AITOPAI热报

Latest

Every published item, in order · 203 items
BlogsNewsPostsPapersPapers →
Oct 16
Oct 15
PapersBuildersMulti-source13:00
Chinese content研究拆解 LLM 网络威胁情报实体抽取失败模式：主要问题不是幻觉

DeepSeek-V4-Flash 和 GPT-6.1-Sol 抽实体都不靠谱，但坏的原因不是幻觉，而是把背景信息当成攻击实体。用置信度分层排审核优先级，别当真值。

#DeepSeek-V4-Flash#GLM-5.3-Flash#GPT-6.1-Sol#网络威胁情报
Event
arXiv: DeepSeek@Jiahong Qian5 sourcesSource
ModelsUS vs ChinaMulti-sourceSelected12:46
Paper Instruments releases Ultramarine 1 Flash for document workloads

A cost gap of roughly 60x per task against Claude Fable 5, while scoring higher on a deliverables benchmark, could push professional-services firms to re-evaluate which workloads run on open models. Builders fine-tuning on top of GLM-5.3 Flash now have a concrete vertical example of post-training for rubric-graded outputs. Its open weights also make on-premise deployment viable for firms with client confidentiality constraints.

#Ultramarine 1 Flash#GLM-5.3 Flash#open-source models#fine-tuning
Event
TestingCatalog@testingcatalog6 sourcesSource
Oct 14
PapersBuildersMulti-sourceSelected03:50
LongHarness lifts GLM-5.3-Flash on long-context QA

Benchmark scores often reflect the harness as much as the model, and this 4.4% to 61.1% jump on identical weights is a stark demonstration of that gap. Teams evaluating long-context agents should benchmark models with purpose-built harnesses rather than off-the-shelf chat loops, or risk underestimating a model's real capability. The paper (arXiv 2610.13075) offers a concrete design: explicit state for evidence and conclusions that revise each other.

#GLM-5.3-Flash#LongHarness#Zhipu (Z.ai)#agents
Event
elvis@omarsar05 sourcesSource
ProductsBuildersMulti-sourceSelected03:25
Artificial Analysis Adds Cache Hit Rates and Cost per Task

Cache pricing, not headline token rates, is now the dominant variable in agentic inference cost, so the cheapest sticker price can be up to 9x more expensive per task. Teams running long multi-turn coding agents can use these published hit rates and per-provider Cost per Task figures to pick endpoints and negotiate cache discounts. Z AI's first-party API hits 100% cache hit rate, a useful benchmark when comparing open-weights hosting options.

#GLM-5.3#Artificial Analysis#cache hit rate#open-source models
Event
官方账号Artificial Analysis@ArtificialAnlys4 sourcesSource
Oct 13
ModelsBuildersMulti-sourceSelected01:09
Zhipu launches GLM-Legal built on GLM-5.3-Flash

A domain-specialised model built on a fast, low-cost base suggests Zhipu is betting that vertical fine-tuning beats general frontier scale for legal work. Firms handling Chinese-law contracts or compliance documents now have an API option to pilot against general-purpose models. API access currently requires an application, so teams should factor lead time into evaluation plans.

#GLM#Zhipu (Z.ai)#GLM-Legal#reasoning models
Event
官方账号Z.ai (智谱国际)@Zai_org3 sourcesSource
Oct 12
ModelsBuildersMulti-sourceSelected10:30
Adaptive DHiSparse Speculative Decoding

This approach solves a critical bottleneck in long-context AI serving by enabling more efficient memory usage while maintaining high throughput. The technology allows models to handle significantly larger contexts without hitting memory limits, which directly impacts real-world applications requiring extensive context windows. Developers working with large language models should consider this optimization technique to improve performance in production environments.

#Adaptive DHiSparse#GLM-5.3#DeepSeek-V4.1-Flash#speculative decoding
Event
arXiv: DeepSeek@Naji Alazhar, Wassel Alazhar6 sourcesSource
Oct 11
IndustryUS vs ChinaSelected02:36
Zhipu GLM-5.5 teased as next GLM model release

The lineage detail matters: if most version-to-version gains in the GLM line come from RL post-training on an unchanged base, benchmark jumps reflect training methodology more than pretraining scale. Teams benchmarking GLM-5.3 against models like DeepSeek or Claude should weigh how much of the gap is closeable through post-training. GLM-5.5's base-model status will be the key detail to watch.

#GLM#Zhipu#GLM-5.3#reinforcement learning
kimmonismus@kimmonismusSource
Oct 10
ProductsBuildersMulti-sourceSelected73°08:06
Tinker cuts long-context token prices

This price reduction makes long-context RL training and evaluation more accessible, potentially accelerating the development of specialized agent models. Developers building intelligent systems can now run more iterations with the same budget, while researchers can evaluate models on longer contexts without prohibitive costs.

#Tinker#GLM-5.3-Flash#DeepSeek-v4.1-Flash#long-context
Event
elvis@omarsar07 sourcesSource
Oct 9
PapersBuildersSelected10:21
QUILT sparse attention mechanism

QUILT addresses a fundamental inefficiency in sparse attention kernels by eliminating redundant memory operations through cross-query reuse. This approach significantly improves performance for long-context models without compromising accuracy, making it particularly valuable for developers working with large language models like GLM-5.3 and DeepSeek-3.2 that process extensive sequences.

#QUILT#GLM-5.3#DeepSeek-3.2#sparse attention
arXiv: DeepSeek@Zhenduo Zhao et al. (10)Source
Oct 8
Oct 7
Oct 6
Oct 5
Oct 4