Post B45NWTUc6wJf4rUkAy by abucci@buc.ci
 (DIR) More posts by abucci@buc.ci
 (DIR) Post #B45NWTEJ5ZGwGHril6 by ngaylinn@tech.lgbt
       0 likes, 0 repeats
       
       A lab mate shared this write up of Don Knuth using LLMs to solve a math problem: https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdfIt's clear that using Claude did help them arrive at some new understanding here, which is wonderful. I'm happy for them.However, I'm upset by how much they personify Claude and attribute the solution to "him."From this narrative, it's clear that the humans were very actively involved from beginning to end. Claude was a helpful tool, but it did not solve this problem on its own. What role did it actually play? How was it like or unlike a human collaborator on this problem?It did generate a crucial insight, but where did that come from? Was it plagiarized from some unknown source? Did it "just emerge" from text completion and interpolation in latent space? Do we need some other explanation for Claude's apparent creativity?These folks don't care. They just wanted a solution, which they attribute to Claude, and leave it at that. I think that's a serious problem.#llm #ai #science #math
       
 (DIR) Post #B45NWTUc6wJf4rUkAy by abucci@buc.ci
       0 likes, 0 repeats
       
       @ngaylinn@tech.lgbtA lab mate shared this write up of Don Knuth using LLMs to solve a math problem ...There are at least three strong reasons to look at this document with significant skepticism.First, it's possible this document is Anthropic marketing. All the major AI makers have been paying influencers and high-profile people to make enthusiastic claims about this technology, using the sample playbook that was used for cryptocurrency and NFTs. Obviously I have 0 evidence of this, but I do believe, given this industry's past behavior and vested interests, that a level of skepticism is warranted. Apparently this version of Claude Knuth refers to was designed to be especially sycophantic. The fact that he mentioned the company and product by its exact name raises a yellow flag, for me (obviously there are other, innocent explanations for that, but it's consistent with influencer marketing).Second, and a little more "objectively", claims about machines doing or inventing stuff, or solving problems, are unscientific. They are not falsifiable. There is no counterfactual to compare to. An obvious question: if the same set of people put as much energy into solving this problem using more conventional means and without Claude, could they have solved the problem in the same amount of time or faster? People have been claiming that machines solve problems or invent things for a long time, from Doug Lenat's Automated Mathematician to John Koza's claims that GP generates patentable inventions.  It's a classic "marketing" move, like a flashy demo often is.Third, there is a rhetorical move here supporting a metaphysical claim that conflates a human activity with the activity of a machine. This, again, is not scientific; it also demands explanation and justification that goes beyond presenting evidence. If someone rides a bicycle down the road, nobody says that the bicycle walked down the road. If someone flies a simulated plane from Boston to Chicago in a flight simulator, nobody says the person traveled to Chicago. Yet somehow when people think with the aid of a certain kind of AI machine, we're meant to refer to that as the machine doing the thing humans do (thinking, solving a problem, inventing, or what have you). We're meant to believe that what the machine is doing is not meaningfully different from what humans do despite the obvious layers of metaphor involved. This conflation is not scientific, it's metaphysical. It demands an explanation and justification that goes beyond just presenting evidence because it is making a claim about how the world works or is structured.