1,029 karma · joined June 23, 2025
>>"This sentence has five words" is going to appear far more often than "This sentence has four words".
it's not about this at all. your point is about data quality. you need to take a step back. the point is that if you trained a language model just on this data set which has sentences akin to "this sentence has two words" - the model is going to learn that. this shows that the language modeling itself doesn't truly provide an understanding of the real world. you can train a language model with the most advanced technology on shitty data and the model will start providing shitty outputs confidently - the model will never say "hey there is something wrong with the data i am trained on". thats what 'understanding of the world' meant in that paper.
>> > LMs are not performing natural language understanding (NLU), and only have success in tasks that can be approached by manipulating linguistic form
did you read the paper carefully??? this line is directly cited from another paper (pay attention - it's from july 2020) . the full line is - LMs are not performing natural language understanding (NLU), and only have success in tasks that can be approached by manipulating linguistic form [ 14 ]
the paper its cited from is titled - Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data
>>>She did say that, as clear as day. Now she's trying to rewrite history. Ugly behavior.
did you read the article after that line or was there a shortage of attention span??
i think one thing that's going for google is the network effects and what it's able to do.
>>And my criticism is that if you observe Claude code with opus 4.8 working through an entirely novel problem that nobody has ever worked on before and which certainly wasn’t in its training data, the choice to even metaphorically call them stochastic parrots turned out to be egregiously wrong.
First of all, for your own benefit - Claude code stopped showing the real reasoning. It only shows a summarized version of it now. So don't ever ask someone to observe the model. Second, do you know when this paper came out? do you know when opus 4.8 came out??? how do you know what is novel? did opus 4.8 tell you it was novel? how do you know no one has worked on it before?
did you think this through?
imagine the sentence was "This sentence has four words", now extrapolate that to all the shit that can exist in a dataset and train a model on that dataset - do you know what will happen? - go ahead and think it through.
why? that's a bit of an overreaction from your alarm bells honestly. i didn't know about organic maps because i never looked up for this solution and never came across this specific app in my readings. what i am working on (and want to figure out) has less to do with commercial angle of privacy (which organic maps solve) and more to do with navigating without relying on any external signal. i truly think we should be able to navigate earth (land/sea/air/subs) without any external human controlled signal (GPS/Internet/radio etc); that in my mind is true offline navigation. but you can argue that organic maps is also an offline navigation app and i think for an everyday user that's true.
your comments are helpful. thanks for being so thorough with your thought process. i am trying to write on https://trulytyped.com/user/4f448758/deepan-wadhwa - but i have not written anything technical yet and there is no RSS button there yet. let me add it today haha. thank you.
are you sure this is true? - i can still run the phone on airplane mode(with GPS off) after i have the maps downloaded and can go anywhere afterwards. i don't think there is any communication with the outside world after that.
I am not sure if you looked at the repo I linked. In the intro I clearly state that I am doing this work for my own curiosity and for a hypothetical scenario (where GPS is denied). Moreover, the tech I am relying on (TERCOM) has applications in other fields as well such as navigating under water (submarines use a version of this). Or missile systems such as tomahawks which use tercom extensively.
i can sense the lack of capacity for reasoning in your comment. it's okay. the paper and the article are written by two very different people. the article is written by someone in the communications team, who communicate "what's going on at the university" to the outside world . the paper is written by the actual scientists.
do you have any other explanation other than tells? - maybe your tarot cards might have better explanation.
can you please share your methodology for detecting ai please?
But here's a bigger problem - Imagine a teacher grading 50 students - 40 of them use AI to write their answers and 10 write honestly. All 40 of them use the hack that I used and get 100% human from pangram and the other 10 also get 100 human from pangram. what is the product really adding to the workflow? Nothing, zero or zilch!!! - but then you come along and say hey at least those 10 honest students also got 100% from pangram. This whole argument of false positives being very very low works if your false negatives are tight which they are not.
the text in above is all ai generated. i uploaded a picture of two kittens to gemini and asked it to guess their age. then asked gemini to change the language to of someone who is born in deep Appalachia but has Japanese parents. then asked it to change the bullets to paragraphs written by someone who is talking fro life experience rather than grounded in published science. pangram gave it 100% human generated. go ahead, try it. it boggles me how people believe these ai detectors.
if it makes you feel good the original response from gemini gets 100% AI detection and for that I think pangram should get the credit but in practical terms the empty pots in my yard are more useful than pangram at this point.
'To my knowledge' - okay. In my experience, they're all horrible. I have fooled them both ways - ai generated content that they say is purely human and human generated content that they say is partially AI or just AI.
They can't produce anything on their own. They have to be prompted which is initiated by humans at this point, so the patents can be owned by the initiator(human) not the tool.
Well, we now have the best model of our time (trillions of $$$ of investments) telling us something completely different(and wrong) from a human expert. I would really like someone calling out dario, sam, elon on these things and hear their explanations but alas, a man can only dream.