Xavier Hansen’s Post

In 1951, the brilliant #ClaudeShannon wrote a paper on “predicting the next word based on previous text”. The idea behind #GenAI has roots. Like my bandmate Jalil S. wrote back in the day: “Out with the old, and in with the new, and back to the old ‘cause you missed a thing or two.” https://lnkd.in/emVJ56wf

Interesting contradiction: one of the main use cases for #GenerativeAI in enterprise is #Retrieval & #Analytics (where the "generation" part is merely the translation of the valuable discovery process). Many teams sit on vaults full of historical documents and/or work with processes that produce enormous quantities of real-time data. The "unstructured data" (first and foremost, lots of words) got stored and mostly forgotten because the level of effort required of individuals to review thousands of pages of anything wasn't worth it or even realistic. #GenAI represents a radical change of capabilities. And -because LLMs are probabilistic machines- a thorny challenge to get to consistent and reliable results at scale...

One more post on this topic. From Prof. Ethan Mollick below. "1) The majority of students are using AI to help with schoolwork, whether you consider this cheating or not. 2) AI detectors do not work & have biased false positives that falsely accuse students. AI cannot detect AI writing. Asking the AI to detect AI use gives false positives. 3) Teachers and graders can't tell when AI is being used." #GenerativeAI #Education #LLMs #ChatGPT

Follow up post: Generative AI will transform education… because it already has. See the paper below: “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students.“ Hat tip Prof. Ethan Mollick https://lnkd.in/eihzNeWx

One important conversation that #GenerativeAI (re)sparked: the ability to write long-form essays is likely not the marker of intellectual ability we previously thought. This has big, BIG implications for school and work. As many among the neuro-diverse already know, you can be wicked smart and struggle with writing an essay. Claude Sonnet 3.5 writes better essays than most high-schoolers today. So now we'll have to focus on more creative ways to engage and assess "intelligence" than passing the written exam.

Surveys show that most people in the US have only used #ChatGPT (occasionally). Try #Claude Sonnet 3.5 if you haven't yet. #Anthropic is investing in a different user experience: fast, fluid and conversational. As Prof. Mollick explains, the future paths for LLMs likely differ for the coding community and for everyone else that is looking for intuitive AI.

My dude. :)

A brief summary of the "race to AGI" (aka "super AI"): #SamAltman: OpenAI is raising $ trillions to build bigger and bigger LLMs with our secret sauce. #IlyaSustkever: My new cracked SSI company will go to superintelligence in one straight shot. #DarioAmodei: Only Anthropic can build "safe" AGI. #YannLecun: Everyone is high on GenAI. LLMs are still dumber than house cats. META is developing new AI approaches to reasoning and world-modeling. It's gonna take a while. #DemisHassabis: Our AI psychologist said we already created a sentient AI at Google DeepMind. How did we let a startup steal our lunch? #OpenSourceAI: History is on our side.

Our paper is now available in the MIC 15 Conference Proceedings (published by Springer). Definitely didn't have "read 200+ papers on metaheuristics and co-author a Variable Neighborhood Search article with my son Aidan Riordan and my father Pierre Hansen" on my bingo card for 2023-2024. Always remain open to unexpected adventures... :) https://lnkd.in/ed-nbj-c

Adding to my reading list for today: "Transcendence: Generative Models Can Outperform The Experts That Train Them". One area I am exploring: "expert" training is likely to fall short on non-linear problems and tail risks... because this is a blind spot among many experts in the real world. This could lead to overconfidence in "generally" performant #AI systems that are fragile to sudden extreme changes in conditions. (Read: #NassimTaleb) https://lnkd.in/erWt-aYm