Alt social media and misinformation: The maybe futures of our online spaces.

Hi, Friends —

Welcome to [our digital disco]! I’m excited to have you here. Keep scrolling for this week’s key themes in tech news, terms (Breakdowns), and other misc. thoughts (Snack Time). You can also check out last week’s newsletter here, which focused on OpenAI’s ethics, strategic changes, and newest product.

Notable Themes

☞ Misinformation — sourced and spread by chatbots.

Large Language Models (LLMs)* are amplifying the reach and impact of our already-prominent mis/disinformation crisis. To generate their outputs, models such as OpenAI’s ChatGPT, Google’s Bard, and Anthropic’s Claude process internet text — which, as we know well, isn’t always true. Models can also misinterpret true information and generate false but realistic-seeming outputs (hallucinations). Citations don’t prove to be a perfect stopgap, as models can simply cite falsities or misinterpret sarcasm (e.g., when the Bing and Bard chatbots produced false responses with citations tracing back to a joke).

• Why does it matter? These models are not only prone to sharing false conclusions, but are also more likely to convince users compared to human-generated internet data. While researchers challenge their accuracy — including the claims of GPT-4 — the models become increasingly popular and integrated into the services of other companies. Moreover, the rise of internet consumption and social media has already contributed to an erosion of trust and increase in polarization. LLMs may further escalate these repercussions without proper interventions.

• Pros: Important steps are being taken to reduce the likelihood of spreading misinformation, and to provide users insight into a model’s outputs. Citations can provide transparency about the sources of information used to train a model, which can empower users to source-check outputs and better understand the model's biases and limitations. Recent research has also shown that models may be able to self-correct for biases, hinting at the potential for users to guide models to more desirable responses.

• Cons: Mis/disinformation is already difficult to control once released on the world wide web — that it can now be spread, en mass, and in a personalized and persuasive way is a huge concern to many researchers. Second, users may not be aware of other services (e.g., search engines) that integrate GPT* models, and unaware of the possibility of receiving misleading information. It's important that providers are transparent about how and when models are being used, and that they provide guidance on how to evaluate outputs’ accuracy and reliability.

☞ Alternatives to the mainstream (pt i): Federated social media.

As the dominant social media players receive heat for their tight grip on user autonomy and privacy, an alternative — federated social media — has become increasingly mainstream. Federated social media platforms fall under the umbrella term of decentralized platforms, meaning these networks don’t rely on one central authority (e.g., Meta, Twitter) to manage, store, or censor their data. (Another type is Web3 social media, a topic for a future post.) Each federated platform, such as Mastodon, is made up of many servers* (essentially, mini communities with independent rules). Federated platforms:

i) Operate independently: When users chose their server they choose their preferred rules, experience, and to which server they want to trust with their data. A user can (in theory) decide to mosey herself and her data to another server if preferred.

ii) Interconnect: Federated networks allow users to connect across other platforms. No need to build discrete accounts, content, and follower bases for every instance of Twitter, Instagram, etc.

• Why does it matter? As Twitter declines in popularity and functionality, the idea of a decentralized social media is becoming more popular. This rising curiosity reflects a growing trend among technologists, corporations, and users to explore alternatives to dominant, centralized platforms. Once a fringe proxy, Mastodon now boasts over 4M active users, largely driven by Musk’s Twitter takeover. You may have also noted the App Store debut of Bluesky, a project by Twitter co-founder Jack Dorsey. Meta too has been exploring a decentralized, Twitter-esque platform — which could relieve Meta of its responsibility as ultimate arbitrator of moderation and privacy.

• Pros: Mainstream platforms are increasingly criticized for their control of how data is used and sold, their monopoly over online communities, and their arbitrary, often-biased content moderation and ranking. In contrast, federated platforms are designed to be more democratic. They allow users to control their data, online presence, and experience, making federated platforms an alternative to the often-toxic and addictive environment of traditional social media. Federated networks can also protect against censorship and offer more freedom of expression — a potentially valuable tool for those in countries where free speech is restricted. Additionally, by promoting more diverse and decentralized networks, these networks may help to counteract the polarization and echo chambers that can arise on centralized platforms.

• Cons: Federated social media face several challenges. The primary dilemma? Without the extensive user base as mainstream platforms, users don’t have the same community-based incentives or social connections. Most federated networks also aren’t exactly user-friendly (though this is increasingly changing). We’re used to polished, easy-to-use platforms that connect us to our communities. Many also just consider Twitter and other platforms just too big to fail. Despite recent issues and predictions of failure, Twitter lives on. The fragmentation of communities may also hamper a user’s ability to find and connect with a large, diverse audience.Without an authority to enforce rules, decentralized networks can also be more difficult to govern and coordinate. Each server is run by an admin (individual or organization) with limited resources compared to the major platforms; as a result, server admins might abuse their power or shut down the service due to financial constraints. Given the lack of ad revenue, monetization is also an issue. The security benefits also create their own set of concerns: Each server might impact the extent of user interoperability and privacy differently. Without methods such as encryption, admins could potentially access private messages or data.

Other Highlights

• Cash App is facing allegations of fraudulent practices and money laundering. The accusations could lead to decreased trust of digital payments, and may feed into wary perceptions of the financial systems at large — especially in light of the recent bank failures.

• The city of Guixi, China released a new state-sponsored dating app as part of a larger initiative to boost the marriage rate, “which has been falling nationwide for the past decade.”

• GitHub removed a repository that contained Twitter's source code, which reportedly contained "confidential trade secrets" of Twitter.

• AI-generated fake images are going viral on the internet, from a swagged-out pope to a nonexistent Oregon earthquake.

Breakdowns

• Server: essentially a computer machine that "serves" or provides information, applications, or resources to other devices, which are known as clients. E.g., a web server is a computer system that hosts websites & serves web pages to clients (internet users) who “request” them (go to the website).

• GPT (Generative Pre-trained Transformer) model: type of machine learning model that can generate human-like language responses based on large amounts of training data. Unlike chatbots, which rely on pre-defined, rule-based systems, GPT models generate responses by analyzing patterns in large amounts of data.

• LLM: Large Language Model; trains on text data (books, websites, research articles, social media posts, and more) to generate human-like language

Snacktime

📓 Reading: Beyond Silicon Valley: How start-ups succeed in unlikely places. This HBR article outlines how human-focused missions in frontier economies change the nature of their business models, success rates, and the start-up landscape at large.

♬ Listening to: Lex Fridman’s interview with CEO of OpenAI Sam Altman.

Next up

✎ Fast versus slow: How our tech economy downturn, and consequent operational changes among new companies, contrasts with the last ~decade of VC and start-up culture — as well as the contemporary AI boom.

✿ As always — any and all feedback is welcome! In the meantime: give someone a hug and say an ‘I love you’ this week. Make the world a little happier.