flockonus
You're comparing apples to oranges there!
To hallucinate is arbitrary, you can use a 2B image param model, it will hallucinate an OS no problem* running real time!
What this post is about is completely different, which is actually emulating the complete system in an LLM - that is hundreds (or more) times slower.
*absulutely crappy quality tho, unusable OS in practice
About 2y ago, before AI was all the rage - lookup: controversy, scientific fraud networks, "paper mills"
Suppose we have a very convincing math PhD impersonator, agreeing and making some prepared smart-sounding comments saying they understood the proof and it seems to make sense - would it?
Well, most would say - NO! Obviously not, knowing this person was an impersonator. Others now knowing could be convinced of the validity.
Very simple thought exercise that allows to conclude, someone "saying they understood" (or agree) bears absolutely no weight to something's validity.
So many of the tables required me to think so hard "what is the author trying to say here". Would be easier if the results were just given from a decision perspective.
One example: "Its own server or twenty people?" and the axis is "share of versions" ... makes no sense until read the fine print and interpret it.
Great model to architect and resolve eng. problems in a harness loop tho. Poor handoff.
Still, unjustified in the pipeline and feature as a whole. Makes no sense to add another format to the pipeline, even if that format saves ~30% on size. Now i need to store 170% of the image size from before & have extra complexity to deal with.
*yes i know, not apples to apples.
I prompt it to:
"Attached (or underneath) is the transcript of a self recording i've done with tons of rambling and some incorrect words transcriptions, please do a pass clearing out and arranging any typos or possible misunderstandings. Keep original in parenthesis when not sure if it's a misunderstanding. Do not summarize or alter the nature of the content, simply tidy the transcript."
It absolutely is relevant.. but only in about 6 months :)
Suppose the obvious case, i have some resizing pipeline for images. Would i include support today in my resizing pipeline today and pay a few extra cycles and storage for someone that my clients can't read?
It would be cool to have a subscribed alert for when the tech crosses a certain % threshold adoption so i can look at it again, otherwise it just relies on my brain's flawed long term cron job
Anything below ~90ish% is no go, unless i have a fallback strategy of some sort.
Open weights models giving a distant salute from afar
Sadly, still far from adoptable today's web: https://caniuse.com/jpegxl = 17% https://caniuse.com/?search=webp = 97%
Evidence of high levels of power political pressure - US latest 25% tariff pressure came out explicitly around this topic - https://www.reuters.com/business/finance/brazil-us-clash-ove...
At first i thought it was going to be a pure counter-AI rant, but no, the author admits using AI everyday in a variety of ways.
Another argument, maybe OSS of smaller packages is on its way out as we get to generate whatever is needed on the fly?
500B params performing worse than other OSS of the same size is pretty meaningless if no one will use it.
I wonder what's the footprint and overall efficiency of making a new font vs. combining old ones on client side (typically JS)?
Jev is not interesting if it's not "smart", a 1B param model is most definitely not smart.
"What are the odds a thunder will strike in Paris at 1pm UTC of 2026-09-16" - that could be a 0.001 chance going from blind historical measurements; 0.01 if it's raining; or 1 or 1 after the date has passed.
But more so it seems there is Fear of missing out (FOMO) in our behaviours. The reality is, if whatever model you are using are good for your purpose, well, keep on it.