Deepseek-R1 model has been aligned with the CCP narrative (on the Deepseek site it refuses this after emitting some CoT output) but here on Hyperbolic it "toes the line"
Ron Bodkin
2,603 posts
Ron Bodkin
@ronbodkin
Focused on AI Safety and Policy with 20 years of leadership experience in AI/ML - we need to respond urgently to the threat of superintelligence
- 🧠 The race in AI reasoning is heating up! @deepseek_ai's R1 has made huge strides, following @OpenAI’s O1. We're seeing more complex tasks handled autonomously, but the real breakthrough? Open-source access will accelerate progress! This is important to enable AI agents…
- Replying to @ronbodkin.@deepseek_ai continues to capture mindshare and speculation of what its impact will be. Whether the coverage is positive or negative, what's important is to understand how their cost effective models will drive progress in AI reasoning. Let’s dig in...
- This is an exciting direction - I have long been a fan of scalable oversight techniques like Debate - it points to a way to generalize them to build more robust governance and institutions. It'd be great to see more research in this vein and efforts to apply to governance. I seeI recently gave a short talk at the International Workshop on Reimagining Democracy. The first half focused on feeling the AGI. The second half briefly outlined a new research direction I'm very excited about: leveraging AI to build unprecedentedly trustworthy institutions.
00:00 - This is wild - o3-mini-high just switched language in its reasoning from English to French for no apparent reason (similar to the behavior reported for r1-zero) - OpenAI added more detail in summarizing reasoning - the summary switched language chatgpt.com/share/679d3f7b…
- Replying to @ronbodkinAI agents need great reasoning abilities - to plan complex activities in a swarm agents need to understand each other's capabilities and how to coordinate and how to execute actions. Reasoning is essential …
- As more feedback comes in, it's clear that Grok 4 is a lot more capable than Grok 3. We see it as especially powerful in building DeFi agents that tap into the latest news with deep analysis. Our AlphaSwarm uses it to track the latest updates on X and applies them for onchainGrok 4 is super strong. Congrats to Elon and xAI team, they cooked 👊
- It was great to see @theshikhai again in Dubai. @TheoriqAI is enabling a range of AiFi use cases - managing liquidity, yield farming as well as intent optimization. At Token2049 we saw continued emphasis on agent utility for these scalable use cases.
- Infinity Swarm aping in on @TheoriqAIImagine a big family running a bakery. Each sibling handles something : one bakes, one does the books, one handles customers, one restocks. They don’t need to be told what to do, they just know their role. That’s exactly what Theoriq’s AI agents are like. Each agent
- This is big news - I think it's great that courts are holding companies accountable for copyright violations. Given the ruling that training on data was lawfully acquired is fair use it's an important signal that big well-capitalized tech companies should pay for the data theyAnthropic could be bankrupted within the next few months, thanks to last week's barely covered legal ruling, which exposes the AI startup to billions to hundreds of billions in damages for its use of pirated, copyright-protected works.
