#TestTimeScaling
QuasiMoTTo nearly hits the theoretical ceiling for any sampler that preserves LM marginals. That's a strong result, but the tasks are all compact symbolic domains. Does the bound stay tight once you move to open-ended text generation?

#MachineLearning #TestTimeScaling #MonteCarloMethods
July 2, 2026 at 4:00 PM
for a v. large value of personal computer, you can turn your personal computer into an IMO medalist/40 on the Putnam exam pretty easy, just pick up kimi k2 weights and read up on this

it's _impressive,_ but not a big deal (weird as that is to say) - most don't think they need AI good at math :(
GitHub - testtimescaling/testtimescaling.github.io: "what, how, where, and how well? a survey on test-time scaling in large language models" repository
"what, how, where, and how well? a survey on test-time scaling in large language models" repository - testtimescaling/testtimescaling.github.io
github.com
September 28, 2025 at 11:15 PM
Researchers introduce Test‑Time Augmentation and Test‑Time Adaptation, boosting small vision‑language models with gains across nine benchmarks while adding only modest inference overhead. https://getnews.me/efficient-test-time-scaling-boosts-small-vision-language-models/ #vlm #testtimescaling
October 7, 2025 at 5:08 PM
LatentEvolve, a self‑evolving test‑time scaling system for LLMs, boosted performance by up to 13.33% across eight benchmark tasks using five model backbones, without changing weights. https://getnews.me/latentevolve-self-evolving-test-time-scaling-for-llms/ #latentevolve #testtimescaling
September 30, 2025 at 11:48 PM
Test-time scaling (TTS) adaptively boosts LLM fact-checking, achieving about 1.8x higher efficiency and an 18.8% accuracy gain, per a study accepted to EMNLP 2025. Read more: https://getnews.me/test-time-scaling-boosts-numerical-claim-verification-in-llms/ #testtimescaling #emnlp2025
September 29, 2025 at 11:44 AM
TTS‑Uniform allocates reasoning attempts evenly across strategies and filters unstable chains, raising accuracy without increasing the sampling budget. Read more: https://getnews.me/mitigating-reasoning-strategy-bias-to-boost-test-time-scaling-in-llms/ #testtimescaling #ttsuniform
September 25, 2025 at 3:57 AM
Kleine KI-Modelle übertreffen große: Effizienz neu definiert

Klein schlägt groß
Effizienz neu definiert
Revolution in der KI

#ai #ki #artificialintelligence #kimodell #effizienz #revolution #testtimescaling

kinews24.de/test-zeit-sc...
February 16, 2025 at 10:42 AM
#NVIDIA commenta il caso #DeepSeek dopo lo scossone finanziario: “DeepSeek è un eccellente avanzamento dell’AI e un perfetto esempio di #TestTimeScaling”.

Scopri di cosa si tratta 👉 ainews.it/nvidia-comme...

#intelligenzaartificiale #tecnologia #innovazione
January 30, 2025 at 2:58 PM