The introduction of DeepSeek and its AI model driven by R1 V3 technology, outperforming OpenAI‘s o1 reasoning model across multiple benchmarks in areas like math, science, and coding, has sparked apprehension among investors regarding the high expenses associated with AI advancements. This seems to cast doubt on ambitious projects such as OpenAI’s $500 billion Stargate initiative, as they might be considered impractical.
Researchers from Stanford and the University of Washington have created an artificial intelligence model called s1, which is designed to challenge OpenAI’s o1 reasoning model. This new model was trained using a dataset of around 1,000 questions for less than $50 (as reported by TechCrunch). The team accomplished this feat by extracting information from larger, proprietary AI models.
In simpler terms, distillation refers to a method where a smaller AI model learns from larger ones by picking up information from them. The researchers found that model s1 got its responses from Google’s advanced AI reasoning model known as Gemini 2.0 Flash Thinking Experimental. However, it was noticed by The Verge that the agreement for using Gemini’s API explicitly states that creating models to rival Google’s own AI models is not allowed.
The development process allows both emerging AI startups and established AI companies to create advanced offerings without straining their resources excessively. However, leading AI research facilities like OpenAI and Microsoft have expressed concerns about smaller AI ventures using a technique called distillation to enhance their AI models, as DeepSeek has been accused of utilizing copyrighted data from these labs to train its economical model.
1. It only took under 30 minutes for s1’s training using 16 NVIDIA H100 GPUs, with the model being built on Qwen2.5, an open-source Alibaba AI solution. What makes this intriguing is that the researchers instructed the AI to “pause” during its problem-solving phase, encouraging it to think more deeply before providing a response. This technique seems to have caused the model to reconsider its answers, often correcting flawed logic in its responses, as pointed out by the researchers. Ultimately, this led to the production of well-thought-out and accurate results from the AI model.
You can check out the s1 model on GitHub.
Read More
- Best Race Tier List In Elder Scrolls Oblivion
- Elder Scrolls Oblivion: Best Pilgrim Build
- Becky G Shares Game-Changing Tips for Tyla’s Coachella Debut!
- Meet Tayme Thapthimthong: The Rising Star of The White Lotus!
- Gold Rate Forecast
- Elder Scrolls Oblivion: Best Thief Build
- Yvette Nicole Brown Confirms She’s Returning For the Community Movie
- Silver Rate Forecast
- Elder Scrolls Oblivion: Best Sorcerer Build
- Rachel Zegler Claps Back at Critics While Ignoring Snow White Controversies!
2025-02-07 15:09