Overview: Science&Technology

Classic example of a negative view on LinkedIn: "Unlike many others, I don’t think any reinforcement learning or reward algorithm is at play... appears to be a generic Chain-of-Thought (CoT) process that breaks tasks into several steps... Subsequent steps ... generated based on context... subsequent interactions concatenated into the context.…

“If you read 1,000 papers and build a powerful representation, humans can interrogate, mine, ask questions, and even get the system to generate new hypotheses...What you will learnAccelerating scientific discovery with generative knowledge extractionUnderstanding ontological knowledge graphs and their creationTransforming information into knowledg…

“About one in four people who began the experiment believing a conspiracy theory came out the other end without that belief... challenges conventional wisdom that evidence and arguments rarely help to change believers’ minds... [as they're] adopting such beliefs to meet various needs – such as a desire for control."The researchers devel…

"examines the impact of AI on human false memories--recollections of events that did not occur or deviate from actual occurrences... through suggestive questioning in Human-AI interactions, simulating crime witness interviews. Four conditions were tested: control, survey-based, pre-scripted chatbot, and generative chatbot" The 200 people…

A Large-Scale Human Study with 100 NLP Researchers

"AI-powered tool designed to assist users in building software ... [from] natural language prompts ... create applications from scratch.

AI is a wider field than LLMs, so "not all the AI use cases are suitable for Generative AI."

"The most common AI-based functionalities in participation tools are toxicity screening, analysis of inputs and translation. The first two in particular are meant to lighten the workload" of community managers."Toxicity screening ... is used to flag hateful or inappropriate inputs." Text is usually flagged and post-moderated, b…

Starts with a basic outline of model collapse and the challenge it faces new AI development, providing several links: "Research has shown that when generative A.I. is trained on a lot of its own output, it can get a lot worse."It then uses scrollytelling to provide a simple example of how "Just as a copy of a copy can drift away fro…

"A deep dive into why RAG doesn’t always work".It should be easy: "install a popular LLM orchestrator like LangChain or LlamaIndex, turn your data into vectors, index those in a vector database, and quickly set up a pipeline with a default prompt."However, while "quick-and-dirty demos are great for understanding the basics…

"As AI-generated content blurs the line between human and machine online, “model collapse” might help us find new value in well-managed human communities."

"See how visible your brand is in AI-powered search engines. Spot where you are doing well and where you can improve to increase brand awareness and sentiment."

AI "can replace much of Wall Street’s entry-level white-collar work, raising tough questions about the future of finance". On the one hand the work - "Pulling all-nighters to assemble PowerPoint ... Punching numbers into Excel ... esoteric financial documents that may never be read" - sounds a pretty dismal "rite of passag…

I hope you had a good summer. I stayed and worked from home, and got a lot done thanks to the mercifully fewer meetings. I also read a lot of good stuff, and published one piece. Here's a selection.

"does something like ChatGPT actually display anything like intelligence, reasoning, or thought?" or is it just a stochastic parrot? And "if you’re just making a useful tool – even ... a new general purpose technology – does the distinction matter?"Yes. LLMs have a ‘reversal curse’ which means it will "fail at drawing rela…

The EU "risks falling behind because of incoherent and complex regulation", but there's an opportunity: "open-source AI—models whose weights are released publicly with a permissive licence... ensures power isn’t concentrated among a few large players... creates a level playing field".Europe has more open-source developers…

Ethan Mollick outlines "three fundamental truths about today’s AI" - ubiquity, and both capability and risks which are as enormous as they are unclear - and proposes "a pragmatic way to consider when AI might be helpful, called Best Available Human (BAH) standard... asks the following question: would the best available AI in a parti…

"When we attribute human-like abilities to LLMs, we fall into an anthropomorphic bias ... But are we also showing an anthropocentric bias by failing to recognize" what they can do?

Note-taking "software can’t automate your thinking. But I do think it can create the conditions for improved thinking: making new connections between ideas"helping reduce distraction"organizing your reading and making it more useful"Casey then intros a few apps he's started using since his last post, starting with "Ca…

Note-taking apps are "designed for storage, not sparking insights. Can AI change that?".Casey's info overload is worse then ever: "As a journalist, I’ve never collected as much data as I do now ... browsing four or five social feeds ... arXiv and pre-prints ... [leaves] a stack of research that I will never get through. Book ga…

Don't "throw AI at problems that are either poorly defined or better suited to non-AI solutions [which] ... no one took the time to consider... more often than not, traditional programming approaches offering greater control and reliability ... managers feel pressured to “find” some AI, regardless of whether it’s the right solution. How …

https://chatgpt.com/g/g-vZ7SgKBOh-framework-findervia LinkedIn: "Helps locate and apply frameworks to your problem... to help you think through issues by providing frameworks... Here is the prompt for the GPT in full: 'You will help the user find frameworks to help them better understand, analyze, and solve a problem. Frameworks might in…

"What happens to GPT generations GPT-{n} as n increases?... the use of LLMs at scale to publish content on the Internet will pollute the collection of data to train their successors: data about human interactions with LLMs will be increasingly valuable."

Ludicity (aka Nikhil Suresh), a developer with real AI chops ("formal training as a data scientist, going so far as to dominate a competitive machine learning event at one of Australia's top universities and writing a Master's thesis where I wrote all my own libraries from scratch in MATLAB") has had enough: "it is with gr…

Some good reasons to not integrate generative AI into your software (like everyone's doing), from Membrain, which " builds tools and workflows that help you build structure to support your sales strategy and create positive behavior change on your sales teams".Their arguments:"It’s Not Free or Even Cheap" and the quality i…

"it’s surprisingly unclear what the AI search startup actually is"- according to its CEO it is, depending on when he's asked:"an answer engine...almost like Wikipedia and ChatGPT had a kid,,,a mere “aggregator of information”"All three cannot be true. The chatobot itself "is more specific... “Perplexity AI is an AI-po…

After so many years of spam, splog, slop, phishing and above all trolling, "Our guard is up. Our filters are activated. Our default mode is suspicion", something that's only going to get worse with AI: are those photos too good to be true? Can that video actually be real? Is that a spambot, or just "someone who isn’t neurotypic…

"There are now several Obsidian plugins available that allow you to use local LLMs instead of a commercial LLM provider", so PKM explorer tries Copilot for Obsidian plugin on his Windows laptop, because this plugin "not only offers the possibility to have a question and answer dialogue ... but also lets you [index your vault]... you…

Applying Dawkins' selfish gene perspective to the evolution of AI

First in a series on "how AI tools can transform intricate concepts into clear, practical frameworks and diagrams", starting with Napkin AI, which "generates interesting diagrams automatically from text input".Plenty of crap circles follow (" Upon completing each document, I exported them as PDFs. Find them here"), bu…