News for hackers. Stuff that matters.
Re:Gemini 4 Neon is the new #1 on the Models Engineering Index by heavensteeth on Thursday November 19, 2026 @11:05PM (#50570018)
From Qwen:

> I'll write the script now. Let me write it. I'll write. OK let me just do this now. I'll write it. Let me write it. I'll stop. Writing. OK.

> my decode of #1 said off=0x1281. Hmm - hmm - hmm. Hmm. Hmm, hmm - but the displayed size was 22255... hmm

(I think this is my quant's fault tbf)

Re:Gemini 4 Neon is the new #1 on the Models Engineering Index by 0xbadcafebee on Thursday November 19, 2026 @11:13PM (#50570067)
[delayed]
Re:Gemini 4 Neon is the new #1 on the Models Engineering Index by novacrazy on Thursday November 19, 2026 @11:21PM (#50570124)
You gave the model ADHD. My brain might just be over-quantized.
Re:Gemini 4 Neon is the new #1 on the Models Engineering Index by dannyw on Thursday November 19, 2026 @11:35PM (#50570227)
The “stolen CoT” paper has some thinking traces of closed frontier models.

The ‘summarisers’ mask a lot of raw thinking you’d probably be frustrated by too.

Re:Gemini 4 Neon is the new #1 on the Models Engineering Index by Youden on Friday November 20, 2026 @02:47AM (#50571440)
Which Qwen? For Flash-Next there are several tunes that reduce the thinking tokens, like ThinkingCap or Swift. I've been running a Frankenstein version of the latter on my DGX Spark.