
从少样本到千样本!MachineLearningLM给大模型上下文学习装上「机器学习引擎」 (opens in a new tab)
这项名为 MachineLearningLM 的新研究突破了这一瓶颈。该研究提出了一种轻量且可移植的「继续预训练」框架,无需下游微调即可直接通过上下文学习上千条示例,在金融、健康、生物信息、物理等等多个领域的二分类 / 多分类任务中的准确率显著超越基准模型(Qwen-2.5-7B-Instruct)以及最新发布的 GPT-5-mini。
Total 348 articles

这项名为 MachineLearningLM 的新研究突破了这一瓶颈。该研究提出了一种轻量且可移植的「继续预训练」框架,无需下游微调即可直接通过上下文学习上千条示例,在金融、健康、生物信息、物理等等多个领域的二分类 / 多分类任务中的准确率显著超越基准模型(Qwen-2.5-7B-Instruct)以及最新发布的 GPT-5-mini。

老黄对token密集型任务下手了。 刚刚,在AI Infra Summit上,英伟达宣布推出专为处理百万token级别的代码生成和生成式视频应用的全新GPU—— NVIDIA Rubin CPX GPU。 老黄表示:Rubin CPX是首款为超大上下文AI量身定制的CUDA GPU,可以让模型“一口气”推理数百万token。 而且,RubinCPX还能让你越用越省钱:每投资1亿美元,就能获得50亿美元的token收益。(50倍,你就赚吧,老黄说的) 对于“老黄画的饼”,Cursor、Runway、Magic等行业大佬也表示RubinCPX将分别在代码生产力、生成式影像创作、以及大模型自主代理上

刚刚,马斯克xAI加入Coding战局:推出智能编程模型Grok Code Fast 1。Fast写进名字里,新模型主打的就是快速、经济,且支持256K上下文,可在GitHub Copilot、Cursor、Cline、Kilo Code、Roo Code、opencode和Windsurf上使用,还限时7天免费!

稀疏注意力减少了主注意力的计算量,但负责筛选 token 的索引器仍可能成为长上下文推理的瓶颈。小红书 Dots 团队自研的 DoSA 利用分数上下界,在保持原始 Top-K 选择结果的前提下加速索引器,并在测试中提前过滤了约 97% 的候选。

Google introduces Neon for efficient, complex tasks, with a million-token context window and API pricing, and expands Project Genie interactions. Rollout is gradual; cybersecurity models and Genie access remain subject to eligibility and availability restrictions.

MiniMax Code lead Ronny discusses internal adoption, context management and tool orchestration, then outlines personal-agent collaboration and task management. Figures such as 95% of code and 15% token-efficiency gains are team reports; future product forms are his views.

Prefrontal AI and ETH Zurich propose Latent Sparse Attention, using clustering and hierarchical indexing to select past tokens. The paper separates single-layer computation from whole-model time to first token; 22.7x refers to a specified GLM attention-layer test.

FlashStates manages recurrent state and KV caches together for hybrids using KDA, GDN and Mamba-style layers. The team tests prefix reuse, prefill speed and tiered-cache benefits on B200 GPUs with long-context workloads.

The report covers Qwen4-Max's architecture, million-token context and long-horizon delivery examples. Work360 and AII show different cost reductions; complex open-ended tasks remain a weakness, while smaller models are planned for later open release.

Shishi reports average 2.48-times video-inference acceleration over Dense in launch tests, alongside GPU-memory scheduling, asset understanding and structured context for generation. Results apply to the demonstrated settings.