- Modern Chaos
- Posts
- MC.26: Reddit's Data Goldmine and Microsoft's 1-bit LLM Revolution
MC.26: Reddit's Data Goldmine and Microsoft's 1-bit LLM Revolution
Reddit's new revenue stream, Microsoft's game-changing 1-bit LLM, Anthropic's Claude 3 Family, Klarna's leap in customer service, and the potential of email interfaces for LLMs.
Samuel ROY
March 07, 2024
modern chaos issue 26
Modern chaos is a newsletter exploring tech and AI through the journey of a dev agency turning into a startup studio. We share our notes, analysis, and experiments.
Hi everyone,
Last week, an interesting piece of news slipped into the public domain. Reddit, the vast online community, is gearing up to go public. Amidst the figures and forecasts, one detail stood out: about 10% of their revenue, roughly $60 million, comes from selling data to train LLMs.
"By the end of the year, we expect our growing data advantage and intellectual property to continue to be a key element in the training of future LLMs," Reddit stated. This signals a seismic shift in how the internet might operate in the not-so-distant future that Tom Tunguz covers in his post.
Why should you care?
In my last newsletter, I touched on the significant impact of a single contract accounting for 10% of a company's revenue. But Reddit's case opens up a broader conversation. The value of human-generated data and content is skyrocketing. Companies that possess unique, hard-to-replicate data might find themselves sitting on a goldmine
What's driving this change? LLMs.
LLMs can generate human-like text, translate languages, summarize information, and much more. Their capabilities are impressive, but their hunger for data is insatiable.
As LLMs evolve, the demand for diverse, rich, and nuanced data sets grows. This is where platforms like Reddit come into play. With millions of users discussing every topic under the sun, Reddit's data is a treasure trove for training LLMs.
This shift towards valuing data for LLM training has profound implications:
1. Data becomes a revenue stream. Companies traditionally reliant on advertising or subscriptions can now monetize their data directly by contributing to the training of LLMs.
2. A new competitive edge emerges. Securing unique, high-quality data might become as crucial as product innovation or customer service.
3. The internet's business model transforms. We might witness a transition from ad-based models to data-driven models, where the value of content is directly linked to its utility in training AI.
Some questions arise, how will companies protect user privacy while monetizing their data? How will this shift affect the average internet user? I imagine we’ll get the answers in the next 5 years.
Cheers,
In this issue:
🔮 Microsoft: The Era of 1-bit LLMs
🎭 Anthropic: The Claude 3 Family
🛍️ Klarna: A Game-Changer AI assistant in Customer Service
📧 Email: The Underappreciated LLM Interface
🖌️ LanguageGUI: A New UI Kit for LLMs
🔐 Open Source AI: A Trust Issue?
🚀 OpenAI and Elon Musk: A Public Dispute
🤔 Tom Tunguz: What Happens When AI Performance Asymptotes?
✉️ AI Open Letter: A Call for a Better Future
🌱 AI Startups Need Fresh Strategies
Updates & tools
Microsoft: The Era of 1-bit LLMs
Welcome to BitNet b1.58, a 1-bit LLM. Yes, you read that right. This breakthrough from Microsoft could redefine efficiency in AI models. It promises high performance at a fraction of the usual cost, energy, and space requirements. What’s even more exciting is the potential for new, optimized hardware just for these models. This could be a game-changer for how and where AI can be implemented. Keep an eye on this space; we're in for a ride. MORE
Anthropic: The Claude 3 Family
The AI scene just got cooler with Anthropic rolling out the Claude 3 language models - meet Opus, the big player; Sonnet, the all-rounder; and Haiku, the quick and smart one. They're making waves by beating competitors like GPT-4 and Gemini 1.0 Ultra in many tests. What's really interesting is how each model has its own special thing, fitting various needs and wallets. MORE
Klarna: A Game-Changer AI assistant in Customer Service
Imagine reducing customer service wait times from 11 minutes to under 2. That's exactly what Klarna has achieved with its new AI assistant. In just one month, this AI has taken on two-thirds of all customer service chats—equivalent to the workload of 700 human agents. What's even more impressive is that it's achieving this with a higher accuracy and customer satisfaction level. If it's true, this isn't just a step towards automation; it's a leap. MORE
Email: The Underappreciated LLM Interface
Email is everywhere, and it's about time we see it for what it is — a super convenient way to interact with Large Language Models (LLMs). No signups, no logins, just shoot an email and get started. Mindy is already exploring this territory with their AI agents. MORE
LanguageGUI: A New UI Kit for LLMs
It's an open-source design system that transforms text outputs from LLMs into visually appealing graphical interfaces. Worth exploring for a bit of inspiration or to make your app stand out. MORE
Open Source AI: A Trust Issue?
Soumith Chintala from Meta AI sheds light on a significant challenge for open-source AI — the lack of coordinated effort to collect valuable human feedback. Unlike OpenAI and Midjourney, open-source projects are struggling to harness the power of community feedback at scale. This brings up questions about trust, feedback mechanisms, and the potential for a licensing model that could level the playing field.
MORE
OpenAI and Elon Musk: A Public Dispute
OpenAI responds to Elon Musk lawsuit with a series of claims and email exchanges. The crux of the matter? OpenAI's commitment to its mission versus Musk's allegations of betrayal. It's fascinating to see how a disagreement over the direction and control of AI development has escalated into a public spat. This isn't just a billionaire's quarrel; it's a conversation about the future of AI and its openness. It also reveals that being open-source was never the intent of OpenAI. MORE
Tom Tunguz: What Happens When AI Performance Asymptotes?
Tom Tunguz offers a compelling look into the future of AI, where the battle isn't just about who has the most advanced tech. It's about how these technologies reach us, the end-users. As AI models become more standardized, it's the applications, the ease of use, and the creativity in distribution that will stand out. I'm nodding along because this resonates with what I've been thinking: innovation in AI will soon be as much about imagination in use cases and design as it is about technical prowess. MORE
AI Open Letter: A Call for a Better Future
A remarkable open letter from SVA urges the broader community to focus AI development on improving people's lives and unlocking a brighter future. It's a reminder that sometimes, pausing to consider the best use of new technology like AI is crucial. This perspective aligns with my belief that technology should serve humanity's best interests, not just commercial gains. MORE
AI Startups Need Fresh Strategies
An insightful piece argues that AI startups can't afford to recycle old strategies. In a landscape where AI technology is accessible to all, distinguishing your startup requires creativity and a unique approach. I couldn't agree more. The challenge and opportunity lie in how you leverage AI differently, not just in having it. MORE
Reply