Michael Miener
@boredbede.bsky.social
Cognitive Psychologist (Bremen) with an MBA (Henley). Left a career in management. Graduate student of Medieval Englishes at UoN. About to finish my MA thesis on early medieval battle poetry. Father of four. Cats. Chickens.
The ARC-AGI-1 is more interesting (though I need to read up on it). Here the efficiency ($) is shrinking fast, the capability scores always move in increments. Could this have been "cracked" by G3? ARC-AGI-2, remains a much harder challenge it seems.
December 1, 2025 at 10:47 AM
The ARC-AGI-1 is more interesting (though I need to read up on it). Here the efficiency ($) is shrinking fast, the capability scores always move in increments. Could this have been "cracked" by G3? ARC-AGI-2, remains a much harder challenge it seems.
Thank you. The models in the METR report use predetermined and standardized programming environments. The chart shows no exponential growth in performance. The report itself mentions a 15-20% capability gain over GPT-5, but no capability jumps. It's incremental, predictable progress - no?
December 1, 2025 at 10:42 AM
Thank you. The models in the METR report use predetermined and standardized programming environments. The chart shows no exponential growth in performance. The report itself mentions a 15-20% capability gain over GPT-5, but no capability jumps. It's incremental, predictable progress - no?
How is the improvement exponential?
November 30, 2025 at 5:06 PM
How is the improvement exponential?
It's a business school case study for change (not only range) anxiety...
November 28, 2025 at 2:01 PM
It's a business school case study for change (not only range) anxiety...
You should have at least come up with your own title...
November 20, 2025 at 7:14 PM
You should have at least come up with your own title...
It is about to morph into Biene Maja
November 18, 2025 at 11:05 AM
It is about to morph into Biene Maja