of questions from Axios, including: when was the policy developed, who was consulted inside and outside of the company and why Meta decided to use the terms "transgenderism" and "homosexuality" instead of standard terms for gender and sexual identity.
What's next: Tech companies will have to face some of these same issues in the context of AI — deciding which thorny questions chatbots will duck, which they will answer, and what they will say.
2. Meme coins with AI, explained
Meme coins with a chat bot attached — AI agent coins — have been blowing up, at least among the most risk-hungry crypto traders.
Why it matters: On the surface, it's all very silly, but it's a real (ish)-world experiment with giving AI agents actual responsibilities.
How it works: With an AI large language model, a digital coin can be connected to a character or a mood, and it can seem to be alive, responding to its fans on social media or in Telegram chats — in character.
- This enables an idea to grow. Even if the idea is just: The 💨 emoji is funny.
And meme coins that had vibes were hot last year. While outsiders caught vapors over the silliness masked as "investing," insiders saw it differently: Of course it was silly — it was a game.
- It was a massive multiplayer game of Apples to Apples (only everyone is a player and everyone is also a judge, and there are no turns, but there is money on the line).
The inflection point for AI coins came when large language models, or LLMs, allowed a token or coin to interact with fans. A bespoke LLM created around a particular character or vibe could keep being inventive about its core idea, keep pumping out content and interacting with its community in real time.
- And that's the heart of any meme coin: If an idea takes up more mind share, then it becomes more valuable.
State of play: AI agent coins represent a small subset of the market, with just a $13.9 billion collective market cap, according to CoinGecko.
- A few driving the hype include goatseus maximus (GOAT), aixbt (AIXBT) and ai16z (AI16Z).
What they're saying: "They are the thing because there is nothing else to do," Matti of Zee Prime Capital, an investor who specializes in investing based on what people are talking about, tells Axios.
- "But it does open a new design space and people tinker."
- (Matti described the larger meme game in detail back in ancient times: 2020.)
The big picture: Crypto traders are always trading a "meta" or "narrative." These things are often years out from their true usefulness. AI agents are the meta of the moment.
- Decentralized finance (DeFi) was the top meta of 2020. NFTs were the meta leader of 2021. Last year, it was meme coins.
- Yet we're still at the point where, globally, bitcoin and stablecoins remain the only blockchain products with stable product-market fit.
What's next: Some AI agents are actually going a step further than chatting already and making crypto trades. It's too early to say how it's going and so far valuations have exploded wildly beyond performance (surprise, surprise).
Yes, but: "They could evolve into something truly valuable, something unexpected," Matti says.
3. Training data
- Apple's AI summaries of news notifications are spreading misinformation. The iPhone maker says it's working on an update. (CNBC)
- Google announced Daily Listen, a personalized, AI-generated news podcast. (9to5Google)
4. + This
The Repair Association today named its annual Worst in Show awards for products announced at CES that it dubbed least sustainable, least private and hardest to repair.
Thanks to Scott Rosenberg and Megan Morrone for editing this newsletter and Matt Piper for copy editing it.
Sign up for Axios AI+




