I'm not even going to try to go into reasons, though I agree the decline appears to be longer than machine learning has been around. It could be how we are measuring rather than an actual decline as I don't always trust how people gather statistics, but I would hope they did a good job. I'm just looking at the numbers which show reading scores have declined since 2015, and maybe even 2013, not just the regular perception of "people were smarter/worked harder (whatever) back in my time.".

"In 2024, the average reading score for the nation at grade 12 was 3 points lower than in 2019. Compared to the first reading assessment in 1992, the average score was 10 points lower in 2024. NAEP scores are also reported at five selected percentiles to show score trends by lower- (10th and 25th percentiles), middle- (50th percentile), and higher- (75th and 90th percentiles) performing students. Compared to 2019, scores were lower at all selected percentiles except for the 90th percentile."

https://www.nationsreportcard....

Or maybe in the near future, drones spraying gasoline across a city and someone tossing a match. Never forget the past...

And about understanding that maybe 90% of what LLM coders provide is good, 10% is somewhat off to outright bullshit and LLMs can be very good at making things do what they are told to do without actually being right when a solution is hard to produce. Ultimately, it is about being a good manager who knows in detail what the solution should be, not just accepting what an LLM or junior coder produces. At least for being a good engineer, many companies just want faster code and are happy to provide 10% bugs to a customer.

Not thrilled, but doesn't sound much different from what has already been done in wars too often. Make everything here dead, rather than asking the AI to decide who to kill and who is not a "legitimate" war target.

Heck, I don't trust a minimum wage store employee to pick out produce for me with a delivery order, but I'd trust them more than an LLM that keeps telling me a gitlab feature exists until I tell it what happens when I actually do what it suggests.

I wonder what the AI would think of the weeks I spent tracing requirements in DOORS every day. Mind numbingly boring, but stuff that needed to be done since automation at that time was not sufficient. Or is the AI basing "too much work" on its own runtime exceeding 40 hours a week on the same repetitive task? I suspect pattern matching coming up with false equivalencies when trying to find the most relevant match for something. Not like I've ever seen an LLM do that before...

Yeah, and this one got me because I wanted to know how any judge actually qualifies human intelligence, which the article had nothing to do with. Smack myself on the hand for believing the article title :p

I live in a 1966 house that is fairly up to date and the estimated (take that with a grain of salt) value has doubled in the past ~10 years. We used to live in basically the same house as you, same age, ungrounded outlets, one more bathroom, that house has an estimated value that has gone up 3x in the past 10 years. Our current house is on the edge of the city and country, a few minutes further from everything, a much quieter area where we prefer to live. The previous house is in the middle of the suburbs. Just a one off case but I feel like location plays a bit more than quality of the house when it comes to value.

From the summary, it seems like the judge is ruling that AIs are not human and not official lawyers, therefore they do not qualify as either. I don't see anything stating the judges think the AI does not have human intelligence.

Pretty much the same here. Our neighbors can get AT&T and Comcast, but we are one house down the side road and only Spectrum runs lines down here (we are less than 50 feet from them and there are a couple dozen more homes before the dead end). We pay $90 for internet that struggles to try to meet the supposed 100 MB, and hasn't been stable since they upped our speed (and bills) from 25 MB.

Basically my thoughts. Is it the appeal of physical media, or the increasing enshittification as streaming services fracture, remove content, increase prices... I need to stop talking about all the crap they are doing.

I think (/s) AIs just gravitated to market speak during its training. AI has spent years training on the internet, market speak has been developed to misdirect with meaningless words for centuries.

Agreed, when I say chasing its tail I'm not talking about a simply output to the command line (which I really wasn't clear on). I created a program that generates a maze using Kruskal's algorithm with C++ and asked Claude to convert this to intel assembly with comments to cover where the equivalent functions and variables would be to make it easier for a class to match the assembly with the C++. It mostly worked, but Claude kept chasing one bug in circles until I realized what it was doing wrong and pointed out where the actual problem was. The LLM is quite good, but it has a tendency to get stuck and need help getting back out more than I like.

I've used LLMs for generating web interfaces and it does pretty good. Try to get it to code in assembly or have a container log traffic between other containers and the LLM will keep chasing it's tail trying to fix bugs.