Ukv

26 ms·

301.
by Ukv 2y ago
I don't personally believe that there's anything fundamentally preventing machines from being intelligent in the same way biological life is. Not to say that LLMs currently are. But, if you want, you can replace "some intelle
302.
by Ukv 2y ago
I think the core idea is reasonably solid. For as long as there's some intellectual capability that humans have and machines don't, it should in theory be possible to use that to distinguish the two. Turing gave the example of fee
303.
by Ukv 2y ago
Figure 16 onwards in the paper.
304.
by Ukv 2y ago
Human win rate would be 1 minus the model win rate, to my understanding. So 77% against ELIZA, 27% against GPT-4.5 with a human persona.
305.
by Ukv 2y ago
People choosing to use an LLM for therapeutic support probably do so with the belief that it is, while maybe not as beneficial as a human therapist, better than nothing. Early RCT results[0] so far do seem promising. It's not equivalen
306.
by Ukv 2y ago
The authors of the books didn't get anything because this article is an April fools joke ("Update: April fools" at the bottom). Hypothetically if it were a real scheme, it doesn't seem like it'd displace a sale give
307.
by Ukv 2y ago
A little hard to tell what's happening in the video due to lack of indicator of where the user is actually pressing/dragging.
308.
by Ukv 2y ago
> I am frequently asked for my PowerPoint slides, which basically function for me as lecture notes. It is unimaginable to me that I would have ever asked one of my professors for their own lecture notes. No, you can’t have my slides. I d
309.
by Ukv 2y ago
Agree with Sébastien's stance that understanding isn't just zero or one. I'd argue current LLMs do already to some extent satisfy Emily's definition of "understanding" by mapping from language to an internal se
310.
by Ukv 2y ago
> That's not telling you what you think it is. A lot of those deaths are that person in a car on their own. [...] Sure - going off of NHTSA figures it looks around 35%. There's also a lot of car passenger deaths (~15%), pedestr
311.
by Ukv 2y ago
> So far the AI use cases that have succeeded are those where that accuracy requirement is not stringent. Image generation and coding assistants, e.g. Cursor, Copilot, and similar, are the only ones that come readily to mind. Or where ac
312.
by Ukv 2y ago
> I don't want my pencil to draw 2% away from where my hand goes: I want it to be the exact result of what I do If you have a pencil that, due to the graphite core being loose, draws 0.5mm from your intended position and you're
313.
by Ukv 2y ago
> If it’s 100% accurate, it’s a fantastic time saver. If it is anything less than 100% accurate, it’s useless. I don't think 100.0% accuracy is a reasonable requirement for anything. Should at the very least be "do I end up wit
314.
by Ukv 2y ago
From what others have posted, this sample was apparently 35 nanograms salvaged from a smoke detector - and was sold by the online store embedded within an acrylic cube labelled "plutonium" in addition to still being in the origina
315.
by Ukv 2y ago
Worldwide stats from https://www.who.int/news-room/fact-sheets/detail/road-traffi... : > Approximately 1.19 million people die each year as a result of road traffic crashes. > Road traffic injuries are t
316.
by Ukv 2y ago
Works fine for me. As you scroll down/go through the pages you'll get looser matches, but the first topic results I get for "USMLE" all seem exact: 1. "ChatGPT Has Passed the United States Medical Licensing Examina
317.
by Ukv 2y ago
Feels like an issue that could be approached with a randomized trial, assigning some "high-tech" schools (every student gets a laptop) and some "low-tech" schools (mostly paper-based), to get a better idea of whether ed-
318.
by Ukv 2y ago
To add anecdotally based on logging on my portfolio site, all major US players (OpenAI, Google, Anthropic, Meta, CommonCrawl) appeared to respect robots.txt as they claim to do (can't say the same of Alibaba). Sometimes I do still get
319.
by Ukv 2y ago
> I think whenever people say "zeroth" they speak in jest and doubt that there is any disagreement on the fact that the element without predecessor (nowadays in programming most often assigned index 0) is the first element. &
320.
by Ukv 2y ago
I don't think it's "wrong" per se - it's zero-based indexing and ranges are consistently inclusive of start, exclusive of end. But, it breaks the "fence-post" mental model that some people use, and is less
321.
by Ukv 2y ago
> How many years old are you when you’re born? You have lived zero full years and are in the first year of your life. In most (but not all) countries the former is considered "your age". That's consistent with both zero-ba
322.
by Ukv 2y ago
> Yes, negative indexing as in e.g. Python (so basically "from the end") can be incredibly convenient and works seamlessly when indexes are 0-based. I'd claim 0-based indexing actually throws an annoying wrench in that. Co
323.
by Ukv 2y ago
I think they mean that it's a tool accessing URLs in response to a user request to present to the user live - with that user being a human. Like if you used some webpage translation service, or non-ML summarizer. There's some gray
324.
by Ukv 2y ago
For those exercising their GDPR right to rectification, my general feeling is that a filter applied onto the model to prevent the invalid output should be considered sufficient here. Judging a system by an incomplete internal representation
325.
by Ukv 2y ago
> AI companies can also not just 'hide' false information from users while they internally still process false information To my understanding of this: 1. The false claims were hallucinations, i.e. do not exist in training data
326.
by Ukv 2y ago
Are sites that currently try to enforce real name usage, like Facebook, noticeably better for it than mostly-pseudonymous sites like HN/Reddit? It doesn't really fix people being gullible (so will spread spam/scams or fall fo
327.
by Ukv 2y ago
> But I did specify "start of 2022". Your initial comment didn't: > > GOOG is around the same price as it was in 2022 That's the comment pb7 replied to by saying it's up 90% since 2022 (which is true, or ev
328.
by Ukv 2y ago
> Nothing of value is lost when these models are made private again To list a few uses I see for voice-generation/cloning tools: * Real-time translation of a user's voice, maintaining emotion and intonation * Professional-quali
329.
by Ukv 2y ago
I wouldn't say we'd be quite back to pre-photo evidence days. I feel a lot if not most of the value in a video/audio recording is not just that the medium has traditionally been difficult to edit, but that it's attesting
330.
by Ukv 2y ago
Feels like the underlying issue is Mozilla's dependence on Google. I can understand why Mozilla took and don't want to give up the search deal - it gives significant funding for only a minor concession, so it's easy to justif

More ›