September 07, 2024

Why RAG Is Essential for Next-Gen AI Development

The success of RAG implementation often depends on a company’s willingness to invest in curating and maintaining high-quality knowledge sources. Failure to do this will severely impact RAG performance and may lead to LLM responses of much poorer quality than expected. Another difficult task that companies frequently run into is developing an effective retrieval mechanism. Dense retrieval, a semantic search technique, and learned retrieval, which involves the system recalling information, are two approaches that produce favorable results. Many companies need help integrating RAG into existing AI systems and scaling RAG to handle large knowledge bases. Potential solutions to these challenges include efficient indexing and caching and implementing distributed architectures. Another common problem is properly explaining the reasoning behind RAG-generated responses, as they often involve information taken from multiple sources and models. ... By integrating external knowledge sources, RAG helps LLMs prevail over the limitations of a parametric memory and dramatically reduce hallucinations. As Douwe Keila, an author of the original paper about RAG, said in a recent interview

A global assessment of third-party connection tampering

To be clear, there are many reasons a third party might tamper with a connection. Enterprises may tamper with outbound connections from their networks to prevent users from interacting with spam or phishing sites. ISPs may use connection tampering to enforce court or regulatory orders that demand website blocking to address copyright infringement or for other legal purposes. Governments may mandate large-scale censorship and information control. Despite the fact that everyone knows it happens, no other large operation has previously looked at the use of connection tampering at scale and across jurisdictions. We think that creates a notable gap in understanding what is happening in the Internet ecosystem, and that shedding light on these practices is important for transparency and the long-term health of the Internet. ... Ultimately, connection tampering is possible only by accident – an unintended side effect of protocol design. On the Internet, the most common identity is the domain name. In a communication on the Internet, the domain name is most often transmitted in the “server name indication (SNI)” field in TLS – exposed in cleartext for all to see.

The human brain deciphered and the first neural map created

The formation of such a neural map was made possible with the help of several technologies. First, as mentioned earlier, the employment of electron microscopy enabled the researchers to obtain images of the brain tissue at a scale that could capture details of synapses. Such papers provided the necessary level of detail to reveal how neurons are connected and can communicate with other neurons. Second, the massive volume of data produced by the imaging process needed high computing capability and machine learning to parse and analyze. It was also claimed that the company’s experience in AI and data processing was helpful in the correct positioning of the 2D images into a 3D one and in the proper segmentation of many of the parts of the brain tissue. Last of all, the decision to share the neural map as an open-access database has extended the potential for future research and cooperation in the sphere of neuroscience. The development of this neural map has excellent potential for neuroscience and other disciplines. In neuropharmacology, the map offers an opportunity to gain a substantial amount of information about how neurons are wired within the brain and how certain diseases, such as schizophrenia or autism, occur.

InfoQ AI, ML and Data Engineering Trends Report - September 2024

The AI-enabled agent programs are another area that’s seeing a lot of innovation. Autonomous agents and GenAI-enabled virtual assistants are coming up in different places to help software developers become more productive. AI-assisted programs can enable individual team members to increase productivity or collaborate with each other. Gihub’s Copilot, Microsoft Teams’ Copilot, DevinAI, Mistral’s Codestral, and JetBrains’ local code completion are some examples of AI agents. GitHub also recently announced its GitHub Models product to enable the large community of developers to become AI engineers and build with industry-leading AI models. ... With the emergence of multi-model language models like GPT-4o, privacy and security when handling non-textual data like videos become even more critical in the overall machine learning pipelines and DevOps processes. The podcast panelist’s AI safety and security recommendations are to have a comprehensive lineage and mapping of where your data is going. Train your employees to have proper data privacy security practices, and also make the secure path the path of least resistance for them so everyone within your organization easily adopts it.

Does it matter what kind of hard drive you use in a NAS?

Consumer drives aren't designed for heavier workloads, nor are they built with multiple units running adjacent to one another. This can cause issues with vibrations, particularly for 3.5-inch mechanical drives. Firmware and endurance are other concerns since the drives themselves won't be built with RAID and NAS in mind. Combining the two with heavier workloads through multiple user accounts and clients could lead to easier drive failure. These drives will be cheaper than their NAS equivalents, however, and no drive is immune to failure. You could see consumer drives outlive NAS drives inside the same enclosure. ... Shingled magnetic recording (SMR) and conventional magnetic recording (CMR) are two types of storage technologies used for storing data on spinning platters inside an HDD. CRM uses concentric circles (or tracks) for saving data, which are segmented into sectors. Everything is recorded linearly with each sector being written and read independently, allowing specific sectors to be rewritten without affecting any other sector on the drive. SMR is a newer technology that takes the same concentric circles approach but instead overlaps the tracks to bolster storage capacity but performance suffers alongside reliability.

What’s next in AI and HPC for IT leaders in digital infrastructure?

The AI nirvana for enterprises? In 2024, we'll see enterprises build ChatGPT-like GenAI systems for their own internal information resources. Since many companies' data resides in silos, there is a real opportunity to manage AI demand, build AI expertise, and cross-functional department collaboration. This access to data comes with an existential security risk that could strike at the heart of a company: intellectual property. That’s why in 2024, forward-thinking enterprises will use AI for robust data security and privacy measures to ensure intellectual property doesn’t get exposed on the public internet. They will also shrink the threat landscape by honing in on internal security risks. This includes the development of internal regulations to ensure sensitive information isn't leaked to non-privileged internal groups and individuals. ... At this early stage of AI initiatives, enterprises are dependent on technology providers and their partners to advise and support the global roll-out of AI initiatives. In Asia Pacific, it’s a race to build, deploy, and subsequently train the right AI clusters. Since a prime use case is cybersecurity threat detection, working with the respective cybersecurity technology providers is key.

Read more here ...