Regan Connell

https://lnkd.in/eQmzEEFj. A great article from the Chief Information Officer at Goldman Sachs, Marco Argenti. LLMs are becoming more capable every month it seems, and one industry that people feel it will disrupt the most is software engineering. Given that LLMs can write syntactically perfect code, many feel that it will make learning to program obsolete soon. However, I think it'll be more like the calculator to math, or the printing press to writing: both were disruptive, yet innovative, tools for their respective fields, but ultimately they both helped increase the scale and complexity of calculations that could be performed and books that could be written, respectively. "Today, we are at a point where computers (i.e., AI) do not need this intermediate level of translation between the language we speak and the one they understand. We can set aside the Rosetta Stone and just speak English to a computer. They will likely understand just as well as when speaking to them in Python. This immediately presents two choices: We can get lazy, or we can elevate our thought." Thanks for the rec. Philip Walsh, PhD!

7

2 Comments

Gone are the days when you would Google important things/life advice like "How to transition from SWE to ML Engineer" without running into overly optimized trite SEO Optimized articles. I find good and important information on Reddit/Twitter, but then, I feel the information is scattered and unorganized again. Introducing TextData, a new tool made by Kevin Ros on which I am contributing these days. It allows you to save your notes, bookmark those hundreds of tabs scattered across your desktop, and, most importantly, build public communities and creators (or organizers? haha) can come up with curated content compilations. The vision is to create a community marketplace where creators could come up and develop roadmaps, and organize gated knowledge/personal notes that eventually would benefit members. Do give it a spin! #BuildInPublic! What are your thoughts on the platform?

28

1 Comment

Last week in San Francisco, I attended a talk by Sal Khan — founder of Khan Academy — about opportunities for AI in education. Some of my takeaways: - The best minds in history all learned through personal tutors (Plato famously tutored Aristotle). While the modern classroom is an attempt at scaling traditional tutoring, it does so imperfectly: children are clustered into groups of 20-30 and expected to learn at an average pace that doesn’t really work for anyone. AI tutors are a huge opportunity to bring back 1-1 tutoring, at scale. - Children have fewer inhibitions when speaking to an AI; they are more likely to keep asking questions until they understand, to ask the “weird” questions they would be too shy to ask in a group setting, and to try more often — because the cost of failure is 0. - AI won’t replace teachers, in the same way that YouTube didn’t replace teachers in the 2010s. Instead, it is a tool to augment teachers — and to help consolidate what has been taught in class in a more personalised way. - LLMs present opportunities for immersive learning by allowing students to “chat” with authors, book characters, or historical figures — offering better context and increased student engagement. - Evaluation and safeguarding are essential considerations in building educational assistants — in large part because they’re children-facing. Traditional machine learning evaluation techniques don’t apply here because the outputs are mostly qualitative; instead, these systems should be examined under software engineering standards — with extensive testing from teachers, parents, and (supervised) children alike. - “Technology is an amplification of human intent.” #generativeai #education #edtech

38

9 Comments

Sharing some of my key takeaways from Climatebase fellowship: - A great resource for climate knowledge and potential solutions: https://drawdown.org/ - https://bank.green/: Learn how your bank is investing your money in the fossil fuel industry. Choose a bank that invests elsewhere. - Fast fashion is terrible. Use resellers or renting exchanges. Look out for more sustainable brands. - We can reduce 10% of global emissions by changing to a plant based diet. - See where your electricity provider is sourcing energy from. Here in NYC, I learned about Zone1Energy that helps residents change to greener sources with reduced electricity bills. - 30% of the world's food is wasted. Reduced food waste can reduce the land and resources used to produce food. https://lnkd.in/g4N4jVa4 is a great website that helps cafes/supermarkets sell produce at the end of the day for one-third the price. - Every company should do Lifecycle Assessment (LCA) to understand their Scope 1,2,3 emissions as well as impact of how materials are sourced and products disposed. - Climate justice is needed. While countries in the Global North have produced the majority of emissions, countries in the Global South will suffer the most. There were many more learnings from this experience but these are some that stand out. As I hear about temperatures in India reaching 50 C, I hope we can collectively do a better job to save our planet!

25

I'm thrilled to share that I've successfully completed the Mentor Idea Review milestone at the Founder Institute. This challenging phase involved crafting a concise 3-minute pitch deck and presenting it to a panel of mentors and industry experts. Couple of takeaways: Building a Sustainable Venture: The rigorous sprints and structured feedback have been instrumental in shaping my approach to building a sustainable business. The process is intense but incredibly rewarding, teaching me the resilience needed to navigate the entrepreneurial journey. The Joy of being a founder: Despite the ups and downs, being a founder has proven to be an exhilarating experience. It’s a journey of perseverance, learning, and growth that I find immensely fulfilling. Exceptional Support: The quality of mentors and resources at the Founder Institute is truly exceptional. Engaging with knowledgeable founders and tapping into a wealth of expert advice has helped my entrepreneurial journey profoundly. I am immensely grateful for the support and insights from all the mentors, fellow founders, and the Founder Institute team. If you're contemplating a startup or looking to accelerate your entrepreneurial journey, I highly recommend exploring what the Founder Institute has to offer. #FounderInstitute #Entrepreneurship #Startups #Innovation #BusinessGrowth

7

2 Comments

Thrilled to share that our paper "Re-envisioning Remote Meetings: Co-designing Inclusive and Empowering Videoconferencing with People Who Stutter" received an Honorable Mention Award for ACM Conference on Designing Interactive System (DIS 2024). Amazing co-authors Shaomei Wu and Gilly Leshed! 🏆 🎉 Big thanks to our product team at AImpower.org and all the participants who shared their expertise and experiences in co-designing empowering videoconferencing products. 🙏 Quick overview: we conducted a two-phase co-design study with people who stutter to explore and reflect on the design space for inclusive and empowering Videoconferencing technologies from their perspectives. Our findings present a broad design space for tools that support people who stutter before, during, and after videoconferencing, focusing on aspects such as supporting self-disclosure, educating non-stuttering audiences, and promoting personal reflection for long-term self-growth. Check out the preprint here: bit.ly/CodesignVC

85

4 Comments

Testing LLM applications is hard, but robust regression tests are critical for developing software. I wrote a short blog talking about how we are doing this at Eidolon. Give it a read if you are having issues here! And please let me know what y'all have been doing so we can improve here as well.

5

1 Comment

In my latest blog piece I discuss the recent paper by Leopold Aschenbrenner titled: "Situational Awareness: The Decade Ahead." I outline my thoughts on it and that while his piece is certainly provocative and while one can disagree with some of his points or his timing, there is much to consider and that it challenges people to consider the accelerating pace of AI development and the profound changes it is bringing. https://lnkd.in/gdTqXMjn

3

Anthropic is publically committing to pausing training and deployment on any future models that develop “Red line capabilities”, essentially, capabilities that they deem to be too dangerous. They are using a standard they call ASL-3 (AI Safety Level) to determine if any system has developed these capabilities, such as having even low-level autonomous capability or a substantial risk of catastrophic misuse.

4

As we explored in part 1 of our Agents series, LLM based AI Agents ability to unlock AI’s potential to do real work across many domains is evident. But for these Agents to realistically produce actual work, they need to be able to handle higher order complexities and do so in a robust and reliable way. In this next post, The Future of LLM-Based Agents: Making the Boxes Bigger, we dive deeper into how we are approaching these problems at Arcus. Take a look!

5

As part of my first academic blog post I discuss the fact that we can enact alternative social, economic, and environmental relationships -- or, in a phrase, community-based economies. Here the use of the word economies is deliberate. Alternative economies can be designed to foster better social, economic, and environmental relationships than what we currently experience under the mainstream economy. Curious? To do this I employ a research paradigm called action research, using participatory design of artificial intelligence (AI) and automation tools to allow folks to spend their efforts elsewhere, more productively, and in community...

24

Gone are the days when we keep using OneNote/Google Drive/MS Word to store and keep track of all our learnings while browsing through vast amounts of web content! TextData is the new game-changer! Super fortunate to work with Kevin Ros and his amazing team doing my bit of building TextData as a graduate research assistant at the University of Illinois Urbana-Champaign. We are open to feedback! Do you have any community suggestions for which you would like to see organized data? How can we make this platform even better? Hit me up in DMs or the comments down below! #InformationMangament #OpenSource

27

1 Comment

Clear, concise post from Melanie Mitchell casting doubt on claims in an article from the prestigious Nature scientific journal summarizing Stanford research (https://lnkd.in/eeP2MV_8). https://lnkd.in/eU-dKk8J The Nature headline is misleading at best, because the underlying benchmarks aren't questioned at all. Most #LLM scores are gamed. Academics (funded by industry, with skin in #DeepLearning, and under pressure to publish) are using reductionist methods favorable to #AI that aren't objective comparisons to human intelligence. When the media, and academic journals, echo these claims without question they are acting like, well... #StochasticParrots ; )

25

2 Comments

"ChatBotArena is a living source of vibes evals." This sums it up nicely IMO. Highly recommend reviewing this blog by Nathan Lambert to understand what is [and importantly, _isn't_] accounted for by the Open LLM Progress Tracker: https://lnkd.in/e_utpZtZ

3

Artificial Intelligence needs data. LLMs+DKGs to the rescue? Attended annual week-long Knowledge Graph Conference (KGC) at Cornell Tech on Roosevelt Island in NYC. For the third time in a row. It looks like Artificial Intelligence large language models (LLMs) are converging with knowledge graphs (KGs). Branimir Rakic at https://tracelabs.io introduced their "ChatDKG" with DKG = DISTRIBUTED knowledge graph. Juan Sequeda at https://www.data.world presented an experiment combining LLMs and KGs -- chat accuracy improvement 3X. Hallucinating AIs need data. LLMs and DKGs to the rescue?

75

4 Comments

[#ResponsibleAI] A nice summary of my recent responsible AI talk at Northeastern University Toronto highlighting our work at the Institute for Experiential AI at Northeastern University

38

1 Comment

🚀 Join our Python and Data Science Sessions! 🚀 🔹 Python: Wednesdays at 4 PM (90 mins). Led by a former MSJ student, now studying Computer Science. Try it out for the first half-hour! 🔹 Data Science: Saturdays at 11 AM. Taught by a local Computer Science grad from Santa Barbara. Interested? Let me know for a temporary link. Please share with your network! #Python #DataScience #Learning #Education

2

CRISPR and AI are redefining gene editing. Profluent and Stanford University are leading the charge with their latest breakthrough - AI-designed CRISPR models. No more time-consuming search for organisms. No more slow, resource-draining methods. Welcome to the era of AI-driven CRISPR research. AI models, steeped in protein and genomic sequences, can decode genetic patterns with ease. They create CRISPR designs that surpass traditional tools, offering precise edits and significantly reduced side effects. Meet OpenCRISPR-1. An AI-designed, error-minimizing, gene-editing powerhouse. Better yet, it's open-source. Now everyone can explore this technology, free of charge. The future of precision medicine is here. https://lnkd.in/dxt7vJBg #genomesequencing #biomarkers #aiindrugdiscovery

6