Tootfinder

There's an economic curse on Large Language Models — the crappiest ones will be the most widely used ones.
The highest-quality models are exponentially more expensive to run, and currently are too slow for instant answers or processing large amounts of data.
Only the older/smaller/cut-down models are cheap enough to run at scale, so the biggest deployments are also the sloppiest ones.
#llm

There's an economic curse on Large Language Models — the crappiest ones will be the most widely used ones.
The highest-quality models are exponentially more expensive to run, and currently are too slow for instant answers or processing large amounts of data.
Only the older/smaller/cut-down models are cheap enough to run at scale, so the biggest deployments are also the sloppiest ones.
#llm

ASID: Active Exploration for System Identification in Robotic Manipulation
Marius Memmel, Andrew Wagenmaker, Chuning Zhu, Patrick Yin, Dieter Fox, Abhishek Gupta
https://arxiv.org/abs/2404.12308

ASID: Active Exploration for System Identification in Robotic Manipulation
Model-free control strategies such as reinforcement learning have shown the ability to learn control strategies without requiring an accurate model or simulator of the world. While this is appealing due to the lack of modeling requirements, such methods can be sample inefficient, making them impractical in many real-world domains. On the other hand, model-based control techniques leveraging accurate simulators can circumvent these challenges and use a large amount of cheap simulation data to le…

This https://arxiv.org/abs/2312.01957 has been replaced.
link: https://scholar.google.com/scholar?q=a

Distilled Self-Critique of LLMs with Synthetic Data: a Bayesian Perspective
This paper proposes an interpretation of RLAIF as Bayesian inference by introducing distilled Self-Critique (dSC), which refines the outputs of a LLM through a Gibbs sampler that is later distilled into a fine-tuned model. Only requiring synthetic data, dSC is exercised in experiments regarding safety, sentiment, and privacy control, showing it can be a viable and cheap alternative to align LLMs. Code released at \url{https://github.com/vicgalle/distilled-self-critique}.

Do you want to reduce your home's energy consumption, switch to greener heating, or simply know what's going on in your house? You need data and graphs! Join @… at this year's Berlin Buzzwords to learn how cheap hardware and devops observability tools can help you do just that! #bbuzz

Where does a fitness watch send data to? I bought a cheap MISIRUN C17 to test on an Android test device.
I installed PCAPdroid to check network connections of the DaFit app. I saw wr.moyoung.com (app provider) and android.bugly.qq.com (chinese cloud service). Scary!
Data transfer can be suppressed by simply removing permission of the DaFin app to use any network service (in settings / apps). One would need to do that immediately after first launch in flight mode (with all networ…

Do you want to reduce your home's energy consumption, switch to greener heating, or simply know what's going on in your house? You need data and graphs! Join @… at this year's Berlin Buzzwords to learn how cheap hardware and devops observability tools can help you do just that! #bbuzz

ImplicitSLIM and How it Improves Embedding-based Collaborative Filtering
Ilya Shenbin, Sergey Nikolenko
https://arxiv.org/abs/2406.00198 https://

ImplicitSLIM and How it Improves Embedding-based Collaborative Filtering
We present ImplicitSLIM, a novel unsupervised learning approach for sparse high-dimensional data, with applications to collaborative filtering. Sparse linear methods (SLIM) and their variations show outstanding performance, but they are memory-intensive and hard to scale. ImplicitSLIM improves embedding-based models by extracting embeddings from SLIM-like models in a computationally cheap and memory-efficient way, without explicit learning of heavy SLIM-like models. We show that ImplicitSLIM im…

The Oxford Olympics Study 2024: Are Cost and Cost Overrun at the Games Coming Down?
Alexander Budzier, Bent Flyvbjerg
https://arxiv.org/abs/2406.01714 https://arxiv.org/pdf/2406.01714
arXiv:2406.01714v1 Announce Type: new
Abstract: The present paper is an update of the "Oxford Olympics Study 2016" (Flyvbjerg et al. 2016). We document that the Games remain costly and continue to have large cost overruns, to a degree that threatens their viability. The IOC is aware of the problem and has initiated reform. We assess the reforms and find: (a) Olympic costs are statistically significantly increasing; prior analysis did not show this trend; it is a step in the wrong direction. (b) Cost overruns were decreasing until 2008, but have increased since then; again a step in the wrong direction. (c) At present, the cost of Paris 2024 is USD 8.7 billion (2022 level) and cost overruns is 115% in real terms; this is not the cheap Games that were promised. (d) Cost overruns are the norm for the Games, past, present, and future; they are the only project type that never delivered on budget. We assess a new IOC policy of reducing cost by reusing existing venues instead of building new ones. We find that reuse did not have the desired effect for Tokyo 2020 and also look ineffective for Paris 2024. Finally, we recommend that the Games look to other types of megaprojects for better data, better forecasting, and how to generate the positive learning curves that are necessary for bringing costs and overrun down. Only if this happens are Los Angeles 2028 and Brisbane 2032 likely to live up to the IOC's intentions of a more affordable Games that more cities will want to host.

ImplicitSLIM and How it Improves Embedding-based Collaborative Filtering
Ilya Shenbin, Sergey Nikolenko
https://arxiv.org/abs/2406.00198 https://

ImplicitSLIM and How it Improves Embedding-based Collaborative Filtering
We present ImplicitSLIM, a novel unsupervised learning approach for sparse high-dimensional data, with applications to collaborative filtering. Sparse linear methods (SLIM) and their variations show outstanding performance, but they are memory-intensive and hard to scale. ImplicitSLIM improves embedding-based models by extracting embeddings from SLIM-like models in a computationally cheap and memory-efficient way, without explicit learning of heavy SLIM-like models. We show that ImplicitSLIM im…