> at Q4_K_M, stock-style quantization is retaining ~99–99.8% of BF16 accuracy

That's a tall claim. By that measure, even NVIDIA's QAD, which is AFAIK is currently SOTA for 4-bit quantization (albeit NVFP4 instead of INT4) would be worse than Q4_K_M RTN quantization. :D

https://arxiv.org/abs/2601.20088

The most salient thing about these models is that they're non-reasoning models. This makes then very token efficient and particularly well suited for local inference where decoding is usually slower than with datacenter GPUs.

Link to HF collection: https://huggingface.co/collections/ibm-granite/granite-41-la...

Probably worse than Gemma 4 or Qwen 3.6 with thinking off.

I can't believe I joined Github back in 2009. I was a hardcore Mercurial fan and user back then. :)

Someone on X seems to have found data leaks from the future in its training set.

https://xcancel.com/deredleritt3r/status/2048977698832241060

They discuss this in the article, in respect to its knowledge of FDR.

One of the first bits of infosec advice I give to my non-technical friends and family, when they ask for it, is to turn off background location access for all apps on their phones.

Needless to say, I know plenty of technical people who don't care about it.

because someone made background location access a "necessary" part of the the bluetooth stack?

The cost of opting out is very high.

"Mark of the beast"-- you want to participate in society, you need it.

Modern people seem to be incredibly weak and dependent. "I can't protect my freedom if it means giving up bluetooth!" It almost reads as satire.

I've seen people getting fired in BigTech for using the platform to stalk their ex-es. It's usually an alert that goes off when employees access internal dashboards for a certain profile, too many times.

BigTech is far more competent than a Telco though.

having worked and consulted at both... debatable.

level competency is higher at BigTech but laziness, vanity, selfishness, ego, and learned-helplessness happens plenty too.

e.g. for all of the BigTech brilliance plenty of them fall for mildly complex phishing efforts or bribes, etc.

Yeah, OpenAI has been attaching C2PA manifests to all their generated images from the very beginning. Also, based on a small evaluation that I ran, modern ML based AI generated image detectors like OmniAID[1] seem to do quite well at detecting GPT-Image-2 generated images. I use both in an on-device AI generated image detector that I built.

[1]: https://arxiv.org/abs/2511.08423

Exactly, I grew up playing with BC547 and BC337s (my father was an electronics engineer) and only later found 2N2222 and 2N3904. Those were almost entirely unheard of in India.

Merely implies a very good fitness function.

Yes. Though according this fitness function we're not necessarily more successful than a jellyfish or a tapeworm.

Arguably much less successful since jellyfish have been around 700+ million years ands it’s not clear if humans will make it even the next couple thousand. But the jury is still out on that one

aka A Metacircular Interpreter

Do you think God stays in heaven because he too lives in fear of what he's created?