AI
Two Computer Scientists Debunk A.I. Hype with Arvind Narayanan and Sayash Kapoor (via Adam Conover)
This was an informative podcast on why there is so much hype around AI. There are too many details to list here, but some things that stood out are (1) much of the “problem” we have with AI is more of a people issue rather than a technology issue, (2) there are misaligned incentives, primarily driven around profit rather than real solutions, (3) in several scenarios it’s easier/cheaper to exploit human labor than to use AI tech, (4) there are very real impacts on people (e.g., multiple companies using the same AI-based HR screening companies that reject candidates), (5) there is concern where these tech CEOs say that AI could be used for ill and only they can prevent that from happening.
Conference
- This is a single-track conference in Knoxville, now in its second year. It was good to see some familiar faces and meet new people. The food was fine, parking wasn’t an issue (maybe because I arrived early), and the venue was quite cold (temperature). Note: Not many mentions of Microsoft tech. The quizzes about presentation knowledge were a nice touch, but had serious pacing problems. I’d like to see how this conference matures over the next few years.
- Keynote (Beverly Wright) Evolving from Business Intelligence to Artificial Intelligence: Transforming Mindsets and Organizations. Skippable. It felt like this talk was put together on the drive/flight to Knoxville. The speaker spent lots of time humble-bragging about her degrees and work she’s done over her career. There was no cohesive story to her talk other than maybe “people don’t understand statistics.” Despite having AI listed in her abstract four times, no application of AI appeared in her presentation.
- Amos Manneschmidt: How I built the Screw Sorting Machine. Physics major turned software engineer describes a passion project of building a screw sorter using 3d printed parts, commodity hardware, and Python + ChatGPT. He described a compelling experience of using ChatGPT iteratively as a pairing companion. I was impressed how far he got with open-source tools on hardware that wasn’t top-tier, and it reminded me that I used to know how fast Fourier transforms worked.
- Amy Fletcher: Never Say Die: Digital Afterlives in an Algorithmic Age. [Favorite talk] Death bots — products that train on images/text/videos of deceased people to make them “live forever” in a sense are a thing now. I also learned about “hauntology.” These kinds of talks — how humans use and think about tools — excite me more than the underlying technology that makes them possible. One can do incredibly useful things or make extremely potent snake oil. (Unfortunately Amy’s comment about a CGI James Dean appearing in movies within the next 2-3 years is out of date — the movie was cancelled: https://www.fudzilla.com/news/57450-james-dean-finally-stars-in-a-new-movie.)
- Gaurav Mittal: RealTime Defects Categorization using ML/AI. He demonstrated an application of natural language processing (NLP) to triage incident reports. The concepts shown, such as sentiment analysis and image recognition, have been around for decades. The speaker’s accent made this one hard to follow.
- Levi Smith: Blockchain for Game Development. Nothing new to see here — blockchains and NFTs — concepts that have been around for 15+ years.
- Wil Wade: The Post-AI Future of Thinking. Not much meat on the bones for this talk. It almost feels like it was written by ChatGPT.
- Jeremy Davis: Scaling AI Solutions: How GraphRAG is Changing the Game for Enterprise Applications. This was the most applied technical (and thorough) talk of the day. It was a bit of a pitch for GraphRAG, while clearly laying out the downsides to traditional retrieval augmented generation (RAG). One of the memorable moments was how the speaker applied modern AI techniques to common datasets with 95% accuracy, but that accuracy fell to 5.5% when applied to their own internal data. It turns out you can’t fire all the data scientists after all! (After further searching, GraphRAG was developed by Microsoft Research — a detail was not mentioned in the talk.)
- Cody Lambert: Building a Second Brain: Leveraging Obsidian for Personal Knowledge Management in the Digital Age. A product pitch for Obsidian as a PKM (personal knowledge management) tool. This tool seems very capable but feels like the answer to the question, “What if Linux made a productivity tool?” Its strength lies in its customizability and not using proprietary tech or data format. (I think Cody needed several good nights of sleep, as the delivery seemed as if someone else wrote his talk and he had to deliver it.)