Stiv Sterjo

“𝗗𝗼𝗲𝘀 𝗚𝗲𝗻𝗔𝗜 𝗥𝗲𝗽𝗹𝗮𝗰𝗲 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗛𝗮𝗿𝗺𝗼𝗻𝗶𝘇𝗮𝘁𝗶𝗼𝗻? 𝗡𝗼𝘁 𝗦𝗼 𝗙𝗮𝘀𝘁!” 🤓 The rise of generative AI, particularly LLMs, has sparked a misconception: that these models can bypass the need for semantic harmonization through business information models. However, the reality is quite the opposite—semantic harmonization remains critical, even with GenAI as a powerful enabler. 💡 And here are the reasons why: 1. 𝙀𝙣𝙝𝙖𝙣𝙘𝙚𝙙 𝙇𝙇𝙈 𝙋𝙚𝙧𝙛𝙤𝙧𝙢𝙖𝙣𝙘𝙚: LLMs deliver higher accuracy and efficiency when operating on semantically harmonized data. Without it, they expend unnecessary effort trying to “understand” data meanings. 2. 𝙉𝙤𝙩 𝙀𝙫𝙚𝙧𝙮 𝙐𝙨𝙚 𝘾𝙖𝙨𝙚 𝙞𝙨 𝘼𝙄-𝙙𝙧𝙞𝙫𝙚𝙣: Generative AI isn’t a one-size-fits-all solution. Many tasks still rely on standardized, well-structured data products beyond AI’s immediate scope. 3. 𝙁𝙤𝙪𝙣𝙙𝙖𝙩𝙞𝙤𝙣𝙖𝙡 𝙛𝙤𝙧 𝘿𝙖𝙩𝙖 𝙈𝙖𝙣𝙖𝙜𝙚𝙢𝙚𝙣𝙩: Harmonized semantics play a pivotal role in automating processes like master data management and ensuring data quality—cornerstones of any robust data strategy. 4. 𝙂𝙚𝙣𝘼𝙄 𝙖𝙨 𝙖 𝘾𝙖𝙩𝙖𝙡𝙮𝙨𝙩: While GenAI doesn’t replace the need for semantic harmonization, it can accelerate the process. It’s a game-changer for optimizing models and mapping them to physical data representations. 𝙎𝙤 𝙬𝙝𝙖𝙩 𝙙𝙤𝙚𝙨 𝙩𝙝𝙖𝙩 𝙢𝙚𝙖𝙣 𝙛𝙤𝙧 𝙮𝙤𝙪? To truly conquer your data challenges, embrace the synergy: use semantic harmonization as the foundation and GenAI as the accelerator. Together, they’re the ultimate duo for unlocking the full potential of your data. #DigitalTransformation #GenAI #DataStrategy #SemanticHarmonization #AI

45

7 Comments

One of the most promising directions in software engineering is merging stateful architectures with LLMs to handle complex, multi-step workflows. While LLMs excel at one-step answers, they struggle with multi-hop questions requiring sequential logic and memory. Recent advancements, like O1 Preview’s “chain-of-thought” reasoning, offer a structured approach to multi-step processes, reducing hallucination risks—yet scalability challenges persist. Configuring FSMs (finite state machines) to manage unique workflows remains labor-intensive, limiting scalability. Recent studies address this from various technical approaches: 𝟏. 𝐒𝐭𝐚𝐭𝐞𝐅𝐥𝐨𝐰: This framework organizes multi-step tasks by defining each stage of a process as an FSM state, transitioning based on logical rules or model-driven decisions. For instance, in SQL-based benchmarks, StateFlow drives a linear progression through query parsing, optimization, and validation states. This configuration achieved success rates up to 28% higher on benchmarks like InterCode SQL and task-based datasets. Additionally, StateFlow’s structure delivered substantial cost savings—lowering computation by 5x in SQL tasks and 3x in ALFWorld task workflows—by reducing unnecessary iterations within states. 𝟐. 𝐆𝐮𝐢𝐝𝐞𝐝 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧 𝐅𝐫𝐚𝐦𝐞𝐰𝐨𝐫𝐤𝐬: This method constrains LLM output using regular expressions and context-free grammars (CFGs), enabling strict adherence to syntax rules with minimal overhead. By creating a token-level index for constrained vocabulary, the framework brings token selection to O(1) complexity, allowing rapid selection of context-appropriate outputs while maintaining structural accuracy. For outputs requiring precision, like Python code or JSON, the framework demonstrated a high retention of syntax accuracy without a drop in response speed. 𝟑. 𝐋𝐋𝐌-𝐒𝐀𝐏 (𝐒𝐢𝐭𝐮𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐀𝐰𝐚𝐫𝐞𝐧𝐞𝐬𝐬-𝐁𝐚𝐬𝐞𝐝 𝐏𝐥𝐚𝐧𝐧𝐢𝐧𝐠): This framework combines two LLM agents—LLMgen for FSM generation and LLMeval for iterative evaluation—to refine complex, safety-critical planning tasks. Each plan iteration incorporates feedback on situational awareness, allowing LLM-SAP to anticipate possible hazards and adjust plans accordingly. Tested across 24 hazardous scenarios (e.g., child safety scenarios around household hazards), LLM-SAP achieved an RBS score of 1.21, a notable improvement in handling real-world complexities where safety nuances and interaction dynamics are key. These studies mark progress, but gaps remain. Manual FSM configurations limit scalability, and real-time performance can lag in high-variance environments. LLM-SAP’s multi-agent cycles demand significant resources, limiting rapid adjustments. Yet, the research focus on multi-step reasoning and context responsiveness provides a foundation for scalable LLM-driven architectures—if configuration and resource challenges are resolved.

168

7 Comments

🚨 [AI RESEARCH] "Regulating Multifunctionality," by Cary Coglianese & Colton Crum, is an EXCELLENT read for everyone in AI governance and a great way to kick off 2025. 👉 Important quotes to understand the paper: 1️⃣ Use Heterogeneity in AI "The uses for which AI can be deployed are virtually limitless. It is this use heterogeneity—especially with the multifunctional nature of foundation models and generative AI tools—that presents the most substantial challenge when it comes to regulating AI. Use heterogeneity arises from a) the existence of a vast array of different AI tools being designed for specific but varied purposes in mind, and b) AI tools that are themselves designed for multiple purposes, perhaps only some of which can be fully anticipated." 2️⃣ Heterogeneity & AI Governance "The heterogeneity of problems posed by AI creates a core challenge for AI governance. This is true for both varied single-function AI tools as well as for multifunctional AI tools. The problems presented by a specialized AI tool that misreads MRIs used to identify cancer, for example, will obviously be quite different from those presented by a specialized AI tool assisting in the automatic braking of an autonomous vehicle. Likewise, the problems created by an LLM tool hallucinating about legal cases will be different from these same foundational LLM tools put to use in social media that contribute to teenagers engaging in self-harm. The question thus becomes how society should approach regulation when both uses and problems can be so varied. In other words, how does society regulate the AI equivalent of the Swiss army knife?" 3️⃣ Flexible Regulatory Strategies "Instead of rigid, prescriptive rules, the future of AI regulation will likely depend on more flexible regulatory strategies. At least 4 more feasible regulatory strategies exist that could be considered for the governance of multifunctional AI: performance standards, disclosure regulation, ex post liability, and management-based regulation. (...) Moreover, regulators can rely on a combination of these strategies depending on their overarching goals and context. Importantly, none of these strategies necessarily demand a 'one size fits all' approach." 4️⃣ Vigilance & Agility "(...) But selecting a particular regulatory strategy—or a combination of strategies—will be only the beginning. Perhaps the most difficult work in regulating multifunctionality will lie in providing ongoing vigilance and agility. Effective AI governance will require constantly adapting, issuing alerts, and prodding action. Regulators need to see themselves as overseers of dynamic AI ecosystems, staying flexible, remaining vigilant, and always seeking ways to improve." 👉 Download the paper below. 👉 NEVER MISS my updates on AI policy, compliance & regulation, including excellent research papers: join 46,900+ readers who subscribe to my newsletter (link below). #AI #AIGovernance #Multifunctionality #AIRegulation #AICompliance

56

5 Comments

#EarlyCareer #talk at #IJCAI2024 by🗣️Julia Neidhart, Technische Universität Wien: Transforming Recommender Systems: Balancing Personalization, Fairness, and Human Values ➡️https://lnkd.in/dK3Nieyr Abstract: Recent advancements in recommender systems highlight metrics beyond accuracy, including diversity, serendipity and fairness. This paper examines the integration of Large Language Models into recommender systems, focusing on challenges such as interpretability, bias, and user autonomy. LLMs offer enhanced personalization capabilities but also introduce concerns regarding transparency and ethical implications. This work discusses ongoing research efforts aiming to develop transparent, fair, and contextually aware systems. Our approach seeks to prioritize user wellbeing and inclusivity, contributing to a more equitable and functional digital environment through advanced technologies and interdisciplinary insights. #XAI #LLMs #InclusiveTech #NextGenAI #AI TU Wien Informatics

3

🚀 Unlocking Artificial Agency in LLMs 🚀 In the evolving landscape of AI, the discourse around Large Language Models (LLMs) is taking a pivotal turn towards the realization of Artificial Agency. But what truly defines agency in the realm of AI, and how close are we to achieving this milestone with LLMs? 🤔 Artificial agency refers to the ability of AI systems not just to perform tasks, but to autonomously set goals, make decisions, and adapt in dynamic environments—traits currently explored within the framework of LLMs like GPT-4 and its successors. 🧠 A recent study, "Artificial Agency and Large Language Models" (https://lnkd.in/gMrxyzPY), delves into whether our advanced LLMs embody genuine agency. While these models excel at understanding and generating nuanced text, they lack true agents' self-awareness and persistent intentionality. Critics argue they function on sophisticated pattern recognition rather than authentic cognitive reasoning. 🔍 To transcend current limitations, the research advocates integrating Neuro-Symbolic AI, which marries neural networks with symbolic reasoning. By doing so, we can hope to endow AI with true adaptability and intentional cognition, unlocking roles in real-world applications where decision autonomy is crucial. 🗣 According to AI luminary Fei-Fei Li, "Aspiring towards agency in AI isn't just about autonomous action—it's about cultivating systems that can genuinely collaborate with humans in ways that enhance both parties' capabilities." Similarly, Yann LeCun champions adding grounded perception and intuitive reasoning to the AI narrative. 💼 The race is on to build more interactive and adaptive AI systems. Industry giants are iterating on LLMs to push boundaries in context-awareness and engagement, moving towards an era where AI doesn't just understand instructions—it reasons through them. 🔗 What are your thoughts on the trajectory of AI systems in achieving true agency? How will these advancements change the landscape of AI-driven solutions in your sector? 👩‍💼 Share your insights below! #ArtificialAgency #LargeLanguageModels #AIResearch #Innovation #MachineLearning #AITechnology #FutureOfAI

20

4 Comments

The 3-Step Guide to Making Your Company AI-Forward 🚀: 1. Provide Access to ChatGPT Plus / Claude Pro($20/user/month) 2. Empower People to Experiment 💡 3. Encourage Sharing 📢 Sure, it sounds simple—maybe too simple—but many executives are overthinking it, bogged down by "CEO guides." Every day, people are discovering nuanced AI workflows that collectively save weeks of time across the company. This isn't a far-off dream; it’s happening now. Why wait until 2040? The AI tools you need are available today. And guess what? Early adopters are eager to experiment. The excitement around AI tools is still very much alive. Let your team experiment. Give them a platform to share their insights. It's that simple! Transform your business with AI today! 🚀 #ArtificialIntelligence #Startups #Technology #AIFuture #Innovation

5

1 Comment

AI practitioners and leaders, do you know about 𝘀𝗽𝗲𝗰𝘂𝗹𝗮𝘁𝗶𝘃𝗲 𝗱𝗲𝗰𝗼𝗱𝗶𝗻𝗴? As a fan of running open-source AI, I wanted to try speculative decoding for the 𝟭𝟬-𝟯𝟬𝟬% 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁 in inference speed that it provides. Techniques for improving inference are commonly used in cloud deployments but rarely in local AI tools. So, without a lot of thought, I forked llama.cpp—a popular local inference backend—to add it to their local server functionality. Something must be in the air, because a few days after I started working on this, they released the feature themselves! Here's a 𝘀𝗶𝗺𝗽𝗹𝗲 𝗲𝘅𝗽𝗹𝗮𝗻𝗮𝘁𝗶𝗼𝗻 of speculative decoding along with some 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗶𝗱𝗲𝗮𝘀 for using it. 𝗪𝗵𝗮𝘁 𝗜𝘀 𝗦𝗽𝗲𝗰𝘂𝗹𝗮𝘁𝗶𝘃𝗲 𝗗𝗲𝗰𝗼𝗱𝗶𝗻𝗴? Speculative decoding is a recent technique for addressing a fundamental performance issue with LLMs: • LLMs 𝗽𝗿𝗼𝗰𝗲𝘀𝘀 𝗶𝗻𝗽𝘂𝘁 𝘁𝗼𝗸𝗲𝗻𝘀 𝘀𝗲𝗾𝘂𝗲𝗻𝘁𝗶𝗮𝗹𝗹𝘆, but the hardware they run on is optimized for 𝗺𝗮𝘀𝘀𝗶𝘃𝗲 𝗽𝗮𝗿𝗮𝗹𝗹𝗲𝗹 𝗰𝗼𝗺𝗽𝘂𝘁𝗮𝘁𝗶𝗼𝗻. This mismatch wastes performance potential. 𝗛𝗼𝘄 𝗗𝗼𝗲𝘀 𝗦𝗽𝗲𝗰𝘂𝗹𝗮𝘁𝗶𝘃𝗲 𝗗𝗲𝗰𝗼𝗱𝗶𝗻𝗴 𝗙𝗶𝘅 𝗧𝗵𝗶𝘀? 1. A 𝘁𝗶𝗻𝘆 “𝗱𝗿𝗮𝗳𝘁” 𝗺𝗼𝗱𝗲𝗹 generates 𝗡 𝗰𝗮𝗻𝗱𝗶𝗱𝗮𝘁𝗲 𝗼𝘂𝘁𝗽𝘂𝘁 𝘁𝗼𝗸𝗲𝗻𝘀. Example for N=5: • Input: "The quick brown fox" • Output: "[jumps, over, the, lazy, dog]" 2. The large model evaluates these tokens in a 𝘀𝗶𝗻𝗴𝗹𝗲 𝗽𝗮𝘀𝘀:   • If the tokens match what the larger model would have produced,     they’re accepted, and generation continues from Step 1—𝗡 𝘁𝗼𝗸𝗲𝗻𝘀     𝗳𝘂𝗿𝘁𝗵𝗲𝗿 𝗮𝗹𝗼𝗻𝗴.   • If the tokens are rejected, the large model produces a token     as usual, and Step 1 repeats. This allows the large model to produce 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗲 𝘁𝗼𝗸𝗲𝗻𝘀 𝗽𝗲𝗿 𝗽𝗮𝘀𝘀 without changing output! 𝗥𝗲𝗾𝘂𝗶𝗿𝗲𝗺𝗲𝗻𝘁𝘀 Speculative Decoding Works If: 1. The draft model uses the 𝘀𝗮𝗺𝗲 𝘁𝗼𝗸𝗲𝗻 𝘃𝗼𝗰𝗮𝗯𝘂𝗹𝗮𝗿𝘆 as the larger model. 2. The draft model is a 𝗴𝗼𝗼𝗱 𝗺𝗶𝗺𝗶𝗰 of the larger model--for example, Llama 1B for Llama 70B. 𝗪𝗵𝗲𝗿𝗲 𝗜𝘁 𝗦𝗵𝗶𝗻𝗲𝘀 The performance improvement is especially noticeable for 𝘀𝘂𝗺𝗺𝗮𝗿𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝘁𝗮𝘀𝗸𝘀. Why? Imagine a summary that quotes the input: the draft model will likely generate the exact same tokens as the larger model for quoted words—leading to high token acceptance rates. 𝗧𝗵𝗲 𝗧𝗿𝗮𝗱𝗲𝗼𝗳𝗳 The main downside of speculative decoding is that you now need to 𝗿𝘂𝗻 𝘁𝘄𝗼 𝗺𝗼𝗱𝗲𝗹𝘀: the small draft model and the large main model. This adds: • A bit more 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁 𝗰𝗼𝗺𝗽𝗹𝗲𝘅𝗶𝘁𝘆. • A small increase in 𝗺𝗲𝗺𝗼𝗿𝘆 𝗰𝗼𝘀𝘁𝘀. But if inference speed is critical to you or your customers, it’s absolutely worth trying! I’ll share more practical tips for running speculative decoding soon. Are you using speculative decoding? I’d love to hear about your experience!

9

3 Comments

Excellent post on the difference between the o and 01 models. Thanks, Ashish Bhatia, for this insight into the two models.

8

+1000 to this below Now is the time to be doubling down on your domain expertise to outlive AI tools. Human fundamentals are here to stay.

70

2 Comments

Interesting! Read carefully this post from Tony Seal Since starting my PhD in 2004, I’ve advocated for a structured knowledge representation hierarchy—Data ➔ Information ➔ Knowledge ➔ Ontology/Knowledge Graph—as the optimal approach for feeding AI models (not just Generative AI). In my experience, this combination is the only one that consistently delivers controlled accuracy (validated by rigorous testing) and efficiency gains (improving productivity, reducing costs, and minimizing CONQ). Soon, I’ll be publishing a paper detailing the construction and scaling of a specialized Knowledge Graph Retrieval-Augmented Generation (RAG) system, alongside a comparative study. Excited to share insights on the path to more reliable and efficient AI. Stay tuned!

20

AI Agents aren't going to be easy, and the problem is the LLMs. LLMs are trained to provide an answer to a question or an output for a task. "write this code", "categorize this list". They are not, however, great at making decisions - navigating a fork in the road. In fact, they're trained to *not* make decisions for you. What makes an agent an agent is its ability to decide, and the current models are fairly terrible at it. There isn't as much training data on making decisions. The fine tuning isn't being done on decision making. For a while, agents are going to be rudimentary at best.

16

1 Comment

One year on... and executives remain bullish on #GenerativeAI. But the real story is not about hype vs exportations but the details of adoption, the value #GenAI is generating and the emerging divide between leaders and laggards ● 𝗦𝗹𝗼𝘄𝗹𝘆 𝗺𝗼𝘃𝗶𝗻𝗴 𝗳𝗿𝗼𝗺 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝘁𝗼 𝘀𝗰𝗮𝗹𝗶𝗻𝗴 Majority of companies is generating real business value from GenAI, but most remain stuck in experimentation model. Thus far only around 20% of companies manage to scale their GenAI AI bets. The highest percentage of scaling success is seen in customer service (25%) and software Development (25%), followed by new natural language interaction interfaces (20%), IT (18%) and new products (17%) ● 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝘃𝗮𝗹𝘂𝗲 𝗶𝘀 𝗴𝗲𝘁𝘁𝗶𝗻𝗴 𝘁𝗮𝗻𝗴𝗶𝗯𝗹𝗲 Vast majority of companies say that Generative AI has already delivered business value: In 59% of cases it has met expected goals and in 14% even exceeded them. It has only fallen short of companies' expectations 25% of the time. Interestingly for 10% of firms the technology has even been a catalyst for broader change, which is a number I expect to steadily increase as more companies succeed in their scale efforts ● 𝗟𝗲𝗮𝗱𝗲𝗿𝘀 𝘃𝘀. 𝗹𝗮𝗴𝗴𝗮𝗿𝗱𝘀: 𝗧𝗵𝗲 𝗲𝗺𝗲𝗿𝗴𝗶𝗻𝗴 (𝗮𝗻𝗱 𝘄𝗶𝗱𝗲𝗻𝗶𝗻𝗴) 𝗱𝗶𝘃𝗶𝗱𝗲 There is a large gap between organizations that are scaling (Gen)AI effectively and those that aren't. Leading adopters are seeing bottom-line impact across 90% of use cases and they’re moving beyond a focus on productivity to reimagine their core offerings and build new business models. They also have 1.7x larger AI budgets, which they translate into 1.2x higher bottom line impact and they are 2x less likely to be held back by internal barriers to adoption internal barriers to adoption Great survey and insights by Bain & Company. Link to full study in the comments. CC: Tony, Theodora, Conor, JOY, Noelle, Matt, Efi, Alex, Chuck, Conor, Arjun, Zvonimir, Mike, Marc, Dinis, Simon, John, Dr., Fabio, T., Ravit, Edgar, Eduardo, Tarja, Jaroslaw, Bernard, Pascal, Alex, Adam, Nicolas, Noelle, Dr. Debashis, Louis-François, Fabio, Prof. Dr., Dr., Tyler, Bugge, Ravi

191

14 Comments

OpenAI Employees Don’t Trust OpenAI 😱 https://lnkd.in/erEkdrcD Several current and former employees from OpenAI and Google DeepMind have publicly called for openness and transparency from AI companies. They have suggested that: 📣 employees shouldn’t be silenced if they have concerns about AI 📣 there needs to be better protection for employees who speak up about their AI concerns There is also concern that OpenAIs new safety division isn’t sufficiently independent because Sam Altman (the CEO) and other senior executives are members. This is a good call for more accountability and explainability in relation to AI models - something that all organisations that use or develop AI could commit to. It perhaps confirms that regulatory controls are insufficient. These concerns from Open AI employees are worrying. What do they know that they think we should know? 🤔 How serious are their concerns? Does your organisation have an AI policy that demonstrates transparency and openness to help with building digital trust? #airisks #OpenAI #womeninai

14

6 Comments

A good Agentic AI solution needs an arbitration and delegation logic with authority when 2 agents overlap in function and have equal rights to execute autonomous tasks defined by an unambiguous sequence of execution. Should a tie-breaker situation arise, which may be unavoidable, a deadlock resolver AI agent is needed with higher powers to override the default chain of command.

5

#Google #DeepMind researchers have introduced #InfAlign, a groundbreaking #ML framework designed to enhance language model alignment. InfAlign addresses the challenge of aligning generative #LLMs with #inference-time strategies, ensuring reliable and efficient #AI performance. Key benefits include: ✅️ Improved inference-time win rates. ✅️ Enhanced robustness and consistency. ✅️ Maintained computational efficiency. This advancement is a significant step towards more reliable and efficient AI systems 🚀🚀 #AIinnovation

1

1 Comment

Improving #RAG performance with #KnowledgeGraphs. Knowledge Graphs (KGs) represent information in an interconnected network of entities and relationships, enabling more complex reasoning across content. How do KGs augment retrieval? Explicit Facts — KGs preserve key details by capturing facts directly as nodes and edges, instead of condensed into opaque vectors. Contextual Details — KG entities possess rich attributes like descriptions, aliases, and metadata that provide crucial context. Network Structure — KGs capture real-world relationships - rules, hierarchies, timelines, and other connections - between entities. Multi-Hop Reasoning — Queries can traverse relationships, and infer across multiple steps, to connect and derive facts from diverse sources. Joint Reasoning — Entity Resolution can identify and link references that pertain to the same real-world object, enabling collective analysis. Explainable Relevance — Graph topology lets us transparently analyze the connections and relationships that determine why certain facts are retrieved as relevant. Personalization — KGs capture and tailor query results according to user attributes, context, and historical interactions. In sum, whereas RAG performs matching on disconnected nodes, KGs enable graph traversal search and retrieval of interconnected contextual, search for query-relevant facts, make the ranking process transparent, and encode structured facts, relationships, and context to enable complex, precise, multi-step reasoning. As a result, compared to pure vector search, KGs can improve relevance and explanatory power. Know more about improving RAG performance with knowledge graphs: https://lnkd.in/g2wPRQSX

81

Efficient. Fast. Autonomous. And one day it will erase humans: #AI I personally always said there is another perspective to artificial intelligence and the only thing that is super is the outcome for humans. Philosopher Nick Bostom has a new book, and it’s finally acknowledging the potential of a harmonious human-AI relationship and its problem solving capabilities. AI = augmented intelligence #design #ai #problemsolving #innovation #creativeai

11

1 Comment