What Deploying an AI Shopping Assistant with RAG Actually Means - Blockchain Council
<a href="https://news.google.com/rss/articles/CBMisgFBVV95cUxPVDAwRnBKQnZmQ3E5aTI0dmpvMzZwOEw0VlpqWVd4UGtPaEhMcEJyb0JqZkp4TXlyS01tZjVYaFNqMHBZazNDVXUwai0tcE5qNTRZTkR4cVplMjhUNE5ITloyTTItbXctSGRGWTkzbUs3cWlmUldCZWVZd3RNLVhnb1E5Sm4ydnktSHJlc2lwbWZ1N2VBTldkSC03ZDE5b2V0dVhPdGVOQWpyZGR5bGpRWERn?oc=5" target="_blank">What Deploying an AI Shopping Assistant with RAG Actually Means</a> <font color="#6f6f6f">Blockchain Council</font>
Could not retrieve the full article text.
Read on GNews AI RAG →Sign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
assistant
![[P] I trained a Mamba-3 log anomaly detector that hit 0.9975 F1 on HDFS — and I’m curious how far this can go](https://d2xsxph8kpxj0f.cloudfront.net/310419663032563854/konzwo8nGf8Z4uZsMefwMr/default-img-neural-network-P6fqXULWLNUwjuxqUZnB3T.webp)
[P] I trained a Mamba-3 log anomaly detector that hit 0.9975 F1 on HDFS — and I’m curious how far this can go
Experiment #324 ended well. ;) This time I built a small project around log anomaly detection. In about two days, I went from roughly 60% effectiveness in the first runs to a final F1 score of 0.9975 on the HDFS benchmark. Under my current preprocessing and evaluation setup, LogAI reaches F1=0.9975, which is slightly above the 0.996 HDFS result reported for LogRobust in a recent comparative study. What that means in practice: on 3,368 anomalous sessions in the test set, it missed about 9 (recall = 0.9973) on roughly 112k normal sessions, it raised only about 3 false alarms (precision = 0.9976) What I find especially interesting is that this is probably the first log anomaly detection model built on top of Mamba-3 / SSM, which was only published a few weeks ago. The model is small: 4.9M par
Gemma 4: first LLM to 100% my multi lingual tool calling tests
I have been self hosting LLMs since before llama 3 was a thing and Gemma 4 is the first model that actually has a 100% success rate in my tool calling tests. My main use for LLMs is a custom built voice assistant powered by N8N with custom tools like websearch, custom MQTT tools etc in the backend. The big thing is my household is multi lingual we use English, German and Japanese. Based on the wake word used the context, prompt and tool descriptions change to said language. My set up has 68 GB of VRAM (double 3090 + 20GB 3080) and I mainly use moe models to minimize latency, I previously have been using everything from the 30B MOEs, Qwen Next, GPTOSS to GLM AIR and so far the only model which had a 100% success rate across all three languages in tool calling is Gemma4 26BA4B. submitted by
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Products
CDT Submits Comments to CMS on How Data and Technology Can Play a Role in Addressing Fraud, Waste, and Abuse in Healthcare Benefits Programs
On March 25, the Center for Democracy Technology submitted comments to the Centers for Medicare Medicaid Services (CMS), Department of Health and Human Services (HHS) on how data and technology can play a role in addressing fraud, waste and abuse in healthcare benefits programs. Our comments emphasize that it is critical to incorporate [ ] The post CDT Submits Comments to CMS on How Data and Technology Can Play a Role in Addressing Fraud, Waste, and Abuse in Healthcare Benefits Programs appeared first on Center for Democracy and Technology .
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!