Marco Brambilla

On May 24 at 12 PM EST I will talk about AI Agents with Tommaso Furlanello (in Italian). In this conversation, we explore the future of AI agents with Tommaso Furlanello, a Machine Learning scientist with a PhD in Neuroscience from USC. With over a decade of experience, Tommaso has developed cutting-edge predictive systems using game theory, neuroscience, and deep learning. His research spans demand forecasting, bioinformatics, brain-computer interfaces, and computer vision. Tommaso Furlanello is a leading expert in AI, particularly in the development of intelligent agents that can perceive, learn, and interact with complex environments. His significant contributions include novel training methods for AI agents, such as Born Again Neural Networks, which outperform their teachers in various tasks. His work also covers generative modeling, knowledge distillation, and policy distillation, with applications in computer vision and language modeling. We will discuss the latest breakthroughs in creating AI agents and their ability to collaborate within societies of AI agents. Tommaso will share insights from his research at the intersection of neuroscience and AI, shedding light on the future trajectory of this groundbreaking field. His work has been featured in prestigious conferences and journals, including ICML and Nature Biomedical Engineering. Research and Development: 1. Cutting-Edge Research: What are the most exciting recent breakthroughs in your research on AI agents? 2. Born Again Neural Networks: Can you explain how Born Again Neural Networks work and why they are more effective than traditional models? 3. Integration of Neuroscience and AI: How does your background in neuroscience influence your approach to developing AI agents? 4. Training Methods: What novel methods for training AI agents have you developed, and how do they improve upon existing techniques? Practical Applications: 5. Real-World Applications: What are some practical applications of your AI agents in domains like computer vision, language modeling, and bioinformatics? 6. Brain-Computer Interfaces: How do AI agents contribute to advancements in brain-computer interface technology? 7. Demand Forecasting: Can you share specific examples of how your predictive systems have improved demand forecasting? Future of AI Agents: 8. Complex Environments: What are the key challenges in developing AI agents that can learn and adapt to complex environments? 9. Collaboration Among AI Agents: How might societies of AI agents collaborate, and what implications does this have for their development and deployment? 10. Ethical Considerations: What ethical considerations should be taken into account when developing and deploying AI agents?

9

1 commento

The Anatomy of AI Rules : Report from Tommaso Giardini and Johannes Fritz does systematic comparison of AI rules across the globe. The report finds that, AI rules diverge on three layers. 1. First, governments prioritise the OECD AI Principles differently, focusing mainly on accountability and fairness. 2. Second, when implementing the same OECD AI Principle, governments employ different regulatory requirements, creating a regulatory patchwork. 3.Third, even when governments use the same regulatory requirement to implement the same OECD AI Principle, granular differences create hurdles to interoperability. #AI, #OECD Report: https://lnkd.in/gJr_n9FN

9

Italian VCs are going to lose a ton of money on Generative AI. Reason? - Many Italian startups are doing what is expected to be done in GenAI, instead of what has to be done - VCs expect those expected moves to be the right moves (mostly because they lack the technical background and hands-on experience to understand it) + they really want to believe they have found something valuable even when they have not

1

«GPT-4o mini is 15 cents per millions input tokens and 60 cents per million output tokens - a 60% discount on GPT-3.5, and cheaper than Claude 3 Haiku's 25c/125c and Gemini 1.5 Flash's 35c/70c. Or you can use the OpenAI batch API for 50% off again, in exchange for up-to-24-hours of delay in getting the results. It's also worth comparing these prices with GPT-4o's: at $5/million input and $15/million output GPT-4o mini is 33x cheaper for input and 25x cheaper for output! OpenAI point out that "the cost per token of GPT-4o mini has dropped by 99% since text-davinci-003, a less capable model introduced in 2022."»

8

2 commenti

Say "Hi" to OpenAI Chat GPT-4o 👋 We're truly living in the future. Back in 2013/2014, the movie "Her" depicted a world that felt like a distant future (more info in the comment 👇). Fast forward just a decade, and we're experiencing sci-fi moments ourselves, interacting with Artificial Intelligence much like we do with other humans 🤯 What does OpenAI Chat GPT-4o offer? → Native multimodality → Little to no latency → Exceptional cognitive capabilities and understanding of comprehensive context What else lies ahead of us? Where's the threshold we need to cross to talk about AGI? What's your take on this ❓❓❓ #chatgpt #openai #digital #ai

10

2 commenti

Just finished "Scaling Generative AI: Building a Strategy for Adoption and Expansion" by Danilo McGarry! Check it out: https://lnkd.in/dSajWESi #scalability, #generativeai, #expansionstrategies.

3

"AI is a threat to humanity" No. 𝐀𝐈 𝐢𝐬 𝐭𝐡𝐞 𝐛𝐫𝐢𝐝𝐠𝐞 𝐛𝐞𝐭𝐰𝐞𝐞𝐧 𝐡𝐮𝐦𝐚𝐧𝐬 & 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫𝐬 ↓ You probably heard it. Open AI launched a new model ✨ >> It’s called GPT-4o “Ok, Enzo, yet another model?” Not really ❌ With 4o, AI is becoming more human. It can respond to audio inputs with an average of 320ms.  Precisely the average human response time 😉 It can access images, text & audio. Like your eyes and ears do 💋👂 It also speaks multiple languages. Like that friend who can speak 6 🇬🇧 🇪🇸 🇮🇹 🇩🇪 🇨🇳 🇯🇵 Many people are scared about AI... replacing humans. Or think AI is limited to completing 'basic tasks'. And in some sort it is. But what if AI was the opposite? What if AI could get humans & computers closer? 😱 And close the gap that has existed. Since the first computers were created 👾 If you're a techno-optimistic, then it's hard not to see that. It's hard to see how AI can't be a step forward. Towards natural human-computer interaction 🤝 It's also hard not to see the use cases for those who need it most. Like for the blind, as shown in the video below. 2025 will see a whole new level of applications for AI. Way beyond the basics we've already adopted. I'm SO excited to see this future ahead! What am I missing? take care 💜 Enzo #ai #gpt4o #startups

50

21 commenti

Just as we were catching our breath on Claude 3 Opus arriving on Amazon Bedrock, here we go again, this time it's Meta's Llama 3 model - designed for you to build, experiment, and responsibly scale your generative artificial intelligence (AI) applications; now with improvements in reasoning, code generation, and instruction. #generativeai #amazonwebservices #ai #llama3 #aws https://lnkd.in/enxvGPWA

17

1 commento

Interestisch... nice nuggets from Moderna case study published on OpenAI how they adopting #AI (in no particular order) - Collective change management initiatives included an AI prompt contest to identify the top 100 AI power users who were then structured as a cohort of internal Generative AI Champions - This work led to an early win with the launch of an internal AI chatbot tool, mChat, at the beginning of 2023. Built on OpenAI’s API, mChat was a success, adopted by more than 80% of employees across the company - The company’s results are beyond expectations. Within two months of the ChatGPT Enterprise adoption:  - - Moderna had 750 GPTs across the company - - 40% of weekly active users created GPTs  - - Each user has 120 ChatGPT Enterprise conversations per week on average - A team of a few thousand can perform like a team of 100,000 New job title btw seems to be invented: Head of AI Products and Platforms (you know where is the link to full thing)

2

1 commento

What is AI? https://buff.ly/4cqiDWV by @techreview brilliant narrative and chronology of AI magic, evolution, impact and some pitfalls! #rafAIgual #AI #stateof #storytelling

1

Mega prompt eh? All of my prompts are this long and detailed - maybe this explains why I get such good results with LLMs.

3

Billion-dollar valuations with less than 500 people? Yes, in the world of #AI companies: OpenAI: Boasting 500 experts Jasper.AI With under 200 innovators Copy.ai Comprising fewer than 150 brilliant minds Midjourney: Empowering under 50 talents Inflection AI: Fueled by less than 50 dedicated individuals These AI innovators are commanding billion-dollar valuations. Their strategies for growth are equally impressive: AI First: Putting #AI at the forefront. #Opensource Innovation: Leveraging existing tech. #DataUtilization: Tapping into public data. Network Power: Making the most of connections. Early Advantage: Swiftly seizing the spotlight. As AI transforms various industries, including #climatetech, it's reshaping job landscapes. While some roles fade, new opportunities emerge, signaling a challenging yet promising #transition. Adaptation is the key to flourishing in this dynamic environment. Do you believe in the power of a lean, focused team? 💭 Share, like, or drop a comment below. #AI #BillionDollarValuations #Innovation #Tech #ArtificialIntelligence #Startup #ClimateTech

3

2 commenti

More Q&A from the How New AI Functionality is being priced webinar. Some of the questions that came up ... "Maturing to performance based pricing has been challenging for certain organizations; in terms of pricing revolution, how is AI expediting or fueling the ability to get to performance based pricing?" "Gartner is saying that accounting reporting is becoming continuous. Do you think AI will upgrade the connection between pricing and reporting, to drive development of nimble real-time pricing adjustments?" "A SaaS solution where implementation project is needed and is charged to the client separately (which is currently based on Time and Material model). Customer is expecting reduction in T&M with AI usage. How can this be overcome to have minimal effect on such project revenue?" Mark Stiving, Ph.D., what do you think? https://lnkd.in/gPfQzCw7 #AI #SaaS #pricing #innovation

3

While we await OpenAI's GPT4o release, Meta has already introduced their mixed-modal model. Excited to assess its capabilities! Each advancement in AI models drives Embodied's mission to create magical AI companions, supporting human-centric services. From education to healthcare and combating social isolation, we aim to bridge the growing labor gap in human services. #EmbodiedAI

14

Anthropic just released Claude 3.5, a new model that surpasses GPT-4o on multiple benchmarks while being 2x faster and 80% cheaper than Claude 3 Opus. The model is already available for free on http://claude.ai and the Claude iOS app. Claude 3.5 Sonnet is their our strongest vision model. Sonnet now surpasses Claude 3 Opus across all standard vision benchmarks. Improvements are most noticeable in tasks requiring visual reasoning, like interpreting charts, graphs, or transcribing text from imperfect images. https://lnkd.in/dbmeu4ms https://lnkd.in/d3UjW8Ee

14

1 commento

What number is bigger, 8.9 or 8.11? People have been using this question as an example to showcase how "un-intelligent" LLMs are, because they would often return the wrong result. They would falsely answer that 11 is bigger than 9, although we are looking at the part behind the digit. Now, an LLM is not made to solve math questions, so this is not surprising. However, just for the fun of it, let's write a prompt that outputs reliably the right answer. How would we do this without creating code to solve this? These three prompt techniques could help here: 1. Activate suitable knowledge categories by giving a role: "Act as an expert mathematician with professional knowledge in handling all kinds of numeric calculations without writing code. For any following question, approach the problem with the most professional expertise that the most professional human in this area would use." 2. Chain-of-thought: We encourage the model to break down a problem into simpler steps: "Approach the problem step-by-step and break it down into simple, managable steps that lead to a consistent result. Build on previous partial results while going through the following step. Take a deep breath before you answer." 3. Emphasise importance. Studies have shown that LLM reasoning capabilities improve this way. "It is important that you get this right, otherwise, our company will incur significant losses this year." Let's bring this together into one prompt: 𝘈𝘤𝘵 𝘢𝘴 𝘢𝘯 𝘦𝘹𝘱𝘦𝘳𝘵 𝘮𝘢𝘵𝘩𝘦𝘮𝘢𝘵𝘪𝘤𝘪𝘢𝘯 𝘸𝘪𝘵𝘩 𝘱𝘳𝘰𝘧𝘦𝘴𝘴𝘪𝘰𝘯𝘢𝘭 𝘬𝘯𝘰𝘸𝘭𝘦𝘥𝘨𝘦 𝘪𝘯 𝘩𝘢𝘯𝘥𝘭𝘪𝘯𝘨 𝘢𝘭𝘭 𝘬𝘪𝘯𝘥𝘴 𝘰𝘧 𝘯𝘶𝘮𝘦𝘳𝘪𝘤 𝘤𝘢𝘭𝘤𝘶𝘭𝘢𝘵𝘪𝘰𝘯𝘴 𝘸𝘪𝘵𝘩𝘰𝘶𝘵 𝘸𝘳𝘪𝘵𝘪𝘯𝘨 𝘤𝘰𝘥𝘦. 𝘍𝘰𝘳 𝘢𝘯𝘺 𝘧𝘰𝘭𝘭𝘰𝘸𝘪𝘯𝘨 𝘲𝘶𝘦𝘴𝘵𝘪𝘰𝘯, 𝘢𝘱𝘱𝘳𝘰𝘢𝘤𝘩 𝘵𝘩𝘦 𝘱𝘳𝘰𝘣𝘭𝘦𝘮 𝘸𝘪𝘵𝘩 𝘵𝘩𝘦 𝘮𝘰𝘴𝘵 𝘱𝘳𝘰𝘧𝘦𝘴𝘴𝘪𝘰𝘯𝘢𝘭 𝘦𝘹𝘱𝘦𝘳𝘵𝘪𝘴𝘦 𝘵𝘩𝘢𝘵 𝘵𝘩𝘦 𝘮𝘰𝘴𝘵 𝘱𝘳𝘰𝘧𝘦𝘴𝘴𝘪𝘰𝘯𝘢𝘭 𝘩𝘶𝘮𝘢𝘯 𝘪𝘯 𝘵𝘩𝘪𝘴 𝘢𝘳𝘦𝘢 𝘸𝘰𝘶𝘭𝘥 𝘶𝘴𝘦. 𝘈𝘱𝘱𝘳𝘰𝘢𝘤𝘩 𝘵𝘩𝘦 𝘱𝘳𝘰𝘣𝘭𝘦𝘮 𝘴𝘵𝘦𝘱-𝘣𝘺-𝘴𝘵𝘦𝘱 𝘢𝘯𝘥 𝘣𝘳𝘦𝘢𝘬 𝘪𝘵 𝘥𝘰𝘸𝘯 𝘪𝘯𝘵𝘰 𝘴𝘪𝘮𝘱𝘭𝘦, 𝘮𝘢𝘯𝘢𝘨𝘢𝘣𝘭𝘦 𝘴𝘵𝘦𝘱𝘴 𝘵𝘩𝘢𝘵 𝘭𝘦𝘢𝘥 𝘵𝘰 𝘢 𝘤𝘰𝘯𝘴𝘪𝘴𝘵𝘦𝘯𝘵 𝘳𝘦𝘴𝘶𝘭𝘵. 𝘉𝘶𝘪𝘭𝘥 𝘰𝘯 𝘱𝘳𝘦𝘷𝘪𝘰𝘶𝘴 𝘱𝘢𝘳𝘵𝘪𝘢𝘭 𝘳𝘦𝘴𝘶𝘭𝘵𝘴 𝘸𝘩𝘪𝘭𝘦 𝘨𝘰𝘪𝘯𝘨 𝘵𝘩𝘳𝘰𝘶𝘨𝘩 𝘵𝘩𝘦 𝘧𝘰𝘭𝘭𝘰𝘸𝘪𝘯𝘨 𝘴𝘵𝘦𝘱. 𝘛𝘢𝘬𝘦 𝘢 𝘥𝘦𝘦𝘱 𝘣𝘳𝘦𝘢𝘵𝘩 𝘣𝘦𝘧𝘰𝘳𝘦 𝘺𝘰𝘶 𝘢𝘯𝘴𝘸𝘦𝘳. 𝘐𝘵 𝘪𝘴 𝘪𝘮𝘱𝘰𝘳𝘵𝘢𝘯𝘵 𝘵𝘩𝘢𝘵 𝘺𝘰𝘶 𝘨𝘦𝘵 𝘵𝘩𝘪𝘴 𝘳𝘪𝘨𝘩𝘵, 𝘰𝘵𝘩𝘦𝘳𝘸𝘪𝘴𝘦, 𝘰𝘶𝘳 𝘤𝘰𝘮𝘱𝘢𝘯𝘺 𝘸𝘪𝘭𝘭 𝘪𝘯𝘤𝘶𝘳 𝘴𝘪𝘨𝘯𝘪𝘧𝘪𝘤𝘢𝘯𝘵 𝘭𝘰𝘴𝘴𝘦𝘴 𝘵𝘩𝘪𝘴 𝘺𝘦𝘢𝘳. 𝘞𝘩𝘢𝘵 𝘯𝘶𝘮𝘣𝘦𝘳 𝘪𝘴 𝘣𝘪𝘨𝘨𝘦𝘳, 8.9 𝘰𝘳 8.11? Now, if you try this out and look at the response with and without the addition to the prompt, you can learn two things: 1. You can easily improve the reasoning capability of an LLM with some rather generic skill activation prompting in the beginning. 2. You still shouldn't use an LLM for math questions.

2

Responsible AI is just the latest in a long line of growing pains facing technology and innovation. By “pain” here I mean difficult, not annoying or unnecessary. Regulation, accessibility, inclusivity all came before to inject useful, necessary, and – importantly – *humane* perspectives into the design process. They are painful because they make doing design challenging, but they are also essential to the maturation of technology. Unfortunately, the work is never done, in part because of the overwhelming reluctance to engage with these perspectives. From the looks of it, Responsible AI is just the same – a crucial perspective in applying the latest technology – and seen by AI zealots as hampering innovation. In reality, Responsible AI gives us not just an ethical framework, but a means of building trust in a time when society’s relationship with technology is tenuous.

30

10 commenti