Plus friends and family inadvertently give Google/Facebook (etc) your personal info via tagged photos, contacts, DMs without your consent and they gather/link/sell this collateral info to the highest bidder without anyone being the wiser.
It constantly ignores explicit instructions (e.g. do NOT remove existing comments) and it's not nearly as intuitive in knowing the right questions to ask as gpt, claude or gemini in my experience from using all of them.
The reason I'm getting LLM burnout is from dealing with the obvious neutering and opaque downgrading of all the top models.
Prior to the last 12mos AI companies were hell bent on squeezing out the best results from mediocre models.
But... now that the top models have progressed, those same AI companies have switched their efforts into reducing the computation (cost of a producing a result) as much as possible without being too obvious.
What was an exponential slope in the quality of results over the last 36 months has now nearly flat lined.
Addendum: IMHO results have 'flat lined' not because the models aren't much more capable than a year ago, but because conserving the enormous processing cost (of an over subscribed user base) supersedes the goal of following the user's explicit instructions (e.g. especially if that means more processing cost) to generate the best results.
I re-subscribed to GPT's "PLUS" plan after ditching Anthropic for lack luster results... one of the first coding tasks I gave it resulted in a progress/thinking message that said something to the effect of (it vanished too quickly to get a screen shot unfortunately):
Evaluating client value
It took me aback. Note: the code had nothing to do with "client value".
Behind the scenes it is not hard to imagine OpenAI, Anthropic, et al simply minimizing processing for clients - like me - that are hopping from one to another to chase the just released SOTA model.
Looking at the comments for that product I suspect it'll find a home in your junk drawer rather quickly.
I've ordered gizmos from Lilygo in the past and found their support/documentation/updates to be very poor. Once a product is out the door they seem to do little further with it.
Churning out new products seems to be their forte.
I've been using ZFS for about a decade on several systems and can't say enough good things about it: rock solid, feature rich and easy to use are the top benefits.
I wonder if anyone remembers the long running computer club that met at the Forest Hill Collegiate on Eglinton Ave in Toronto?
And there was another one in the basement of the old Toronto School Board building (where the new TDSB building now stands) beside Mel Lastman Sq. in Willowdale.
This may be a more accurate analogy... "The Porsche you rented at $200/mo now only allows you a maximum of 100km of travel. You will be automatically charged extra when you go over that."
It seems that anthropic has added something similar to their browser UI because just in the last few days chat has become almost unusable in firefox. %@$#%
I hit my limit on the project I've been working on (after I let "MAX" run out and moved to "PRO") after about only 2 hours!
TIP (YMMV): I've found that moving the current code base into a new 'project' after a dozen or so turns helps as I suspect the regurgitation of the old conversations chews up tokens.
Your tool's method of returning element references is clever and should greatly improve llm handling of the page components (and greatly reduce token cost).
It inspires me to tackle a project I've been holding off on for many years: OCR my grandmother/great-grandmother's cookbook. It's about 100 pages of collected and annotated recipes from the 1930-1980s.
OCR and AI have become sufficiently capable (as you've demonstrated) to properly scan, index, and classify the recipes into something I can share with relatives online or as an ebook.
This is what I've been doing for a couple years now: having AI help to code/test projects that I've had in my long TODO list but would never realistically started/completed. AI is now pretty capable of producing decent code if your specifications are decent.
I still think that non-programmers are going to have a tough time with vibe coding. Nuances and nomenclature in the language you are targeting and programming design principles in general help in actually getting AI to build something useful.
A simple example is knowing to tell AI that a window should be 'modal' or that null values should default to xyz.
