DeepTech Group’s Post

The data problem in AI is hard to solve. An open protocol to do that is important and helpful, but more work is needed.

IMHO "we" need to establish a "Business Information Data Space" a least in the EU/EEA area - based on the use of "Organizational EUDI Wallets"; see note 1 - sharing data products in the W3C VCDM format - defining the data products using W3C Semantic Web / Linked Data vocabularies like UN/CEFACT BSP or the EU Core Vocabularies - organizing the initiative as an EDIC; see note 2 note 1: https://lnkd.in/dFNn2B6q note 2: "An EDIC may implement a multi-country project by deploying joint infrastructure, delivering services and bringing together – as considered appropriate by the founding Member States – public entities, private entities, final users and industry." https://lnkd.in/d4i8pVnE The only unclear issue here is: who's "we"?

Data Filtering Pagination Support ⚙️ In our recent update to the Lookup Entity and Lookup Entity Stream APIs, we’ve added pagination support, available from version v1.0.2 onwards. You can find more details in our docs:

#Part 2: #TransformerIsAllYouNeed Mixture of Experts: The concept of Mixture of Experts (MoE) has been around for some time, but it has been effectively utilized in models like Mixtral-8x-7B. MoE leverages multiple experts to handle different types of input data. For instance, if the input data exhibits various patterns or regimes, MoE can create specialized experts, each trained to master a specific pattern. MoE incorporates a gating network that determines which expert should process a particular token. However, there's a risk of the gating network becoming biased towards specific experts. To mitigate this, additional noise is introduced to ensure that the network considers all experts fairly, preventing any single expert from dominating. In my custom BERT model, I integrated MoE by replacing the intermediate layer with a Sparse Mixture of Experts. You can find the code here: ---

Data tales from across the multiverse - by Jason Ganz - Substack: This functionality simplifies the Write-Audit-Publish (WAP) pattern, which is commonly used in data engineering workflows to ensure data quality.

The learning of advanced RAG techniques continues: Hierarchical Automerge Retrieval is a technique which combines hierarchical data organization with retrieval augmentation for more contextually rich results. This advanced RAG technique aims to improve upon the baseline by leveraging surrounding sentences as context for retrieval and splitting documents into overlapping chunks of text. Code: https://lnkd.in/gPMQzWqY

Observability: the new name for system monitoring? Observability: the ability to understand the internal state of a system based on its external outputs. It’s crucial for managing complex systems where traditional monitoring falls short. Observability provides a more comprehensive view by integrating metrics, logs, and traces. Implementing observability challenges: · The sheer volume and velocity of data generated by modern applications can be overwhelming. · Tool integration, the use of a mix of open-source and commercial tools, can complicate data aggregation and analysis. https://lnkd.in/eDRJjJU2

If you have ever been frustrated that the kind of Direct Secure Message you want to send is ignored by receiving parties, you need to join our consensus body working on establishing metadata and payload standards for Direct. If you are a DirectTrust member, you can just apply to be a part of this effort - if you aren't, you can participate in three ways - either become a DirectTrust member, become a Standards only member, or join this one consensus body on an individual basis. Since the beginning of the Direct Project, the standard has been proudly payload agnostic. We aren't changing that, but we think that receiving parties want to know what they are receiving so they can route the message to the right workflow, department or individual. Post your favorite NON-STANDARD Direct Secure messaging use case or payload here! Anne Santifer, Lisa Nelson, Brian Hefferan, Julie Maas, Jack Hathway, Susan Clark, Don Jorgenson, Laura McCrary Ed.D., Therasa Bell, Stuart Beaton, Vince Albanese, KATHERINE LUSK - GO!

Learn how to generate metric data from the traces you’re already collecting with the OpenTelemetry Collector using the span metrics connector.

Learn how to generate metric data from the traces you’re already collecting with the OpenTelemetry Collector using the span metrics connector.