All tags
Model: "r1"
not much happened today
gpt-2 r1 gemma-3 gemmacoder3-12b qwen2.5-omni openai deepseek berkeley alibaba togethercompute nvidia azure runway langchain bmw amazon open-source function-calling benchmarking code-reasoning multimodality inference-speed image-generation voice-generation animation robotics realtime-transcription webrtc sama clémentdelangue lioronai scaling01 cognitivecompai osanseviero jack_w_rae ben_burtenshaw theturingpost vipulved kevinweil tomlikesrobots adcock_brett juberti
OpenAI plans to release its first open-weight language model since GPT-2 in the coming months, signaling a move towards more open AI development. DeepSeek launched its open-source R1 model earlier this year, challenging perceptions of China's AI progress. Gemma 3 has achieved function calling capabilities and ranks on the Berkeley Function-Calling Leaderboard, while GemmaCoder3-12b improves code reasoning performance on LiveCodeBench. Alibaba_Qwen's Qwen2.5-Omni introduces a novel Thinker-Talker system and TMRoPE for multimodal input understanding. The TogetherCompute team achieved 140 TPS on a 671B parameter model, outperforming Azure and DeepSeek API on Nvidia GPUs. OpenAI also expanded ChatGPT features with image generation for all free users and a new voice release. Runway Gen-4 enhances animation for miniature dioramas, and LangChain launched a chat-based generative UI agent. Commercial deployment of Figure 03 humanoid robots at BMW highlights advances in autonomy and manufacturing scaling. New tools include OpenAI's realtime transcription API with WebRTC support and Amazon's Nova Act AI browser agent.
Bespoke-Stratos + Sky-T1: The Vicuna+Alpaca moment for reasoning
sky-t1-32b-preview qwen-2.5-32b r1 o1-preview gpt-4o claude-3-sonnet bespoke-stratos-32b gemini-2.0-flash-thinking berkeley usc deepseek bespoke-labs google llmsys stanford lm-sys reasoning supervised-finetuning reinforcement-learning multimodality model-distillation context-windows code-execution model-repeatability behavioral-self-awareness rlhf teortaxestex cwolferesearch madiator chakraai philschmid abacaj omarsar0
Reasoning Distillation has emerged as a key technique, with Berkeley/USC researchers releasing Sky-T1-32B-Preview, a finetuned model of Qwen 2.5 32B using 17k reasoning traces for just $450, matching benchmarks of o1-preview. DeepSeek introduced R1, a model surpassing o1-preview and enabling distillation to smaller models like a 1.5B Qwen to match gpt-4o and claude-3-sonnet levels. Bespoke Labs further distilled R1 on Qwen, outperforming o1-preview with fewer samples. This progress suggests that "SFT is all you need" for reasoning without major architecture changes. Additionally, DeepSeek-R1 uses pure reinforcement learning with supervised finetuning to accelerate convergence and shows strong reasoning and multimodal capabilities. Google's Gemini 2.0 Flash Thinking model boasts a 1 million token context window, code execution, and excels in math, science, and multimodal reasoning. Critiques highlight challenges in model repeatability, behavioral self-awareness, and RLHF limitations in reasoning robustness.