State of the Model Serving Communities - August 2025 by @terrytangyuan
https://inferenceops.substack.com/p/state-of-the-model-serving-communities

`His initial intended uses were for linguistic analysis and other mathematical subjects like card shuffling, but both Markov chains and matrices rapidly found use in other fields.`

https://en.wikipedia.org/wiki/Stochastic_matrix#History

#statstab #391 {sensemakr} Sensitivity Analysis Tools for OLS

Thoughts: No unobserved variables is an untestable assumption, but you can quantify the robustness of your ATE.

#R #causalinference #observational #inference #confounding #bias #sensitivity

https://carloscinelli.com/sensemakr/

Laptop LLM: You can run decent AIs on a small computer
https://www.technologyreview.com/2025/07/17/1120391/how-to-run-an-llm-on-your-laptop/
#inference #hardware #llama #llm #ai #+

#GPUHammer is the first attack to show #Rowhammer bit flips on #GPU memories, specifically on a GDDR6 memory in an #NVIDIA A6000 GPU. Our attacks induce bit flips across all tested DRAM banks, despite in-DRAM defenses like TRR, using user-level #CUDA #code. These bit flips allow a malicious GPU user to tamper with another user’s data on the GPU in shared, time-sliced environments. In a proof-of-concept, we use these bit flips to tamper with a victim’s DNN models and degrade model accuracy from 80% to 0.1%, using a single bit flip. Enabling Error Correction Codes (ECC) can mitigate this risk, but ECC can introduce up to a 10% slowdown for #ML #inference workloads on an #A6000 GPU.

https://gpuhammer.com/

#statstab #383 Berkson's paradox

Thoughts: aka Berkson's bias, collider bias, or Berkson's fallacy. Important for interpreting conditional probabilities. Can produce counterintuitive patterns.

#paradox #collider #bias #inference #causalinference

https://en.m.wikipedia.org/wiki/Berkson's_paradox

#statstab #370 The Problem with “Magnitude-based Inference”

Thoughts: An appealing but flawed approach. Good overview of the error inflation issue.

#MBI #errorrate #sportsscience #sports #inference #critique

https://journals.lww.com/acsm-msse/fulltext/2018/10000/the_problem_with__magnitude_based_inference_.23.aspx

#statstab #368 The FisherlPearson Chi-Squared Controversy: A Turning Point for
Inductive Inference

Thoughts: An overview of the difference between Pearson's descriptive view and Fisher's inferential view of X2.

#fisher #pearson #inference #chisquared

https://genepi.qimr.edu.au/contents/p/staff/1983BairdBJPS105-118.pdf

Tokasaurus: An LLM Inference Engine for High-Throughput Workloads

https://scalingintelligence.stanford.edu/blogs/tokasaurus/

'On Consistent Bayesian Inference from Synthetic Data', by Ossi Räisä, Joonas Jälkö, Antti Honkela.

http://jmlr.org/papers/v26/23-1428.html

#bayesian #privacy #inference

#statstab #346 Jeffreys-Lindley paradox

Thoughts: I like this short explanation of the "paradox" of why frequentist and bayesian inference can differ.

#paradox #frequentist #bayesian #inference #bayesfactor #pvalue #explanation

https://michael-franke.github.io/intro-data-analysis/jeffreys-lindley-paradox.html

'DAGs as Minimal I-maps for the Induced Models of Causal Bayesian Networks under Conditioning', by Xiangdong Xie, Jiahua Guo, Yi Sun.

http://jmlr.org/papers/v26/23-0002.html

#inference #causal #bayesian

NVIDIA Dynamo: Scaling AI inference with open-source efficiency https://www.artificialintelligence-news.com/news/nvidia-dynamo-scaling-ai-inference-open-source-efficiency/ #nvidia #dynamo #opensource #inference #ai #tech #news #technology

“How ‘inference’ is driving competition to Nvidia’s #AI chip dominance”

#NVidia / #reasoning / #inference <https://archive.md/AYHs7>

How to #backdoor a #LLM

https://blog.sshh.io/p/how-to-backdoor-large-language-models

Tenacity, Authority, Plausibility, Inquiry
• https://inquiryintoinquiry.com/2025/02/14/tenacity-authority-plausibility-inquiry-a/
• https://bsky.app/profile/inquiryintoinquiry.bsky.social/post/3li5nmqdc3s2a

Re: Peter Cameron • Mathematics and Logic
• https://cameroncounts.wordpress.com/2010/01/03/mathematics-and-logic/

My favorite polymathematician, Charles Sanders Peirce, gave a fourfold classification of what he called “methods of fixing belief”, or “settling opinion”, most notably and seminally in his paper, “The Fixation of Belief” (1877). Adjusting his nomenclature very slightly, if only for the sake of preserving a mnemonic rhyme scheme, we may refer to his four types as Tenacity, Authority, Plausibility (à priori pleasing praiseworthiness), and full‑fledged Scientific Inquiry.

Reference —

Peirce, C.S. (1877), “The Fixation of Belief”, Popular Science Monthly 12, 1–15.
• https://www.cspeirce.com/menu/library/bycsp/fixation/fx-frame.htm

#Peirce #Logic #Mathematics #Belief #Opinion #Knowledge #Inference
#BeliefFixation #Method #Tenacity #Authority #Plausibility #Inquiry

Cerebras CS-3

Cerebras’ CS-3 aims to cram cluster-level AI performance into a dorm-sized fridge by dint of building off a single wafer-scale chip that carries on-chip memory and interconnects, m(...)

#ai #cerebras #hardware #inference #nvidia #silicon #training #wafer

https://taoofmac.com/space/links/2025/02/11/1300

How is #censorship implemented in #deepseek? A link to #wikipedia referring to the #tienanmen square can spark an #ethical judgment on the #chinese government. Of course it dare not speak its name
Since censorship is active also on locally run model, probably it is implemented toward the last steps of #inference, while the training set was hastily used and not “curated” for censorship. #AI #LLM