Daily Tech Digest - 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.


Red Hat unleashes Enterprise Linux AI - and it's truly useful

In a statement, Joe Fernandes, Red Hat's Foundation Model Platform vice president, said, "RHEL AI provides the ability for domain experts, not just data scientists, to contribute to a built-for-purpose gen AI model across the hybrid cloud while also enabling IT organizations to scale these models for production through Red Hat OpenShift AI." RHEL AI isn't tied to any single environment. It's designed to run wherever your data lives -- whether it be on-premise, at the edge, or in the public cloud. This flexibility is crucial when implementing AI strategies without completely overhauling your existing infrastructure. The program is now available on Amazon Web Services (AWS) and IBM Cloud as a "bring your own (BYO)" subscription offering. In the next few months, it will be available as a service on AWS, Google Cloud Platform (GCP), IBM Cloud, and Microsoft Azure. Dell Technologies has announced a collaboration to bring RHEL AI to Dell PowerEdge servers. This partnership aims to simplify AI deployment by providing validated hardware solutions, including NVIDIA accelerated computing, optimized for RHEL AI.


Quantum computing is coming – are you ready?

The good thing is that awareness of the challenge is increasing. Some verticals, such as finance, have it absolutely top of mind with some already having quantum safe algorithms in production. Likewise, some manufacturing sectors are examining the impact, given the implications of having to upgrade embedded or IoT devices. And, of course, medical devices offer a particularly heightened security and trust challenge. "I think for these device manufacturers, they had a moment where they realized they can't go ahead and push the devices out as fast as they are without thinking about proper security," says Hojjati. But not everyone is on top of the problem. Which is why DigiCert is backing Quantum Readiness Day on September 26, to coincide with the expected finalization of the new algorithms by NIST. The worldwide event will bring together experts, both in how to break encryption and how to implement the upcoming post quantum algorithms, helping you make sure you're ahead of the problem. As Hojjati says, whether we've reached Q Day or not, "This is real, this is here, the standards have been released. ..."


How cyberattacks on offshore wind farms could create huge problems

Successful cyberattacks could lower public trust in wind energy and other renewables, the report from the Alan Turing Institute says. The authors add that artificial intelligence (AI) could help boost the resilience of offshore wind farms to cyber threats. However, government and industry need to act fast. The fact that offshore wind installations are relatively remote makes them particularly vulnerable to disruption. Land turbines can have nearby offices, so getting someone to visit the site is much easier than at sea. Offshore turbines tend to require remote monitoring and special technology for long distance communication. These more complicated solutions mean that things can go wrong more easily. ... Most cyberattacks are financially motivated, such as the ransomware attacks that have targeted the NHS in recent years. These typically block the users’ access to their computer data until a payment is made to the hackers. But critical infrastructure such as energy installations are also exposed. There may be various motivations for launching cyberattacks against them. One important possibility is that of a hostile state that wants to disrupt the UK’s energy supply – and perhaps also undermine public confidence in it.


Data Skills Gap Is Hampering Productivity; Is Upskilling the Answer?

"A well-crafted data strategy will highlight where specific skills need to be developed to achieve business objectives," said Michael Curry, president of data modernization at Rocket Software. He explained that since a data strategy typically involves both risk mitigation and value realization, it's important to consider skill gaps on both sides. Kjell Carlsson, head of AI strategy at Domino Data Labs, said better data prep, analysis, and visualization skills would help organizations become more data-driven and make better decisions that would significantly improve growth and curtail waste. "Imbuing your workforce with better prompt engineering skills will help them code, research, and write vastly more efficiently," he said. "A well-crafted data strategy will highlight where specific skills need to be developed to achieve business objectives," said Michael Curry, president of data modernization at Rocket Software. He explained that since a data strategy typically involves both risk mitigation and value realization, it's important to consider skill gaps on both sides. ... "Imbuing your workforce with better prompt engineering skills will help them code, research, and write vastly more efficiently," he said.



Quote for the day:

"Leadership should be born out of the understanding of the needs of those who would be affected by it." -- Marian Anderson

No comments:

Post a Comment