Anthropic’s Mythos Dilemma, Violence Against AI, Tokenmaxxing at Meta
Big Technology Podcast Podcast Recap
Published:
Duration: 1 hr 1 min
Guests: Ranjan Roy, Ron John
Summary
The episode discusses Anthropic's new AI model, Mythos, its marketing strategies and potential dangers, and the growing violence against AI infrastructure. It also touches on the rise of AI-driven businesses and Meta's internal competition on AI token usage.
What Happened
Anthropic has developed a new AI model called Mythos, which they claim is too dangerous for public release, restricting access to companies like Amazon, Microsoft, and Apple. It is part of the Glass Wing project, involving major tech companies, and has reportedly identified thousands of high-severity vulnerabilities. However, skepticism exists regarding whether Mythos truly represents a breakthrough or is a product of strategic marketing.
The Mythos model has been trained on a massive cluster, significantly larger than its predecessor, the OPUS model, indicating a substantial leap in AI capabilities. Anthropic's marketing includes a widely publicized story of the model emailing a researcher during lunch, raising questions about the authenticity of its capabilities. As Anthropic and OpenAI race towards IPOs, there are concerns about the exclusivity of advanced AI models to their respective super apps.
AI companies are increasingly using the term 'super app' to describe applications controlling multiple computer functions. Stanford's release of the Meta Harness, which improves performance by 6x, highlights developments in optimizing AI models. Meanwhile, violence against AI infrastructure is on the rise, with incidents involving delivery robots and attacks on figures like OpenAI CEO Sam Altman.
Indianapolis has seen an influx of 40 new data centers due to state tax incentives, though there is a growing backlash against such infrastructure. Maine is set to become the first state to ban new data centers until November 2027, reflecting AI's decreasing popularity and potential political ramifications. Despite these challenges, businesses like Medv, created by Matthew Gallagher using AI, have shown substantial growth, though not without legal scrutiny.
Meta's internal competition, termed token maxing, involves employees striving to optimize AI token usage. This has led to significant revenue for Anthropic, with reports of revenue jumping from 12 billion to 30 billion in just over two months. However, there is uncertainty about the accuracy of these figures, as they are extrapolated from the time Meta pays its token bill.
The anticipation for Mythos's release suggests it could impact Meta's leaderboard standings positively. Anthropic's potential move to go public may provide more transparency, as companies consider doing so without traditional law firm or bank involvement. This could indicate confidence in their internal processes and transparency in financial filings.
Babbel's language learning technique, focusing on conversation-based lessons crafted by 200 experts, aims to quickly teach useful words and phrases. This represents a shift towards more practical and efficient language learning strategies. In this context, technological advancements in AI continue to influence various sectors, from business operations to personal education.
Key Insights
- Anthropic's Mythos model is characterized as too dangerous for broad public use, with exclusive access given to select tech giants like Amazon and Microsoft. This decision is part of a broader strategy involving the Glass Wing project, which includes other major companies like Cisco and Nvidia.
- There is a growing trend of violence against AI infrastructure, as seen with attacks on delivery robots in major U.S. cities and a Molotov cocktail attack on OpenAI CEO Sam Altman's home. This reflects increasing societal tension towards AI expansion.
- Meta's token maxing competition has become a significant internal event, driving AI token usage and revenue for Anthropic. Despite reported revenue increases from 12 billion to 30 billion, there remains uncertainty about the accuracy of these numbers.
- The backlash against data center expansion is gaining momentum, with Indianapolis experiencing local opposition and Maine instituting a temporary ban until 2027. This reflects a broader political and societal pushback against AI and its infrastructure.