June 2022 - cypherpunks - lists.cpunks.org

Towards a comprehensive approach to counter hate speech
by Gunnar Larson 26 Jun '22

26 Jun '22

Gay. https://unric.org/en/towards-a-comprehensive-approach-to-counter-hate-speec…

1 0

Cryptocurrency: Schneier Bleats, Kasparov Correctly Slams Democracy Threat To Freedom
by grarpamp 26 Jun '22

26 Jun '22

Schneier the unrepentant statist bleats for regulation, bashes entire cryptocurrency space, doesn't even HODL... https://www.schneier.com/blog/archives/2022/06/on-the-dangers-of-cryptocurr… Kasparov brings truth... https://www.youtube.com/watch?v=J3V4neeDVtI https://www.youtube.com/watch?v=xWAwK2fHArc CryptoLiberty's Win Knows No Regulators

1 0

Goldman Sachs, Harvard team up with Al Gore’s investment firm | Business and Economy News | Al Jazeera
by Gunnar Larson 26 Jun '22

26 Jun '22

https://www.aljazeera.com/economy/2021/10/27/goldman-sachs-and-others-team-…

1 0

[ot][spam][crazy] notes/daydreams can maddness do convolution
by Undiscussed Horrific Abuse, One Victim of Many 25 Jun '22

25 Jun '22

i was reading looking at this wonky algorithm i found online and pursued a little in discord it's biggest bottleneck is a call to conv1d(). the author cuda-accelerated it by hand; they are new to cuda. thinking about conv1d via this mithral/maddness stuff, dunno. in a matrix, one combines columns of A with rows of B or something like that. so, the algorithm organizes a bunch of columns and rows, and I think it just precalculates their dotproducts, and then does something equivalent to interpolating the precalculations to do future stuff. if that's true, there's likely a basic analogy with conv1d if i can move the concepts through my mind. if it's not true, those concepts will still do some of the process work for whatever the case actually is. not sure what happened to the multiplication step of the interpolation, if that's what it is, but it seems it has been sorted out. with convolution of 1d data, basically there a bunch of dot products taken between sequences that change in length. [ 1 2 3 4 ... 49 ] convolved with a kernel with indices [ 1 2 3 ... 7 ] i think it produces many dot products, all of length 7, along the length of the data. given the conv1d is performed inside a trained model, the data can indeed be collected from the training. basically, the kernel is roughly a matrix, with all the rows the same. i'm guessing you could reuse the mithral/maddness approach with convolution by turning the convolution kernel into a matrix, and extracting all the sequential chunks of the data to process. it's a naive approach, since the data is so repetitive. but it woudl work well enough to consider. to consider it better, one might think of how each data item is multiplied by every item in the kernel, and summed with its neighbor item. it's interesting to me that in the research they found they could lookup faster than multiply. maybe relates to how near the data is. i've probably often worked with fragmented memory.

1 0

[ot][crazy][spam] Notes: Matmul Acceleration
by Undiscussed Horrific Abuse, One Victim of Many 25 Jun '22

25 Jun '22

Goal: Take implementation-oriented notes on https://arxiv.org/pdf/2106.10860.pdf . My understanding of the approach is that it lossily compresses the data into a smaller representation that contains most of the big ends of the information, combines in a way that involves no add-multiplies, then decompresses it and produces a batch matmul result. The goal is to understand the derivation and implementation enough to daydream about implementing a convolution operation rather than matrix multiplication. [multiple people are working on convolution; i don't know whether i will find their results] Planning to skip the abstract, then start through the intro. 1 Introduction - Approximate Matrix Multiplication (AMM) - assumptions: tall, dense, held in a single address space - A is large data, B is linear operator such as model weights - nonlinear preprocessing reduces problem to table lookups - no multiply-adds if B is constant - a family of quantizers used rather than a single expensive on - 3 parts: 1. high-speed quantization functions 2. high-speed low-bitwidth stable summation algorithm 3. matmul algorithm using 1&2 with theoretical quality guarantees 1.1 Problem Formulation A: NxD B: DxM N >> D >= M tau: computation time bound Problem formulation to minimize error epsilon for computation time bound tau: f(g(A), h(B)) * alpha + Beta - AB [interruption]

1 6

[ot] SOTA Synthetic Scene Video [source]
by Undiscussed Horrific Abuse, One Victim of Many 25 Jun '22

25 Jun '22

If you have the capability to write new code off of research papers, we know how to produce arbitrary new video containing both real and novel humans, animals, plans etc, new behaviors and very large environments. https://dellaert.github.io/NeRF22/ more non-scene-generation stuff from that conference at https://blog.roboflow.com/cvpr-2022-best-papers-and-highlights/

1 0

Fwd: FOIL Request No. 2021-085728 (Larson)
by Gunnar Larson 25 Jun '22

25 Jun '22

I. Business Affiliations of Techsprint Registrants The Department is withholding the business affiliations of Techsprint registrants pursuant to Public Officers Law § 87(2)(b), which provides an exemption from disclosure for records or portions thereof that, “if disclosed, would constitute an unwarranted invasion of personal privacy under the provisions of subdivision two of section eighty-nine of this article.” ---------- Forwarded message --------- From: JeanBaptiste, Pascale (DFS) <Pascale.JeanBaptiste(a)dfs.ny.gov> Date: Fri, Jun 4, 2021, 1:31 PM Subject: FOIL Request No. 2021-085728 (Larson) To: G(a)xNY.io <G(a)xny.io> Cc: Garrett, Carmen (DFS) <Carmen.Garrett(a)dfs.ny.gov> Dear Mr. Larson: Please see the attached determination in response to your FOIL request. Thank you.

1 0

[aimcroleplay][ot][spam] Break Room Between Retraining Sessions
by Undiscussed Horrific Abuse, One Victim of Many 25 Jun '22

25 Jun '22

[placeholder]

1 0

Prosecuting Computer Crimes Computer Crime and Intellectual Property Section Criminal Division
by Gunnar Larson 25 Jun '22

25 Jun '22

183 highlights to the Department of Justice guide to reviewing computer crimes: https://drive.google.com/file/d/11tbgHgDg8qagomO-NBffvIFpxXKmBC3g/view?usp=… The Supreme Court has recognized that the mail and wire fraud statutes sweep more broadly than the common law definition of fraud and false pretenses. Durland v. United States, 161 U.S. 306, 313-14 (1896).5 However, the Court also rejected the notion that every “scheme or artifice that in its necessary consequence is one which is calculated to injure another [or] to deprive him of his property wrongfully.”

1 0

Fwd: FOIL Request
by Gunnar Larson 25 Jun '22

25 Jun '22

Via Email: g(a)xny.io Gunnar Larson RE: FOIL Request #4164 Dear Gunnar Larson: Please be advised that we require additional time to complete our response to your FOIL request dated May 19, 2021. We will provide you with a status update on or before July 25, 2022, if we have not completed our response by then. Thank you for your courtesy in this matter. Very truly yours, Records Access Office Via Email: g(a)xny.io Gunnar Larson RE: FOIL Request #4164 Dear Gunnar Larson: The Executive Chamber acknowledges receipt of your FOIL request dated May 19, 2022, seeking: Any and all records related to the Executive Chamber’s oversight and/or approval of the New York Department of Financial Services’ engagement of the Alliance for Regulatory Innovation and Brex, Inc. We have begun to process your request and we will provide you with a status update on or before June 24, 2022. If any documents are located that respond to your request, you will be charged $.25 per page for photocopies. If the response to your request will be provided in another media, we will notify you of any charges. Very truly yours, Records Access Office ---------- Forwarded message --------- From: Records_Access <Records.Access(a)exec.ny.gov> Date: Fri, Jun 24, 2022, 5:00 PM Subject: FOIL Request To: g(a)xny.io <g(a)xny.io> Dear Gunnar Larson, Please see the attached correspondence in relation to your Freedom of Information Law (“FOIL”) request. Regards, Records Access

1 0