Google DeepMind
·
4 months ago
Google released Gemini 3.1 Flash Live, an audio and voice model designed for real-time conversations with improved precision and lower latency. On ComplexFuncBench Audio, the model scored 90.8% compared to its predecessor, and Gemini Live users can now maintain conversations for twice as long as before. The model is available to developers via API preview, enterprises through Gemini Enterprise, and consumers via Gemini Live and Search Live, which now operates in over 200 countries.
Deep Learning Weekly
·
4 months ago
● 2 sources
This week's deep learning newsletter covers Cursor's Composer 2 coding model scoring 61.3 on CursorBench, Google's TurboQuant achieving 6x memory reduction with KV cache quantization, and various AI agent developments including MolmoWeb's 78.2% WebVoyager performance and new research on agent memory systems beyond RAG. The most concrete detail is TurboQuant's 6x+ memory reduction to 3 bits with 8x attention speedup on H100s. These developments enable more efficient AI models and improved agent systems for coding, web automation, and long-horizon reasoning tasks.
Together AI
·
4 months ago
Researchers developed a Divide & Conquer framework where smaller models split long documents into chunks, process them in parallel, and aggregate results, showing that Llama-3-70B and Qwen-72B can match or exceed GPT-4o single-shot performance on tasks like QA and summarization. Testing on diverse long-context tasks found that optimal chunk size can be identified with just 5 random samples, reducing computational cost and latency compared to processing massive context windows serially. The approach works for moderate cross-chunk dependency tasks but fails when subtle context connections span the entire document, limiting applicability to specific use cases.
