Tobias Halloran

Smaller is better The need for powerful #LLMs has fueled their exponential growth. However, large LLMs come with a cost - they require significant computational resources and energy, leading to increased response latency. The largest models today have over 100 billion parameters and can take over 200 gigabytes to load. Industry applications and #AI giants are now shifting their focus towards smaller and faster models, prioritizing efficiency over maximum performance. Learn more:

12

#ObsessivelyEnthusiasticInvestor #UnreasonablyEnthusiasticInvestor #IrrationallyEnthusiasticInvestor at REFASHIOND Ventures: The Industrial Transformation Fund. We look for the people we believe are doing the best work in Industrial Transformation & Supply Chain Technology; Attempting the impossible. Then we go all-in on them. With obsessive and unreasonable enthusiasm. Join us. #IndustrialTransformation #SupplyChain #Innovation #Technology #Startups #VentureCapital #DemandTheImpossible #UnreasonableEnthusiasm #ObsessiveEnthusiasm #IrrationalEnthusiasm

15

1 Comment

Gavin Uberti (CEO Etched) is another talented Harvard drop out, who is betting on a silicon model of accelerated compute which only runs transformers. A great (very narrow) strategy which will have a 20x hardware performance advantage over NVidia Blackwell architecture. No details on their software stack strategy, unlikely to match NVIDIA Cuda. HOWEVER, if the industry move on from Transformers as the central tenet of AI models, Etched will be in a very bad place indeed. The threats and risks to NVIDIA investors starting to emerge. If we also consider that NVIDIA is trading at 21 times forward PS ratio, (Microsoft only 11 times) these are risks to the NASDAQ as a whole.

2

Dan thank you for sharing this. As the UK general election approaches, the rise of deepfakes and disinformation is significantly impacting voter perceptions. Blackbird.AI’s analysis reveals that AI-generated content, including deepfake audio and visual clips, is being widely circulated to undermine political candidates and sway public opinion. Additionally, bot-like amplification on social media is spreading misleading narratives and false information. These tactics are not only creating confusion but also eroding trust in the democratic process. To combat these challenges, platforms like Blackbird.AI are essential in identifying and mitigating the spread of disinformation. The US will definitely see these same tactics in the near future!

3

In the evolving landscape of AI technology, choosing between open-source and proprietary models is more than just a technical decision, its a strategic one. As industries and developers weigh the benefits of accessibility and innovation against security and performance, the future of AI might just lie in a hybrid model that leverages the strengths of both.

16

1 Comment

Promising news for Qualcomm investors: https://lnkd.in/eB6B38ig Readers wanting to learn more about Qaulcomm's Cloud AI 100 product should visit https://lnkd.in/eK2AVf-k I noted 3 points: 1) This product is a candidate for the Inference Acceleration AI job. 2) The product can be added to rack mounted servers with a PCI card slot. 3) The product runs on an ASIC How many other early stage AI businesses are looking for custom operation from ASICs? What impact will news like this have on investor sentiment in Nvidia, Intel, AMD based on their participation in the programmable chip business segment? #asic #qualcomm #g42 #intel #amd #nvidia #programmablechips https://lnkd.in/eB6B38ig

4

LLMs in Autonomous Vehicles - Design Considerations Language Models (LLMs) have the potential to revolutionize autonomous vehicles by enhancing human-machine interaction, improving navigation and route planning, ensuring safety compliance, and enabling sophisticated environmental perception and decision-making. By understanding natural language, LLMs can process voice commands, provide real-time feedback, interpret complex directions, and analyze contextual information, making autonomous vehicles more intuitive, user-friendly, and efficient. However, several key challenges need to be addressed to fully harness their potential: Data Privacy: Ensuring the privacy and security of user data is paramount. Autonomous vehicles collect vast amounts of data, including personal information, driving habits, and real-time location. Protecting this data from breaches and misuse is crucial to maintaining user trust. Processing Power: Integrating LLMs into autonomous vehicles requires significant processing power and efficient resource management. LLMs are computationally intensive, and running them in real-time within a vehicle presents substantial challenges. Efficient hardware acceleration and optimizing algorithms are necessary to ensure smooth performance. Reliability: LLMs must be reliable and robust, especially in critical driving scenarios. Autonomous vehicles operate in dynamic environments, and the LLMs must accurately understand and respond to a wide range of inputs and situations. High accuracy, minimal error rates, and robust fail-safes are vital to prevent any misinterpretation that could lead to safety issues. Cost: The implementation and maintenance of advanced LLMs can be costly. For example, developing and training an LLM can require substantial computational resources, such as high-performance GPUs and cloud computing services, leading to expenses that can reach hundreds of thousands of dollars. Additionally, continuous updates and improvements to keep the model current and effective add to the financial burden. Balancing the benefits with these costs is critical for making the technology accessible and affordable. In summary, LLMs have the potential to greatly enhance the capabilities of autonomous vehicles, making them more intuitive, safe, and user-friendly. As technology progresses, the integration of LLMs into autonomous systems will likely become more seamless and widespread. Addressing these challenges is essential to fully realize the benefits of LLMs in advancing autonomous driving technology. #AutonomousVehicles #AI #MachineLearning #DataPrivacy #Innovation #FutureTech #Robotics #Transportation #TechTrends RoshAi

112

2 Comments

Google Cloud and Anthropic made some exciting announcements: 📣 Claude 3 Opus, Anthropic’s most intelligent model, is now generally available on Vertex AI ⚙️ Tool use is available for the Claude 3 model family on Vertex AI, enabling Claude to act as an autonomous agent 📈Provisioned throughput is available for the Claude 3 model family on Vertex AI

19

1 Comment

Google DeepMind has now released Gemma 2 for researchers and developers globally compatible with major AI frameworks to suit your preferred tools and workflows. Just like the original launch, Gemma 2 gives developers and researchers the ability to share and commercialise their innovations. It has been great to see how people are already using Gemma to build and deploy exciting AI models, and we hope Gemma 2 will continue to unlock new levels of performance and creations. Read more about our latest lightweight state-of-the-art open model here→ https://dpmd.ai/45Q6yba

82

2 Comments

How are you using LLMs in your business? Generative AI is advancing at a rapid pace, with the major Large Language Model (LLM) providers announcing new models and features almost every month. As these technologies become embedded into intelligent vertical applications, we move towards the future of mainstream enterprise AI. In my recent AI Atlas, I took a step back to dive into the current state of LLMs, including which features are helping to move the needle for enterprise adoption, the features still needing improvement, and the most common use cases we’re seeing in the field today. https://lnkd.in/eMhDA6f9

29

2 Comments

Interesting piece on how vector databases (an esoteric technology) have finally found use in the generative AI era ---- Vector databases have been around for decades, but are now emerging as something of an industry standard for AI businesses to use alongside a technique called retrieval-augmented generation, or RAG. When combined, businesses can link their private data with large-language models like OpenAI’s GPT-4, allowing the AI to perform data analysis, summarization and other tasks on their data. Without them, AI models are limited to what they have learned from their initial training on public data online, up to a certain point in time, and are more prone to factual errors called “hallucinations.” Vector databases are different from traditional databases with columns and rows because they are designed to store a massive amount of data as “vectors,” or numerical representations of the raw data New York-based startup Pinecone was an early entrant in the vector database AI space. Valued at $750 million after raising $100 million from Andreessen Horowitz and other investors last April, Pinecone has been building its vector database since 2019—a fortunate few years before OpenAI’s ChatGPT kicked off the AI gold rush in late 2022. ----- This one of those things you may learn about in computer science, along with red-black trees, and never ever use! #ai #chatgpt #llms #databases

8

3 Comments

Enterprise Automations, RPAs, AI, Agents - where and how do they intersect and overlap? Great coverage in this article from Insight Partners on this topic: https://lnkd.in/ghiNGtC8

59

3 Comments

SoftBank has acknowledged AI will be the source of future growth, Mayoshi Son is attempting to reshape SoftBank, and its risk-taking Vision Funds, to remain relevant in what he considers humanity’s next stage. In my view humanity’s next edge is in sustainability and SMC - Sustainable Metal Cloud is best placed to support the delivery of AI sustainably with less energy required to deliver the promise of the future.

9

2 Comments

Thanks for sharing Brent Hoberman … wonderful (but not surprising) how personal networks and amazing mentors have helped the remarkable Alex Kendall and Wayve team become #GlobalChampions plus .. “The company also benefited from five aspects of being in the UK: 🎓Academic thought leadership, with great Professors such as Zoubin Ghahramani and Roberto Cipolla 👩‍💻Talent density, thanks to multiple leading global companies w large UK AI centres 🏛️Regulatory regime & supportive government, with testing allowed on UK roads & autonomous vehicle bill going through parliament 🇬🇧Leading corporates with UK HQs willing to partner 🌎Skilled immigration regime allowing immigrant founders to move to the UK I hope that Wayve acts as a beacon to other international founders, to show how a once contrarian approach can attract the world’s best brains and capital, and how the UK can focus on technology opportunities and encourage broader entrepreneurship across the country to drive more growth. Wayve announces Europe's largest AI raise - $1bn+ led by SoftBank, Microsoft and NVIDIA. This should be celebrated as a pivotal deal for the UK to stake its claim to be a global leader in the platform shift for embodied AI.”  #ScaleUp #AI #7Cs #Leadership #EveryoneDeservesACoachMentor

4

1 Comment

Google has launched a suite of open-source tools to support generative AI development, including MaxDiffusion for diffusion models, JetStream for cost-efficient text-generating models, and MaxText for customizable AI models. These tools, optimized for Google's TPUs and Nvidia GPUs, aim to accelerate and democratize AI projects. In partnership with Hugging Face, Google also introduced Optimum TPU, simplifying the deployment of AI models on TPUs. This move signals Google's commitment to open collaboration and innovation in the AI space, promising to lower barriers and foster a community of development. Read more from Techcrunch here => https://neoswap.cc/nje #GoogleCloudNext #AI #OpenSource #Innovation

1

1 Comment

The Ten Steps to Metaverse Interoperability. Step One: Standardisation of Data Formats (An extract from the upcoming KZero Worldswide report, The Ten Steps to Metaverse Interoperability). ▶ Introduction: The emergence of Web3 technologies and the concept of a ‘connected Metaverse’ have sparked considerable excitement, speculation and debate about the future of virtual worlds. Defined by the convergence of virtual worlds, augmented reality (AR), virtual reality (VR) and blockchain technology, the Web3 Metaverse represents a universe of interconnected digital spaces where users can socialise, create and transact. At the heart of this vision lies the concept of Interoperability, explained as: The seamless integration of diverse virtual world platforms to enable shared experiences and asset exchange. ▶ Step One: Standardisation of Data Formats: In the quest for achieving full interoperability within the Web3 Metaverse, one of the fundamental challenges lies in the standardisation of data formats. Establishing common standards for asset formats, including 3D models, textures and animations is essential for enabling seamless asset exchange between virtual worlds. However, this endeavour is fraught with both technical and commercial hurdles that must be addressed to unlock the full potential of interoperability. ▶ The Diversity of Existing Data Formats One of the primary technical hurdles in standardising data formats is the diversity of existing formats used for asset creation across different virtual worlds and platforms. Each platform may have its own proprietary formats for assets, making it challenging to establish a universal standard that accommodates all variations. ▶ In the scenario where an avatar wishes to move a baseball cap from one virtual world to another, the diversity of existing formats used across different virtual worlds and platforms poses significant interoperability challenges. ▶ These challenges stem from the fact that each virtual world or platform may utilise its own proprietary data asset formats for representing 3D models, textures, animations and other digital content. Here's how this diversity of formats manifests as interoperability issues: Format Compatibility:, Asset Representation, Data Loss and Corruption, Metadata and Additional Information, Licensing and Intellectual Property, User Experience. Link to full article is in the comments. #metaverse #Interoperability #Blockchain #Web3

8

5 Comments

Agents Vs RPA: With the advent of new AI Agentic architectures, will agents replace RPA? Or are these agents fundamentally different? RPA is good for repetitive high volume tasks - it requires more deterministic data and process steps - what robots do on factory floors like pick stuff at Amazon warehouses. The thinking and process has already been done i.e. this is how you process a standard insurance claim. The downside is that because it is deterministic, It is likely unable to handle exceptions smartly i.e. how to tackle incorrectly filled insurance forms? If you add a new field to the Insurance form, you have to rebuild the Robot. Agent’s on the other hand are not deterministic - you give it a goal, an agent observes what humans do, learn, and figure out what data and process steps might be needed to achieve the goal. Hence, they are much better at figuring out how to handle exceptions, how to adapt as things change. My take: We are far from AGI. Agents will start with a narrow set of relatively simple tasks tied to specific systems. That’s how we are building at Statisfy. Incrementally, as models and agentic systems improve, we layer on complexity across data, process and decisions. What do you think? Credit to David Luan (Adept)’s for the inspiration behind this post. Listen to the full 20 minute VC podcast with David here. https://lnkd.in/gWsAMiMt

13

1 Comment

🌐 Today I am releasing a Serena brand new report exploring the dynamics of Commercial Open-Source Software (#COSS). In an era where open-source technologies power the backbone of our digital infrastructure, understanding the symbiotic relationship between #OpenSource ethos and commercial innovation is a must for technology builders, buyers, and investors. Our report takes a deep dive into the landscape of VC-backed commercial open-source companies revealing how market dynamics, community engagement, and strategic foundations interconnect. 🔗 Link to the report in the comments 🔍 Here's a sneak peek of the findings: 1️⃣ COSS companies, comprising only 10% of top GitHub projects, are 50x more likely to dominate the charts. 2️⃣ While the U.S. leads in VC funding (82%), EMEA's share is rapidly rising, fueled by ventures like IA ventures. 3️⃣ The most popular categories (eg. devtools) are not necessarily the ones that receive the most funding (eg. modern data stack). This underscores the importance of understanding that popularity based on metrics like stars and contributors may not align with economic attractivity. 4️⃣ Companies seniority counts, as older COSS companies boast more loyal and vibrant communities, attributed to trust and reliability over time 5️⃣ Nearly half of established projects engage with foundations, enhancing governance and long-term sustainability, thus attracting investors. 6️⃣ Despite hot debates around hybrid licenses, permissive licenses remain the go-to for COSS organization, facilitating broad adoption and seamless integration into proprietary products. Ready to dive deeper into the #COSS ecosystem? Please find the link to the report below in the comments section. A massive shout out to the contributors of the report including but not limited to JB Onofré Damien Garros Miguel Valdes Faura Arnaud GIULIANI Bertrand Diard Gilles Rouvier and a special thanks to Antoine Giacomini for helping out massively on that one.

142

27 Comments

Nvidia is on a shopping spree in Israel - Nvidia has announced it will purchase Run:ai, a deal estimated at around $700 million. This will be its biggest acquisition in Israel since buying Mellanox for $6.9 billion in March 2019. Run:ai is an AI platform that enables enterprise customers to manage and optimize their compute infrastructure, whether on premises, in the cloud or in hybrid environments. Run:ai has developed an orchestration and virtualization software layer tailored to the unique needs of AI workloads running on GPUs and similar chipsets. Run:ai’s Kubernetes-based container platform for AI clouds efficiently pools and shares GPUs by automatically assigning the necessary amount of computing power – from fractions of GPUs, to multiple GPUs, to multiple nodes of GPUs. “Run:ai has been a close collaborator with NVIDIA since 2020 and we share a passion for helping our customers make the most of their infrastructure” Following this acquisition, Nvidia is set to acquire Israeli AI startup Deci. Deci is an end-to-end deep learning development platform that helps AI developers build, optimize, and deploy AI models to any environment such as cloud, edge, and mobile. Nvidia also has a R&D center in Israel, which was first established in 2016. It has grown by about 100%, with over 4,000 employees today. #nvidia #israel #acquisition #ai #run:ai #deci

15

2 Comments