António Afonso

Want a preview of what I'm going to be talking about in my keynote for RubyConf? You can catch a glimpse of it here!

10

A data-driven approach to driving wound healing outcomes Marcella Gomez, University of California, Santa Cruz https://lnkd.in/gAkWZYJT Meet: https://lnkd.in/g4AAWkcK YouTube Stream: https://lnkd.in/gwtqaTmf Talk details: https://lnkd.in/gWn3B3cU Subscribe for talk announcements: https://lnkd.in/g5ciuNuX Abstract: Precision medicine requires an ability to predict the response of an individual to a prescribed treatment regimen a priori. Thus, advancement in the field is challenged by a lack of predictive models and, arguably, a lack of time-series information for a highly dynamic system. Here, we discuss work in wound healing, for accelerating wound closure. We argue that timing of treatments is as critical to consider as the choice of drug or therapy. Due to system size and complexity, data-driven methods need to be explored to develop multi-dimensional quantifiable indicators tracking systemic changes. In this work I discuss how bioelectronic devices enhanced with deep learning can help facilitate real-time sensing and actuation for automated decisions in treatment for wound healing and preliminary work in transcriptomic based classification of wound states. Bio: Marcella M. Gomez is an associate professor at UC Santa Cruz in the department of Applied Mathematics and Associate Dean for Diversity, Equity, and Inclusion for Baskin Engineering. She received her PhD from Caltech in 2015 and a B.S. from UC Berkeley in 2009; both degrees in Mechanical Engineering. Her research interests are in the broad field of bio-control leveraging methods in machine learning and control theory. Applications range from controlling single-cell response to driving complex systems such as wound healing. #modeling #simulation #medicine #healing #trauma #predictivemodeling #datascience #data #woundhealing #biology #bioengineering #systemanalysis #sensing #omics

3

🚀 Law as Code: Progress Update 🚀 Yes, the law can be expressed as code. And yes, LLMs can scale the translation en masse. I've reached the proof-of-concept stage with Json-Lawgic, and while the results are promising, key questions remain: 1️⃣ How accurate is the expression of laws? 👉 Subjective and evolving, but I've built an evaluation tool to begin testing this! 2️⃣ What are the practical applications? 👉 #RulesAsCode systems aren't new - this approach can be a means to build them more quickly. Furthermore, legal chatbots could call validated code as a tool to enhance their accuracy. 3️⃣ What's next? 👉 Feedback and collaboration so I'm not building in a vacuum! Huge thank you to Jason Morris and Damien Riehl for their advice and direction - this project is progressing a lot further than I expected thanks to your insights! How do I plan to enhance this project further? 🔧 Prototype a chatbot using encoded laws as tools   🔗 Build rule linking across statutes   📈 Scale with open-source LLMs (so OpenAI doesn't break my bank 💰!) To read more about my evaluation tool and the feedback I've received from experts on this subject, I've made a full blog post here: https://lnkd.in/ghD7p5je This project is open source, so feel free to check it out and give it a ⭐ to keep track as it grows: https://lnkd.in/gGV_b8pV To play around with what I have so far, here's the showcase: https://lnkd.in/gtTitarQ And if you have thoughts, feedback, or scathing criticisms - let's chat! #AI #ComputationalLaw #AccessToJustice #JsonLawgic #RulesAsCode

17

1 Comment

Anthropic released another paper on their ongoing work towards interpretability of large foundational models like their Claude 3 family, built on previous work by many. Their work continues to be enlightening and inspiring. The findings in this work are generally similar to previous findings on smaller models but scaled up to Claude 3 Sonnet: 💡 they find large numbers of "features" corresponding to combinations of neuron activations and these features map to concepts that we recognize/understand: for example, features for transit infrastructure, particular people, concepts (like bias and "objects lacking sentience"). This might sound familiar from classical machine learning, but the interesting thing is that there is no single thing/neuron encoding each of these concepts, it's a particular combination of neuron activations that embeds these concepts in "superposition" in the network. 💡they observe a sort of "distance" of these feature activations that arrange themselves "near" semantically similar features. When they use their sparse autoencoder approach to learn more and more features they observe that the level of abstraction in the detected features becomes more granular, in ways that are interpretable and meaningful. 💡when they artificially fire these activation patterns during inference, it affects the output in ways that relate to the "feature". For example, asking Claude "what's the most interesting science" normally it responds Physics, but if you clamp the feature activations for the feature "Brain Science" to 10x their max values then Claude responds: Neuroscience. Ways to interpret _and_ influence the model. Great readable work [1] along with an interactive explorer of some of the "features" themselves [2] 🎂 There's an easter egg too: if you scroll in the feature explorer tool and try and select one of the "Randomly selected features from 1M" you get a message "if you want to see the rest of the features, we're hiring!" with a link to their recruiting page :) [1] https://lnkd.in/gZbparwp [2] https://lnkd.in/gC9eus53

19

I found a highly useful book on how to handle "crucial" conversations, for example: how to provide feedback, how to disagree, share alternative point of vew properly: https://lnkd.in/gwjwfB4r It might be useful for someone else too.

13

A few thoughts after visiting the Hadrian library - Hadrian library was estimated to have housed 17000 books. This would make it one of largest in the ancient world. In comparison, the library of Alexandria was estimate to have 200,000 to 700000; Google Books 40 million. - Everyone of us have access to that much information in our pockets. 17000 books would fit on a 32 Gb Kindle Paperwhite. The internet is estimated to have about 150 ZB of information, more than 5 million times. - These cultural treasures/institutions, ancient and modern, are worth preserving. https://lnkd.in/g6vFq-tW - Consider donating (time or money) to organizations such as https://www.wikipedia.org/ and https://archive.org/

8

Hey folks, I'll be running a two-part virtual PD through Nassau BOCES in March 2025 called "Grading for Equity" using Joe Feldman's text, Grading for Equity, as our guide. Please see below for the description and feel free to share with your networks. To register, go to: https://lnkd.in/ehw8gkwn While grading is an everyday practice for classroom teachers, we rarely have the opportunity to interrogate our grading practices and understand how grades impact our students' attitudes, well-being, and academic success. In this transformational course, we will use Grading For Equity by Joe Feldman as our anchor text to 1) understand how traditional grading practices impact our students, and 2) how we can use research-based grading policies that are accurate, bias-resistant, motivational, and empowering. Rather than lectures, this course is active and engaging, using the case study method and rigorous discussion to support learning.

10

If you needed one more reason to use TypeScript in 2025, it looks like every single AI tool works better with it than regular JavaScript. For everything beyond weekend projects TypeScript is the clear choice.

3

1 Comment

Coincidentally, my first programming language was also BASIC. There was a time when "electronic dictionaries" were popular among students in Taiwan, and one of them came equipped with programming capability using BASIC. Having no easy access to computers or laptops as a student, I spent months learning the basics of programming using this tiny device. The physical keyboard was roughly the same size as on a Blackberry, and the screen displayed about 5x20 characters. When things got too complicated, I wrote the code in a notebook and then typed it onto the device. The code was horribly written, with lots of GOTOs, no concept of architecture, and poorly named variables. Still, it gave me a glimpse into computer programming and motivated me to become a software engineer one day.

2

Project update! My application tracking extension is now doing fuzzier matching and injecting new information into LinkedIn (and other sites) to keep me from doubling up on applications. Next up: local storage and bulk import/export operations...

12

2 Comments

Its about solving the problem, not 'having AI' in the system. This is from last March, but the Wave 1 vs. Wave 2 distinction does seem to be playing out. Less 'generate me content', more 'synthesize vast amounts of relevant data and aid in decisonmaking or signal extraction'. https://lnkd.in/ehFBFC32 They don't talk about multi-agent systems here, but looking forward to more startups with a focus on leveraging agentic design patterns (https://lnkd.in/ed5rMmFe) to streamline real world workflows.

4

1 Comment

Conversational AI is quickly becoming the talk of the town, and you definitely don’t want to miss out on the buzz! If you’re in the Peninsula/South Bay area, mark your calendar for an exciting tech talk hosted by my good friend, Simon Sturmer. He’ll be delving into the cutting-edge fusion of Automatic Speech Recognition (#ASR), Large Language Models (#LLM), and Text-to-Speech (#TTS) technologies. It’s a unique opportunity to gain insights into the innovations shaping the future of AI. Make sure you’re there to catch all the action! #ConversationalAI #Meetup #TechTalk https://lu.ma/uitxyhds

112

5 Comments

I'm excited to share a few things that we've been working on lately at Anthropic. 1. Computer use API We've created a new way for Claude to understand and interact with computer interfaces just like humans do – moving the mouse, clicking, typing, and navigating software naturally. Rather than building specialized tools for specific tasks, we're teaching Claude foundational computer skills that can adapt to any software interface. Computer use isn't perfect yet. Claude sometimes struggles with basic actions like scrolling, dragging, and zooming. But we expect rapid improvements in the coming months and want to give developers and users an early preview of this capability. 2. New Claude 3.5 Sonnet We've refreshed Claude 3.5 Sonnet and improved its intelligence and reasoning capabilities. The new 3.5 Sonnet delivers significant gains in coding—an area where it already led the field. Sonnet scores higher on SWE-bench Verified than all available models—including reasoning models like OpenAI o1-preview and specialized models for agentic coding. 3. Claude 3.5 Haiku 3.5 Haiku is the next generation of our fastest model. Haiku now outperforms many state-of-the-art models on coding tasks—including the original Claude 3.5 Sonnet and GPT-4o—at the same cost as before. Claude 3.5 Haiku will be made available in the coming weeks. I can't wait to see what you build with computer use and our new models. Read more on our blog: https://lnkd.in/deJJZFUj

146

5 Comments

This past week, I had a conversation that got me thinking about the future of interviews in an AI-driven world. We discussed deep fakes—images, voices, and now AI-generated personas. What happens when you can create a deep fake of yourself (or someone else) and pair it with AI that responds in real-time? On one hand, it’s unsettling. On the other, it highlights AI’s incredible potential to reshape processes we’ve long taken for granted—like hiring. While this raises important questions about authenticity and trust, it also presents an opportunity for companies to rethink how they assess talent and build stronger, more secure hiring practices. AI is evolving fast, and while challenges lie ahead, there’s plenty to be optimistic about as we explore how these advancements can create more innovative and equitable experiences. #AI #DeepFakes #FutureOfWork #Innovation

8

2 Comments

It is indeed tempting to focus on "high-level understanding" rather than to provide fluency through practice & repetition while teachimg, but the latest studies on neuroplasticity shows otherwise. This is why teaaching STEM disciplines - at least when successful - require a hands-on component. I grew up with no interest in playing golf, so the first time I picked up a club my body moviments and my "understanding" of what I should do, were reflected in the many many many different (wrong) directions in which my golf ball would fly. Watching videos only helped in identifying how much I didn't know. It was only when time was invested in actually attempting to perfect my swing, that I "got it". Therefore the following snippet resonated so clearly with me... "There is an interesting connection between learning math and science, and learning a sport. When you learn how to swing a golf club, you perfect that swing from lots of repetition over a period of years. Your body knows what to do from a single thought—one chunk—instead of having to recall all the complex steps involved in hitting a ball." I still suck at golf, by the way.

4

1 Comment

Having lived in SF bay area for so long, I came to know the big coffee culture scene here. I was totally consumed by it a while ago and spent years trying to attain the “perfect" latte. I published a fun coffee story and my entire home setup, which took weeks of research and multiple iterations, in my blog link in bio. Meanwhile, to all fellow coffee loving tech peeps, here are some top tips that no one told me. Put it in practice at home or in microkitchens. - A big factor in preparing milk suitable for latte art is pressure of the wand. If this is important to you, you need to get a machine that has a strong enough pressure. - When pulling shots, put in 18g of beans and extract 30g of espresso. - The darker the roast, the coarser the setting should be on the grinder, the lighter the roast, the finer the setting should be. - When adjusting the grind setting, do so while the grinder is working if there are beans in the hopper. Otherwise, a bean can get stuck between the burrrs. - Pulling a shot should take 20-30 sec in time, or based on observation. It should look like caramel going out, not a waterfall. - Eventually work your way to using a bottomless portafilter. It punishes you for all the small imperfections when preparing your coffee puck, like any tilt, and uneven distribution. It does so by spraying the espresso all over the place. The more you can get it to flow uniformly, the better your preparation of espresso, objectively. - Seal your beans in a vacuum, so they don’t go stale. The more stale the beans, the more minor adjustment you need to make to grind finer. - When steaming milk in your pitcher, you can tell if it’s done by holding the pitcher and as soon as the heat becomes unbearable i.e. stings, it’s done. - Taking a latte art course at a local coffee shop was the best boost for me to start nailing latte art.

9

🤯 CrewAI has totally kept me up all night! I've been working on this awesome content generator agent that taps into several AI intefrance providers for writing, critiquing, editing, and rating. ✍️ Writing: Groq/llama-3.1-70b-versatile (70B parameters) 🔍 Critique: SambaNova Systems/Meta-Llama-3.1-8B-Instruct ✨ Editing: Google DeepMind/gemini-2.0-flash-exp ⭐ Rating: Cerebras Systems/llama3.1-70b The demo below show the crew agents in action, the topic: "Self reflection in LLMs" Checkout the repo 👉 https://lnkd.in/gsZ53hK7 #AI #agents

7

2 Comments

Haha, I got ya, you pesky pixels! One corner of my rectangle was sticking out, so I slapped a border-radius on and banished it to the ether! #cssbattle #cssgang #css

13

5 Comments