Vitaly Kleban

Innovators can thrive in the AI era The AI landscape is vast, evolving, and will feel overwhelming for many. Amidst the noise, small companies hold the potential to create game-changing solutions tailored to real organisational needs. There’s plenty of room to follow a differentiated path in AI development. #AI #Innovation #Communication #ProjectManagement

1

🚀 Exciting news in the AI world! French AI lab Kyutai has just unveiled Moshi, an open-source GPT-4o competitor that can listen, hear, and speak in real-time! 😲 Moshi is a groundbreaking multimodal LLM with an impressive end-to-end latency of just 200ms. This Voice AI is so advanced that it even interrupted the speaker a few times! 😅 Unlike OpenAI’s GPT-4o, demoed in May, Moshi will be entirely open source, with the code, model, and paper set to be released soon. And here’s the most astonishing part: Kyutai developed this revolutionary voice & text model in just 4 months with a small team of only 8 brilliant engineers. French innovation at its finest! 🤯 This technology is set to transform our world, and we can’t wait to see how it will be integrated into our daily lives. #ArtificialIntelligence #AI #MachineLearning #DeepLearning #VoiceAI #OpenSource #GPT4o #MoshiAI #TechInnovation #FutureTech #AIRevolution #MultimodalAI #Kyutai #AIResearch #Technology #AICommunity #FrenchTech #AIModels #VoiceRecognition #RealTimeAI #AIInnovation #TechNews #FutureOfAI #AIEngineering #VoiceTechnology #AIAdvancements #TechUpdates #ArtificialIntelligenceNews #AIIndustry #AITransformation #NextGenAI #SmartTechnology #AITrends #OpenSourceAI #AIInAction #TechBreakthrough #InnovativeTech #DisruptiveTechnology #CuttingEdgeTech #AILeaders #VoiceAssistance #AIIntegration #AIProgress #EmergingTech #HighTech #TechRevolution #AIDevelopment #AIExploration #SmartAI #AIImplementation #GlobalTech #AIFuture

22

2 Comments

🚀 Looking to streamline fine-tuning for models like Llama 3, Mistral, and Mixtral? Our latest blog shows how Anyscale simplifies LLMs fine-tuning with Ray: 🔑 Key takeaways: • Scalable fine-tuning with LLM-Forge • Model Deployment with RayLLM Read more: https://lnkd.in/gQwD4A2U

34

Filash.io's family of algorithms suggests a comprehensive and adaptive approach to optimize file transfers across diverse network conditions. Here are the family members. 1. Latency Control Algorithm: This algorithm enables high-speed file transfers even over long distances, overcoming latency as high as 150 milliseconds with minimal slowdown compared to shorter delays like 20 milliseconds. Traditional protocols often suffer significant performance degradation with increasing latency, making this capability particularly noteworthy. 2. Packet Loss Control Algorithm: Filash.io's algorithms are claimed to handle very high packet loss rates, up to 50% and even simulated extreme conditions at 80% packet loss. This capability is crucial for maintaining usability and efficiency in networks prone to significant packet loss, such as certain satellite or wireless environments. 3. Jitter Control Algorithm: Designed specifically for wireless networks including WiFi, LTE, and satellite, this algorithm manages unpredictable jitter (variations in packet arrival times). Jitter control is essential for stabilizing throughput and ensuring consistent performance in these environments. 4. Bandwidth Probe Algorithm: Operating without human intervention, this algorithm autonomously probes and maximizes potential bandwidth speed while minimizing self-introduced packet loss. This feature is advantageous for dynamically adapting to changing network conditions without manual adjustment. 5. Fairness & Friendliness Control Algorithm: Ensures equitable bandwidth allocation among sessions while being proactive and friendly towards TCP-based traffic like web browsing. This promotes efficient network utilization and user experience across various applications sharing the network. 6. File Set Alignment Algorithm: Optimizes file transfer times by managing the transfer of mixed file sizes (from millions of small files to large files). By aligning file transfers to reduce impact on storage bandwidth, this algorithm aims to streamline end-to-end transfer times effectively. Filash.io's algorithms are to address key challenges in modern network environments, such as latency, packet loss, jitter, and fair bandwidth allocation, with a focus on maximizing performance and usability across diverse conditions. #filash #filetransfer #congestionckntrol #performance #jitter #latency #bandwidth #packetloss #wireless #satellite #fairness #friendliness #tcp #internet #cloudfileserver #cloud #fileserver

1

There are two interesting and relatively new/emerging revenue models in hardware space. (1) Software Enabled Hardware Hardware is sold at lower but slowly improving gross margins (eg 20-40%), and software recurring revenues layered on top. Software sells at higher margins, enabling blended margins to appear in 50%+ category. These are applied in situations where hardware is not expected to be replaced or greatly improved upon in deployment for a significant number of years to come. Consider your car, or military equipment, or even infrastructure equipment. As always, there are often multiple business/revenue models utilized by different players in the same industry. (2) Hardware As A Service Hardware and software are bundled and sold with annually recurring revenues. Often multi-year contracts are signed to be able to amortize hardware over time, and enable blended gross margins over the length of the contract to be high (approaching 60-75%). This is more common in situations where vendor takes responsibility to upgrade hardware as needed along the way to provide maximum and highest functionality to the customer. Hardware inventory costs are, over time, passed on to suppliers or third party financing parties in lieu of few gross margin points. Consider robots as a service in factories (ala Formic), or AI-enabled equipment (eg Evolv, Lumafield etc). Hardware innovation will require deeper investigation, experimentation, and understanding of business/revenue models to fit customer needs as well as to maximize value for the innovating company. Strong and experienced management teams work hard to minimize sales friction, inventory costs etc, but also maximize gross margins, customer stickiness, and ability to upsell during the life of a contract.

339

37 Comments

Here's an interesting bit of research to keep in your thoughts: According to IEEE Spectrum, the very same Large Language Model (#LLM) that power generative AI platforms such as #ChatGPT and others are actually capable of passing, even outperforming humans in Theory of Mind tests. Before you panic, this doesn't necessarily mean that LLMs actually do possess theory of mind or self consciousness -- but it does pique interest in theoretical purposes and exercises. In reality, while LLMs are scoring well, there is no real method for how to test for the existence of theory of mind -- so again, #AI here is just doing what it does best: #synthesis of data from many different datasets. What this does mean is that AI systems can better mimic and understand human psyche and anticipate what they need, not necesarily that it can make decisions or form thoughts in itself. Is this breakthrough exciting to you? How do you think AI technology will improve with this one aspect?

2

2 Comments

This is one of the best use cases I have seen for GenAI in a sector that is overburdened and in some cases still saddled with IT technical debt. What I find interesting from this CNBC article is that it highlights Google Microsoft NVIDIA and others when in fact much more innovation is occurring outside these companies with impressive results in production environments. Oracle is betting big on healthcare but its not mentioned in this piece. Oracle is building a massive campus in the Nashville area which is also home to various healthcare companies including HCA Healthcare, Community Health Systems, Acadia Healthcare and Vanderbilt University Medical Center. For more on why check out this article. https://lnkd.in/ehMqzkhP OCI Generative AI service includes support for Meta's Llama 2 model and Cohere's models, along with enhanced multilingual capabilities that support over one hundred languages. Oracle Cloud Infrastructure (OCI) introduces the Generative AI Agents Retrieval-Augmented Generation (RAG) service, designed to streamline operations by facilitating interactive and actionable conversations with diverse knowledge bases. The service, with an initial focus on OpenSearch, allows employees to query data across systems like ERP and HCM through natural language, enhancing decision-making with real-time data access and semantic matching. This approach democratizes data access within healthcare and other verticals, enabling users to extract and act on information without specialized skills, thus fostering efficiency and informed decision-making across various departments. Oracle use of retrieval-augmented generation (RAG) is vital. It allows Oracle to leverage specific information from the system of record to address relevant questions effectively. I discuss this in more detail in a research note I published in April after attending the Oracle analyst event in Redwood Shores. Message me directly if you want a free copy of it. https://lnkd.in/eT69n8WM

23

4 Comments

Exabits AI Compute Economy: Democratizing AI Access Exabits is reinventing AI compute economy through a dynamic and interconnected ecosystem of participants, including nodes and subnets. By fostering collaboration, decentralization, and accessibility, Exabits empowers contributors of all scales to shape the future of AI infrastructure actively. Tokenized GPU Power: Exabits leverages blockchain technology to tokenize GPU compute power. Users can earn rewards by contributing their GPUs to the network. This democratizes access to AI compute, allowing individuals and organizations to participate in the AI revolution. Currency of the Future: Exabits aims to create a currency backed by real-world compute resources. It’s not just about financial transactions; it’s about valuing computational work. By tokenizing GPUs, Exabits establishes a new paradigm for value exchange. Decentralization and Accessibility: Exabits promotes decentralization. A central authority does not control it; participants make decisions collectively. This ensures that even small contributors have a voice in shaping the AI infrastructure.

2

Open-source AI models such as Llama2, Llama 3, and Mistral are making strides toward ensuring AI technology isn't monopolized by a few companies. However, the primary barrier remains the hardware requirements and associated costs needed to run these models. Although it's not yet fully democratized—thanks to a continuing hardware monopoly by companies like NVIDIA, reminiscent of the old adage, "sell shovels during the gold rush"—AI is becoming increasingly accessible to most developers through AI platforms like https://huggingface.co/, https://openai.com/, https://www.anthropic.com/, https://groq.com/, ...

51

2 Comments

🚀 See how Handshake cut their LLM GPU costs by 50% with Anyscale. Discover how they: 💰 Reduced LLM GPU costs by 50% or more. 📈 Seamlessly scaled large language models (LLMs) without compromising performance. ⏱ Enhanced operational efficiency, enabling faster development cycles. Check out the full story here: https://lnkd.in/gSiKJDaY

37

2 Comments

New short course: Safe and Reliable AI via Guardrails! Learn to create production-ready, reliable LLM applications with guardrails in this new course, built in collaboration with Guardrails AI and taught by its CEO and co-founder, Shreya Rajpal. I see many companies worry about the reliability of LLM-based systems -- will they hallucinate a catastrophically bad response? -- which slows down investing in building them and transitioning prototypes to deployment. That LLMs generate probabilistic outputs has made them particularly hard to deploy in highly regulated industries or in safety-critical environments. Fortunately, there are good guardrail tools that give a significant new layer of control and reliability/safety. They act as a protective framework that can prevent your application from revealing incorrect, irrelevant, or confidential information, and they are an important part of what it takes to actually get prototypes to deployment. This course will walk you through common failure modes of LLM-powered applications (like hallucinations or revealing personally identifiable information). It will show you how to build guardrails from scratch to mitigate them. You’ll also learn how to access a variety of pre-built guardrails on the GuardrailsAI hub that are ready to integrate into your projects. You'll implement these guardrails in the context of a RAG-powered customer service chatbot for a small pizzeria. Specifically, you'll: - Explore common failure modes like hallucinations, going off-topic, revealing sensitive information, or responses that can harm the pizzeria's reputation. - Learn to mitigate these failure modes with input and output guards that check inputs and/or outputs - Create a guardrail to prevent the chatbot from discussing sensitive topics, such as a confidential project at the pizza shop - Detect hallucinations by ensuring responses are grounded in trusted documents - Add a Personal Identifiable Information (PII) guardrail to detect and redact sensitive information in user prompts and in LLM outputs - Set up a guardrail to limit the chatbot’s responses to topics relevant to the pizza shop, keeping interactions on-topic - Configure a guardrail that prevents your chatbot from mentioning any competitors using a name detection pipeline consisting of conditional logic that routes to an exact match or a threshold check with named entity recognition Guardrails are an important part of the practical building and deployment of LLM-based applications today. This course will show you how to make your applications more reliable and more ready for real-world deployment. Please sign up here: https://lnkd.in/gRU6qhzN

3,515

109 Comments

Nvidia's recent surge in revenue, driven by just a few major customers investing heavily in AI, underscores a critical point: those who invest in AI-compute power are pulling ahead. Giants like Amazon, Microsoft, Tesla and Google are reaping the rewards, but what about traditional European industry leaders? For companies like Volkswagen, Airbus, and Siemens, the potential of AI is enormous. However, without similar investments in AI infrastructure, they risk falling behind in an increasingly competitive landscape. You need compute to stay ahead. When will they start? Interested in how your company can stay ahead? Let's talk about making strategic investments in AI-compute power to ensure you remain competitive. #AI #Innovation #DigitalTransformation #AICompute #BusinessGrowth

5

NLP is the Key to Making AI Work for Everyone LLMs might not reach superintelligence, but AI doesn’t need to. By pairing NLP with practical tools, businesses can unlock AI’s potential without getting lost in the hype. This post shows front-end AI can live harmoniously in business operations. #AI #AIAssistants #DigitalTransformation #Innovation #NaturalLanguageProcessing 

5