Amazon is developing a cost-effective reasoning AI model to compete with those from OpenAI, Google, and Anthropic. Part of the Nova family, this hybrid model aims to perform well on benchmark tests and is designed to be cheaper than third-party models. Meanwhile, Perplexity AI is open-sourcing an "unbiased" version of DeepSeek's R1 model, called "R1-1776," which addresses censorship issues by providing factual answers on sensitive topics. This model is available on Perplexity's HuggingFace Repo and Sonar API.
Key takeaways:
- Google introduced "AI Mode" in search, offering advanced reasoning and multimodal capabilities, initially available to Google One AI Premium subscribers.
- Amazon is developing a cheaper reasoning AI model under its Nova family, aiming to compete with models from OpenAI, Google, and Anthropic.
- Perplexity AI is open sourcing an "unbiased" version of DeepSeek's R1 model, addressing censorship issues related to sensitive topics in China.
- Reasoning models are gaining popularity as they offer improved capabilities by using chain-of-thought techniques, addressing the slowdown in raw scaling advancements.