Mikael Loefstrand

🚀 AI Accessibility: The Trend of Decreasing Costs Continues OpenAI's recent introduction of GPT-4o mini represents another step in the ongoing trend of increasing AI accessibility. This new model continues the progression toward more affordable AI capabilities: 💡 GPT-4o mini: $0.15 per 1M input tokens, $0.60 per 1M output tokens 📉 Represents a 30x reduction in cost compared to GPT-4o 🔥 Even more impressively, it's a 60x reduction when compared to GPT-4-turbo This dramatic decrease in pricing follows a pattern we've seen with other transformative technologies. Just as transistors and smartphones evolved from expensive, specialized tools to affordable, everyday devices, AI models are becoming smaller, cheaper, and more accessible. As the technology proves its utility, we can expect this trend of decreasing costs and increasing availability to continue, potentially leading to widespread adoption across various sectors. 📱💻🤖 This ongoing AI proliferation presents both opportunities and challenges: Potential benefits: ✅ Enhanced accessibility for researchers and developers ✅ Increased innovation across various industries ✅ Broader application of AI in solving complex global problems Considerations: ⚠️ Need for robust security measures ⚠️ Importance of addressing algorithmic bias ⚠️ Potential economic impacts and workforce transitions As AI becomes more accessible, we need to be honest and wide-eyed about its potential and risks. 🔍 The democratization of AI is exciting, but it comes with responsibilities: 🛡️ We must implement robust guardrails to prevent exploitation by bad actors. 🤝 Collaboration between tech companies, policymakers, and ethicists is crucial. 📚 Ongoing education about AI capabilities and limitations is essential for all. There's no denying the transformative power of AI, but its widespread adoption requires careful navigation. How can we balance innovation with responsible development? 🤔 #AITrends #TechAccessibility #ResponsibleAI #AIEthics

The Google DORA 2024 report is out. In this report, Google DORA explores , amongst other topics, the downstream impacts of AI (which I suppose is GPT's or similar concepts - I'm open to hearing otherwise - , the report doesn't specifically define this) on the concept of Software delivery performance (broken down into Software delivery throughput and Software delivery stability). I couldn't help but ponder (and agree with) this quote: "Contrary to our expectations, our findings indicate that AI adoption is negatively impacting software delivery performance. We see that the effect on delivery throughput is small, but likely negative (an estimated 1.5% reduction for every 25% increase in AI adoption). The negative impact on delivery stability is larger (an estimated 7.2% reduction for every 25% increase in AI adoption)". 7.2% reduction in software stability per 25% increase in AI adoption, is quite a bit. My guess is that the large majority of the respondents are not running test driven development, or have their bases covered by unit, functional and integrational testing. If you have your tests (and coverage) in place when you develop software, I hypothesize that you are safeguarded and any AI will at worst come up with wrong suggestions (that wont have an impact downstream because your CI will reject it), at best it will speed up software development. It depends: Can the developer verify the AI suggested code? Personally I am still vary of leaning heavily on GPT's in my software development. I prefer reading documentation by those that built the software or tooling I'm using. What's your experience with AI in software development? #devops & #softwaredelivery @ The Tech Collective

11

1 Comment

Great post and discussion about vector vs graph based RAG for information retrieval in enterprise search applications using gen AI/LLMs.

2

There’s a Cambrian Explosion of Voice AI Agents. They’re proliferating in these categories: existing business call volume, novel LLM-enabled Voice AI applications, and Voice User Interfaces. 1. Existing Business Call Volume. We’re seeing Voice AI eat the market for business phone calls. Examples are Infer (YC S21)'s lead qualification agents, Domu's customer surveys, 11x's inbound sales agents, the afterhours agents and appointment scheduling agents people are building on Phonely (YC S24), and so on.  2. LLM-enabled Voice AI applications. These are products that weren’t possible before LLMs. My favorite is Boardy, an AI that introduces you to other professionals in its network. Voice AI feels like an artist’s medium. I’ve built a few in this category for fun, like an AI podcast host that interviews you and generates social content based on the interview. 3. Voice User Interfaces. Soon it will become glaringly obvious how ineffective it is to furiously thumb-tap our phones to death. Fluidworks.ai is building a voice interface for product demos and user onboarding. Whispr Flow lets you dictate to any device with a keyboard shortcut. Soon Generative Siri will handily accomplish tasks within third-party phone apps. If you’re building any of these, you need to know what’s happening in your calls. We can help! Check out the link in the comments.

44

6 Comments

Supply chain enforcement doesn't seem to be stopping anything. In fact when you place resource constraints around groups of motivated people they will actually out innovate you. #Energy efficiency and #computation performance means a lot and amplifies through a system. This will be interesting to watch. 01.ai trained the #7 model in the world for $3M pre-train cost. And the inference price is $0.14/million tokens! Summary: As Chinese entities do not have access to tens of thousands of advanced AI GPUs from companies like Nvidia, companies from this country must innovate to train their advanced AI models. Kai-Fu Lee, the founder and head of 01.ai, said this week that his company has trained one of its advanced AI models using 2,000 GPUs with just $3 million. “The thing that shocks my friends in the Silicon Valley is not just our performance, but that we trained the model with only $3 million and GPT-4 was trained with $80 to $100 million,” said Kai-Fu Lee (via @tsarnick). “GPT-5 is rumored to be trained with about a billion dollars. […] We believe in scaling law, but when you do excellent detailed engineering, it is not the case. […] As a company in China, first, we have limited access to GPUs due to the U.S. regulations [and a valuation disadvantage compared to American AI companies].” #china #ai #efficiency #llm #usa #gpu #ml Kelly Cohen Zachary Huhn Summer Crenshaw Scott Phillips Lohith Srikanth Pentapalli Carl Fraik Paul Baier Alden Do Rosario Eduardo Ordax Joe Oquist💡 Lori Mazor Monica Turner Christi Brown Dave Hatter Josette Riep Michael Beck, MBA Brian Burke Gregg Burke Cory Warfield https://lnkd.in/d5Kx8N94

22

17 Comments

The AI Product Manager's Handbook: https://amzn.to/3D6OyPd via Packt Publishing (greatly expanded and updated 2nd Edition, just released) 𝙒𝙝𝙖𝙩 𝙮𝙤𝙪 𝙬𝙞𝙡𝙡 𝙡𝙚𝙖𝙧𝙣: 🟠Plan your AI PM roadmap and navigate your career with clarity and confidence. 🟣Gain a foundational understanding of AI/ML capabilities. 🟠Align your product strategy, nurture your team, and navigate the ongoing challenges of cost, tech, compliance, and risk management. 🟣Identify pitfalls and green flags for optimal commercialization. 🟠Separate hype from reality and identify quick wins for AI enablement and GenAI. 🟣Understand how to develop and manage both native and evolving AI products. 🟠Benchmark product success from a holistic perspective. ———— #EnterpriseAI #DataScience #GenAI #MachineLearning #ML #DeepLearning #DataScientist #CDO #CTO #CMO #AIStrategy #GenerativeAI #AI

57

10 Comments

🚀 SAFe® in the Government Sector: Elevating Strategy and Alignment! 🚀 The public sector faces unique challenges when it comes to strategy alignment, transparency, and adapting to rapid changes. That’s why ICON Agility Services, Broadcom, and Scaled Agile, Inc. are coming together to present: 📅 SAFe Made Simple: How SAFe and Rally Brings Visibility & Alignment to Public Sector Strategy 🕑 Tuesday, December 17th | 2-3pm EST Join ICON’s Charlene M. Cuenca, SPCT & SAFe Fellow, and Scaled Agile’s Adam Mattis, SPCT & SAFe Fellow, as they guide you through how the Scaled Agile Framework and Broadcom’s Rally software can bring powerful visibility, accountability, and strategic alignment to government organizations. Gain insights into using agile practices to drive success at scale and support mission-driven initiatives. 👥 Don’t miss this opportunity to learn how the right tools and frameworks can simplify your organization’s journey to agility. Secure your spot today! Register here: https://lnkd.in/gARZ6tw9 Stella O'Brien Katie Snelson Daylon Walton SPC6, SDP, SOCC, EBAS Beth Elliott #SAFe #PublicSectorAgility #GovernmentAgile #ScaledAgile #Broadcom #ICONAgilityServices #Rally

42

15 Comments

Looks like AI is having its "horse-and-buggy to Model T" moment, folks! 🚗💨 An MIT spinoff is revolutionizing the AI world with their non-transformer models, and they're already outpacing the current frontrunners. We're watching the next chapter of tech evolution unfold in real-time. These new Liquid Foundation Models are like the Model T of AI –more efficient, easier to mass-produce, and still outperforming the luxury models on the track. They're processing up to 1 million tokens while keeping the memory usage lower than my coffee intake on a Monday morning. Could this be our ticket to AI-powered cybersecurity that doesn't need its own data center to run? The possibilities are as exciting as finding an open road on your morning commute. One thing's for sure - AI is evolving faster than we can update our threat models. Time to buckle up and embrace the ride! What's your take? Are we witnessing the dawn of AI's new efficient age? https://lnkd.in/gh_ERJEm #AIRevolution #Cybersecurity #LiquidAI #AIInnovation #TechEvolution #FutureOfAI #MachineLearning #AIFuture #TechTrends

13

1 Comment

Beyond "Technology-Driven Myopia" in AI and Tech: A Message for Users and Providers In my 25+ years in this field, I've seen technology become infinitely more powerful and useful. Yet, the path to real value remains consistent. It's not about chasing the latest shiny object – even when that object is as transformative as (Gen)AI. Too often, we fall into the trap of "technology-driven myopia", losing sight of the actual problems we're trying to solve. For users of technology, the key steps to unlocking value haven't changed: - Start with the desired outcome - Frame the problem correctly - Determine necessary actions/initiatives and their optimal execution order - Identify the tools needed to get there Often, the solution doesn't lie in new tech at all, but in: - Process changes - Better training, change management, or governance - Properly leveraging existing tech that hasn't been fully adopted But this truism applies equally to those providing the technology. There's an intense focus right now on the "battle over foundation models," with investors pouring obscene amounts of money into these firms. But to what end? Foundation models are essentially the silicon chips of today – they're merely one of many pieces in the supply chain for solutions to real-world problems. While impressive, they're also very expensive and unsustainably energy-intensive. The real innovation – and the path to creating meaningful and sustainable value – lies in building useful applications that solve concrete, real-world problems. For technology providers (and investors), this means: - Look beyond the hype of foundation models - Focus on understanding and addressing real-world challenges - Innovate in the application layer, where solutions meet specific needs - Prioritize sustainability and efficiency in your technological developments The potential to build useful things that solve real-world problems is where true innovation will flourish. So again, whether you're a user or a provider of technology, start with the problem to create meaningful and sustainable value. In the AI era, our collective challenge isn't a lack of technological capability. It's ensuring that we're asking the right questions, defining our goals clearly, and choosing or developing the most appropriate tools to achieve them. Let's move beyond the "technology-driven myopia" mindset. Instead, let's focus on building comprehensive strategies and solutions that align technology with real business and societal needs, driving sustainable value creation for all stakeholders. If you're navigating these challenges in your organization or looking to collaborate on innovative, problem-centric approaches to AI and data-driven technology, I'd be glad to lend my expertise. Reach out at [email protected] #AIStrategy #TechnologyValueCreation #ResponsibleInnovation #SustainableTech

34

5 Comments

How Eight Authors Ignited an AI Revolution... The Genesis of Transformers In 2017, eight pioneering researchers from Google changed the course of artificial intelligence with their groundbreaking paper, "Attention Is All You Need." This seminal work introduced the world to transformers, a new AI architecture that has fundamentally altered how machines understand and generate human language. The impact of this innovation is evident in cutting-edge technologies like ChatGPT and various advanced graphic generators. The story begins with a collaborative effort marked by a unique approach to academic authorship—eschewing traditional hierarchies for a randomly ordered list signaling equal contribution. This decision reflected the collective nature of their breakthrough. As the paper's seventh anniversary approaches, the technology it introduced continues to evolve, underpinning the most sophisticated AI systems that challenge and exceed human capabilities in many domains. Notably, all eight authors have since moved on from Google, leveraging the transformer technology they developed to launch successful startups, each now valued in the millions or billions. This migration underscores a broader trend in the tech industry, where dynamic shifts and the pursuit of innovative ideas often lead to significant transformations outside the confines of established corporate environments. As we look toward the future, the story of the Transformer Eight not only highlights the potential of collaborative innovation but also serves as a testament to the transformative power of attention-based models. The ongoing evolution of this technology promises to redefine our interaction with machines, making AI an even more integral part of our digital lives. Here are some tailored hashtags that can enhance the reach and engagement of a LinkedIn post about the AI revolution sparked by the transformer technology: #AITransformation #MachineLearning #AI #DeepLearning #TechLeadership #CAIO https://lnkd.in/gXqE8tsB

2

OpenAI revelas its “failing” projection on the scaling laws for gpt-5 (Orion). Ilya Sutskever recently told Reuters that results from scaling up pre-training—the phase of training an AI model that uses a vast amount of unlabeled data to understand language patterns and structures—have plateaued. Today we know OpenAI has revealed visually the projections they did for Orion (gpt5) and don’t understand what has could fail yet! ♻️ Repost if you like it and follow me for a pragmatic (and funny) take on AI! #ai #openai #agi #llms

269

31 Comments

For generative AI users here is a good introduction to Llama and its limitations. "Like every big tech company these days, Meta has its own flagship generative AI model, called Llama. Llama is somewhat unique among major models in that it’s “open,” meaning developers can download and use it however they please (with certain limitations). That’s in contrast to models like Anthropic’s Claude, OpenAI’s GPT-4o (which powers ChatGPT) and Google’s Gemini, which can only be accessed via APIs."

10

2 Comments

Since 2023, 92% of businesses and firms have adopted AI in their practices. (MIT Sloan Review) ❗ However, a group of researchers identified a threshold of at least 25% AI adoption intensity before companies could see an ROI. (MIT Sloan) AI alone does not immediately guarantee an ROI. That’s because firms are not maximizing their AI investments effectively. Instead, using complementary technologies and strategies alongside AI will secure greater gains for your organization. The same group of researchers found that when these tools and strategies were used, firms saw a 63.7% increase in revenue growth: 🔷 Database systems expand data collection, management, and storage capabilities to boost outcomes. 🔷 Cloud computing provides enough computational power to perform tasks and analyze data. 🔷 Research & development (R&D) accelerates the development of new algorithms and applications to meet and align with your market needs. 💫 Having a smarter investment strategy and infrastructure when adopting AI is the key to achieving your growth. It will take some time, but it will pay off. What do you think? #revenuegrowth #growthstrategy #artificialintelligence #startups

41

2 Comments

Great insights from Ethan Mollick on Apple's announcement. Some highlights: * Apple's on-device models are slightly worse than GPT-3.5, and their cloud models are slightly better than GPT-3.5, so for the really complicated stuff they give users the option to call GPT-4o. * A narrow AI that just does what you want could be more useful than a wide AI with many possibilities, but the narrow AI will have a smaller effect on your productivity * Apple's strong emphasis on privacy will get people to trust it more than the other AIs https://lnkd.in/dCtNNWRK

3

o1 vs GPT-4o: Is OpenAI's latest model that Good? Learn more in the video👇 #o1 #gpt4o #openai

15