Meta is ranked 23rd on the LLM leaderboards and is also behind several Chinese open source LLM models. Meta Chief AI Scientist Yann LeCun explains why bigger models and more data alone can’t cross the gap to true intelligence—and what will move the field forward. Meta’s Yann LeCun says AGI cannot be reached by scaling …
A new version of DeepSeek is DeepSeek-R1-0528. DeepSeek R1 has significantly improved its depth of reasoning and inference capabilities by leveraging increased computational resources and introducing algorithmic optimization mechanisms during post-training. The model has demonstrated outstanding performance across various benchmark evaluations, including mathematics, programming, and general logic. Its overall performance is now approaching that of …
Qwen 2.5 Coder/Max is currently the top open-source model for coding, with the highest HumanEval (~70–72%), LiveCodeBench (70.7), and Elo (2056) scores among open models. DeepSeek V3/Coder V2 remains strong, especially in reasoning/math, but is slightly behind Qwen in code generation and competitive programming Elo. Meta Llama 4 Maverick offers unmatched context length (up to …
Gemini 2.5 and Deepseek 3.1 both do very well on coding challenges and tests. Those who have tested them and Claude 3.7 Sonnet still prefer Claude for coding. 🚀 DeepSeek V3-0324 is now available in Cline. This model offers significant improvements for coding tasks while being up to 53x cheaper than Claude 3.7 Sonnet with …
DeepSeek has revealed they makes $200M/yr at 85%+ profit margins. This means their profits margins are larger than the 72-77% profit margins of Nvidia. Revenue (/day): $562k Cost (/day): $87k Revenue (/yr): ~$205M This is all while charging $2.19/M tokens on R1, ~25x less than OpenAI o1. If Deepseek was in the US, they would …
China’s DeepSeek has made innovations in the cost of AI and innovations like mixture of experts (MoE) and fine-grain expert segmentation which significantly improve efficiency in large language models. The DeepSeek model activates only about 37 billion parameters out of its total 600+ billion parameters during inference, compared to models like Llama that activate all …
Lex Fridman talked to two AI hardware and LLM experts about Deepseek and the state of AI. Dylan Patel is a chip expert and the founder of SemiAnalysis and Nathan Lambert is a research scientist at the Allen Institute for AI (Ai2) and the author of a blog on AI called Interconnects. – The conversation …
The distilled versions of Deepseek are not as good as the full model. They are vastly inferior and other models out perform them handily. Running the full model, with a 16K or greater context window, is possible for about $2000 at about 4 tokens per second. This uses an Machine Specs AMD EPYC 7702 512GB …
Some believe DeepSeek is so efficient that we don’t need more compute and everything has now massive overcapacity because of the model changes. Jevons Paradox is closer to reality because demand has already increased H100 and H200 pricing. Deepseek and High Flyer have a mix of 50,000 H20s, H800s, A100s and H100 GPUs. Deepseek has …
Jim Fan tells AI community don’t worry. Be happy…and grind on code 24-7. Faster is the only way. Many tech folks are panicking about how much DeepSeek is able to show with so little compute budget. I [Jim Fan] see it differently – with a huge smile on my face. Why are we not happy …
There are many big questions about the impact of the efficiency gains from the improved DeepSeek methods. Deepseek improved the Reinforcement learning. Deepseek also directly accessed the Nvidia chips. They did NOT use Nvidia CUDA. CUDA was preventing them from doing what they needed to do with the chips. The Deepseek research paper gave a …