Judge.me

judgeme

USA v. Samuel Bankman-Fried Court Filing, retrieved on March 15, 2024 is part of . You can jump to any part in this filing . This part is 31 of 33. HackerNoon’s Legal PDF Series here II. The Court Should Order Forfeiture The Court should impose the proposed forfeiture money judgment. This number is derived as follows: First, the Court should direct the forfeiture of $8,000,000,000, which is a conservative estimate of the proceeds of the defendant’s wire fraud and conspiracy to commit wire fraud on FTX’s customers, and the property involved in his conspiracy to launder the proceeds of that wire fraud. As funds that the defendant obtained through false statements and promises, and then misappropriated, they are plainly “proceeds” pursuant to 18 U.S.C. § 981(a)(1)(C). See Bonventre, 646 F. App’x at 90 (calculating forfeiture based on “an amount representing the total client investment”); United States v. Kenner, 443 F. Supp. 3d 354, 378 (E.D.N.Y. 2020) (defining proceeds forfeiture as amount “received . . . as a result of the fraud.”). The total of $8 billion also represents property involved in the defendant’s money laundering. The reason FTX had such a deficit was that the defendant and his co-conspirators had agreed to launder those funds to and through Alameda. Second, the Court should direct the forfeiture of $1,300,000,000, which represents the proceeds of the defendant’s wire fraud and conspiracy to commit wire fraud on Alameda’s lenders. This sum represents the amount that Alameda owed lenders to whom it had defrauded by making false statements about its balance sheet, calculated as of the date of the bankruptcy filing. (GX1014). Third, the Court should direct the forfeiture of $1,720,000,000, which represents the proceeds of the defendant’s securities fraud. This sum is the amount FTX raised from investors in exchange for equity on fraudulent pretenses. (GX-26 (listing the total amounts invested in the Series B, B-1, and C round investments in FTX.com)). The defendant, however, asserts that he owes no forfeiture because “customers and creditors will receive a return of their funds via the bankruptcy.” (Def. Mem. at 88). This argument is factually unsupported and legally wrong. Starting with the facts, as discussed above, it is far from certain that FTX customers and lenders will be made whole, and the bankruptcy will almost certainly not cover losses by FTX’s equity investors (the victims of the conspiracy to commit security fraud), which were approximately $1.72 billion. As a matter of law, there is no basis for reducing the forfeiture judgment based on repayment to victims in bankruptcy. For money laundering based forfeiture under 18 U.S.C. § 982(a)(1), there is no suggestion in the forfeiture statutes that repayment to victims could reduce the judgment. Indeed, Section 982(a)(1) contemplates the opposite, since it is based on the “property involved” in the money laundering. See United States v. Elfgeeh, 515 F.3d 100, 139 (2d Cir. 2008) (for money transmitting offense, which similarly applies Section 982(a)(1) for forfeiture, affirming judgment for total amount transmitted even though there are no victims); United States v. Waked Hatum, 969 F.3d 1156, 1164 (11th Cir. 2020) (in money laundering case where victims suffered no loss, affirming forfeiture judgment of total sum of laundered money, because “the government’s interest in the [laundered property] vests ‘the moment’ such property is laundered”). The same is true for wire fraud based forfeiture. While the defendant asserts that the proper measure of forfeiture here is the net proceeds under Section 981(a)(2)(B), rather than gross proceeds under Section 981(a)(2)(A), (Def. Mem. at 87-88), he is wrong. As the trial record showed, the defendant embezzled billions of dollars of FTX customer funds, and induced investors to part with money based on false statements. That is “inherently unlawful” conduct and therefore the gross proceeds rule under Section 981(a)(2)(A) applies. See United States v. Bodouva, 853 F.3d 76, 80 (2d Cir. 2017) (Section 981(a)(2)(A) applies to embezzlement); Milton, 2024 WL 779210, at *5 (selling securities for which prices have been inflated by “misleading statements” is “more analogous to cases involving Ponzi schemes” and thus Section 981(a)(2)(A) applies). And under Section 981(a)(2)(A), the law is clear: “there is no statutory authorization for an offset” even for payments to victims. Bodouva, 853 F.3d at 80. The single case cited by the defendant for the proposition that return of funds through bankruptcy should offset victim losses is a decade-old, distinguishable decision from the Northern District of Illinois. See United States v. Hollnagel, No. 10 Cr. 195, 2013 WL 5348317 (N.D. Ill. Sept. 24, 2013). There, the district court reduced a forfeiture judgment to zero because the “[d]efendants returned the investors’ investments over time” before their arrests. Id. at *4. Not only did the defendant not return the $8 billion he stole before his arrest, but the Second Circuit has explicitly rejected the reasoning of Hollnagel, under Section 981(a)(2)(B). United States v. Shkreli, 779 F. App’x 38 (2d Cir. 2019) (summary order). In Shkreli, the Court held that because “‘forfeiture is gain based’, not based on the losses (or gains) to victims,” a defendant should be required to forfeit “the total amount invested by investors” even if investors were paid back by the defendant during the scheme. Id. at 42 (citing Torres, 703 F.3d at 203). The defendant next asserts that “none of the accounts identified for forfeiture were for [his] personal benefit,” “were not profits ‘enjoyed’ by [him], but rather maintained on behalf of the companies,” and therefore are not subject to forfeiture. (Def. Mem. at 88). This is legally and factually wrong. While the word “enjoyed” is found in Hollnagel, the law is clear that the Government can forfeit corporate assets based on a defendant’s crimes when those assets were maintained through and on behalf of a company the defendant “extensively controls” or “dominates,” such that “money paid to the corporation was effectively” under the defendant’s control. United States v. Peters, 732 F.3d 93, 103-04 (2d Cir. 2013) (forfeiture proceeding under Section 982); United States v. Tartaglione, 15 Cr. 491, 2018 WL 1740532, at *21 (E.D. Pa. Apr. 11, 2018) (same for Section 981), aff’d, 815 F. App’x 648 (3d Cir. 2020). The trial evidence of course established the defendant’s control and dominance of FTX, Alameda, and its affiliated entities. Finally, the defendant asserts that the Court should order no forfeiture because he “no longer has custody of the accounts identified for forfeiture in the PSR” since the Government seized the accounts. (Def. Mem. at 88). This is utterly meritless—there is no authority even suggesting that the Government’s seizure of assets traceable or involved in a defendant’s crimes somehow precludes the Government from seeking forfeiture of the defendant’s interest in those assets. Indeed, the forfeiture laws expressly authorize seizure as part of the forfeiture process. In order for the Government to effect forfeiture of the specific property described in the preliminary order of forfeiture, the Court must first order forfeiture of the defendant’s interests in that specific property. United States v. Daugerdas, 892 F.3d 545, 549 (2d Cir. 2018) (“At stage one of [forfeiture process], before entering a preliminary order of forfeiture, the court is directed to adjudicate the government’s interest vis-à-vis the defendant”). Only once the Court has forfeited the defendant’s rights in the property can the Court move on to “stage two,” by “resolv[ing] any third-party petitioner’s interests vis-à-vis the defendant.” Id. If no third-party has a cognizable interest in the property superior to the defendant’s, then the specific property can be finally forfeited to the Government. In the preliminary order of forfeiture, the Government lists specific property beyond that listed in the forfeiture allegations in the S6 Indictment and the bill of particulars. (Dkt. 314).[14] That specific property is all forfeitable as proceeds of the defendant’s wire frauds or as property involved in money laundering. The property listed in Ex. A(1) is another account in the name of Emergent Fidelity Technologies, which was seized at the same time and pursuant to the same legal authority as the property listed in the S6 Indictment ¶¶ 38(a), (b), 40(a), (b). Ex. A(2)-(255) are the specific forms of cryptocurrency maintained in the Binance accounts described in S6 Indictment ¶¶ 38(h), (i), (j), 40(h), (i), (j). Ex. A(258) is a bank account in the name of the defendant and another FTX employee, which received transfers from the account at Silvergate Bank in the name of Alameda ending in -4456, which as described in Government Exhibit 1050 accepted customer fiat deposits. The Government also seeks to forfeit the defendant’s interests in certain contributions traceable to his criminal conduct that he, Salame, and Singh made to political candidates and committees, during the relevant period. Beginning in March 2023, the Government, working with the United States Marshals Service and the FBI, sent letters to those candidates and committees who the Government had identified had received donations from the defendant, Salame, and Singh, requesting return of the donations as forfeitable property, and because the donations were illegal. As of today, 251 candidates and committees have returned $3,348,957.23. [15] The Preliminary Order of Forfeiture lists the identified campaigns, including those that have returned funds to the Government, in Ex. A(259) to (261). As evidenced at trial, the defendant’s interests in the political donations are forfeitable because the donations were made using the proceeds of the defendant’s wire fraud on customers and were themselves manifestation of his money laundering. (GX-1039; 1044; 1088; 1089; 1090). Continue Reading . Here [14] The airplanes listed at Ex. A(256) and (257) are the property listed in the bill of particulars. [15] The Government is also coordinating with the FTX Debtors, so that the estate retrieves some of the political contributions. In some circumstances, a political candidate or committee has sought to or did return funds to the estate rather than the Government. As of today, the Government understands that the estate has received $281,356 from 8 candidates and committees. About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on March 15, 2024, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. storage.courtlistener

Court Imposes $10 Billion Forfeiture Judgement on Sam Bankman-Fried

TOPTAL, LLC v. DENIS GROSZ Court Filing, retrieved on February 26, 2024 is part of . You can jump to any part in this filing . This part is 1 of 7. HackerNoon’s Legal PDF Series here This matter comes before the Court on Toptal, LLC's (“Toptal”) Motion for Summary Judgment (“Motion”). In the Motion, Toptal argues that Denis Grosz (“Mr. Grosz”) cannot prove that Toptal (a) breached the Note Purchase Agreement (the “NPA”), convertible promissory note (the “Note”), or (b) the Advisor Agreement entered into by the parties on November 14, 2012[1]; that Mr. Grosz's declaratory judgment counterclaims are superfluous; and that Mr. Grosz's counterclaims are barred by the statutes of limitations. The Court has considered the parties’ motion papers, the'arguments by the parties on September 29, 2023, and October 9, 2023, and the supplemental briefing submitted by the parties on October 4, 2023. The Court finds that there are no material disputed questions of fact with respect to part (a) of Toptal's Motion - whether Toptal breached the Note Purchase Agreement (the “NPA”) and convertible promissory note (the “NOTE”) and hereby grants that part of Toptal's Motion. Given this determination, it is not statute of limitations. California Code Section 360 cannot extend the 15 day period to elect the conversion even under Grosz's reading of the statute to extend the statute of limitations period for contracts. [2] As to Mr. Grosz's counterclaim for breach of the Advisor Agreement, Toptal argues that Mr. Grosz cannot show breach of contract because Mr. Grosz has not submitted evidence of damages. Grosz rests his damage model on the testimony of expert Gregory Nachtwey. The Court damages. Grosz rests his damage model on the testimony of expert Gregory Nachtwey. The Court will defer ruling on this counterclaim until after the hearing on the Motion in Limine concerning the admissibility of Mr. Nachtwey's opinions. FACTS AND PROCEDURAL HISTORY* In 2010, Mr. Du Val founded Toptal, a single-member LLC. Mr. Du Val is and always has been the CEO and sole member of Toptal. Mr. Grosz provided Toptal $1 million pursuant to the NPA on November 14, 2012, and, in exchange, Toptal issued him the Note. As part of that transaction, the parties also entered into an Advisor Agreement on the same day, pursuant to which Mr. Grosz agreed to advise the company in exchange for the opportunity to purchase 1.5% equity in Toptal subject to certain conditions that have not occurred to this day. Mr. Grosz asserts that Toptal breached the NPA and Note by (i) failing to convert to a C-corporation and give him equity upon his written election on November 24, 2015, and/or (ii) attempting to repay his principals, and accrued interest on the Note in March 2020. Several provisions of the NPA and Note are relevant to the Motion. First, the Note states that “[u]nless earlier converted into Conversion Shares pursuant to Section 2.2 of that certain Note Purchase Agreement dated November    , 2012 among the Company and the Lender (the ‘Purchase Agreement’), the principal and accrued interest shall be due and payable by the Company on demand by the Lender at any time after the Maturity Date.”   The Note contains a provision expressly waiving presentment and demand. The NPA defines the “Maturity Date” as “May 14, 2014.” Second, Section 2.2 of the NPA is titled, .”' Section 2.2(a) provides that “[t]he principal and unpaid accrued interest of the Note will be automatically converted into Conversion Shares upon the closing of the Next Equity Financing....” A Next Equity Financing Conversion Shares upon the closing of the Next Equity Financing....” A Next Equity Financing (a defined term in the NPA) did not take place prior to the Maturity Date and has not taken place to this day. Section 2.2(b) of the NPA states: “ At any time on or after the fifteenth (15th) day prior to the Maturity Date, if the Next Equity Financing has not occurred by that time, the Note, at Lender's election by written notice to the Company, will be converted into Common Stock of the Company such that the Lender receives shares constituting 8.333% of the Fully-Diluted Equity Securities of the Company following such conversion.” “ Rieht to Convert Note Optional Conversion. The parties dispute whether Section 2.2(b) provides Mr. Grosz with a 15-day window  before the Maturity Date in which to make an election to convert his Note to equity. For its part, Toptal argues that Mr. Grosz's ability to make an election under Section 2.2(b) ended on the Maturity Date. On the other hand, Mr. Grosz argues that Section 2.2(b) allows him to elect to convert his Note starting 15 days prior to the Maturity Date and continuing forever into the future. The parties agree that Mr. Grosz did not make an election under Section 2.2(b) prior to the Maturity Date. The parties also dispute the meaning of Section 2.2(c) of the NPA, which states: “ For clarification, the Company agrees to complete a Company Conversion prior to any conversion of the Note pursuant to this Section 2.2.” Mr. Grosz claims that this provision, read in connection with Section 2.2(b), grants him the power to force Toptal to complete a Company Conversion. In other words, Mr. Grosz claims that once he makes an election under Section 2.2(b), Toptal must convert from an LLC to a C-corporation under Section 2.2(c) so that it can issue Mr. Grosz equity. Toptal disagrees that Section 2.2(c) grants Mr. Grosz the power to force Toptal to change its corporate form. Companv Conversion Prior to Note Conversion. The NPA also contains a section titled ” which provides that “[t]he following events shall be considered Events of Default with respect to the Note following, and only upon, the declaration of default by the Note holder under Section 7.2: “ Events of Default, (a) The Company shall default in the payment of any part of the principal or unpaid accrued interest on the Note for more than thirty (30) days after the Maturity Date or at a date fixed by acceleration or otherwise;. (b) The Company shall fail to observe or perform any other obligation to be observed or performed by it under this Agreement or the Note, within 30 days after written notice from the Note holder to perform or observe the obligation.” Section 7.2 of the NPA states: “Remedies. Upon the occurrence of an Event of Default under Section 7.1 hereof, only at the option and upon the declaration of the holder of the Note, the entire unpaid principal and accrued and unpaid interest on the Note shall, without presentment, demand, protest, or notice of any kind, all of which are hereby expressly waived, be forthwith due and payable, and such holder may, immediately and without expiration of any period of grace, enforce payment of all amounts due and owing under the Note and exercise any and all other remedies granted to it at law, in equity or otherwise.” It is undisputed that Toptal did not convert to a C-corporation before the May 14, 2014, Maturity Date or pay any outstanding principal and interest to Mr. Grosz within 30 days after the Maturity Date. Continue Reading . Here [I] There are no fully executed copies of these agreements in the Record. The Court has reviewed a copy of the NPA and the Advisor Agreement containing Taso Du Val's (“Mr. Du Val”) signature on behalf of Toptal, but those agreements are missing Mr. Grosz's signature. The Court has not received an executed copy of the Note. However, the Court does not address whether the lack of signatures by all parties renders the Agreements unenforceable, because Count I of Mr. Grosz's counterclaims fails as a matter of law in any event. [2] Toptal's other arguments that Mr. Grosz's remaining counterclaims are untimely is addressed in a separate order. [3] The facts set forth in this Order are the facts “viewed in the light most favorable to the non-movant,” Mr. Grosz. 121 Nev. 724, 729 (2005) (“[W]hen reviewing a motion for summary judgment, the evidence, and any reasonable inferences drawn from it, must be viewed in a light most favorable to the nonmoving party.”); 118 Nev. 288, 293 (2002); 26 Carrington Mortg. Servs., 139 Nev. Adv. Op. 13 (2023). Wood v. Safeway, Inc., Chavez v. Sievers, see also Wishengrad v. About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on October 11, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. media.licdn.com

Toptal, LLC v. Denis Grosz: Summary Judgement and Court Filing Analysis

ERIK ADOLPH vs. Uber Court Filing, retrieved on July 17, 2023, is part of . You can jump to any part in this filing . This is part 15 of 15. HackerNoon’s Legal PDF Series here CONCLUSION We reverse the judgment of the Court of Appeal and remand the case for further proceedings consistent with this opinion. We limited our review to the question of PAGA standing and express no view on the parties’ arguments regarding the proper interpretation of the arbitration agreement. LIU, J. We Concur: GUERRERO, C. J. CORRIGAN, J. KRUGER, J. GROBAN, J. JENKINS, J. EVANS, J About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case S274671 retrieved on September 22, 2023, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. courts.ca.gov

The Judgement of the Court of Appeals Is Reversed in the Uber Lawsuit

Judgement Day.

On-Line Data-Acquisition Systems in Nuclear Physics, 1969, by H. W. Fulbright et al.  National Research Council is part of the HackerNoon Books Series.
 You can jump to any chapter in this book here . Chapter 4: EXERCISING ECONOMIC JUDGMENT IN PLANNING C. EXERCISING ECONOMIC JUDGEMENT IN PLANNING Since the ultimate criterion is research output, the role assigned to a computer system must depend on the nature of the work being planned. In some cases where a very specific use is intended, for example, in the case of a process-control application such as the Argonne plate scanner or an accelerator controller, the conditions are simple enough to make economic judgment relatively easy to apply. In the case encountered in setting up an accelerator laboratory where a wide variety of experiments is to be performed, conditions are much more complex. It is now widely accepted that any such laboratory should have a computer system, but what is not so clear is how extensive and expensive it should be. In other words, points 1-4 in B are accepted, and point 5 is conceded possibly to be applicable. If sufficient funds are available, one sensible way to proceed is to use the accumulated collective experience outlined above. For example, one can say that experience has shown that the total investment in the computer system will be in reasonable balance with the capital investment in the bare accelerator if the ratio of costs is about one to five. Departures from the rule may then be made to adjust to special circumstances. Following this procedure means extrapolating from past experience, which may not prove a good guide, but this approach is similar to that often used in other matters bearing on the support of research. Probability is involved. It should be noted that the actual expenditures for on-line equipment for nuclear research have far exceeded those projected at the "Grossinger Conference on the Utilization of Multiparameter Analyzers in Nuclear Physics" in 1962. In times of economic stringency it may be necessary to take a hard look at points 3-5 in B above before deciding how large a computer can be justified. A medium-sized computer is sufficient for most data-acquisition demands but not for large-scale calculations of a theoretical nature or for an occasional complicated piece of data reduction. Often it will be advisable to plan on carrying out all large calculations at the computing center, in which case a medium-sized computer will probably suffice for data acquisition, and a saving of about half in capital investment and operating expenses can be achieved. About HackerNoon Book Series: We bring you the most important technical, scientific, and insightful public domain books. This book is part of the public domain. H. W., Fulbright et al. 2013. On-Line Data-Acquisition Systems in Nuclear Physics, 1969. Urbana, Illinois: Project Gutenberg. Retrieved May 2022 from https://www.gutenberg.org/files/42613/42613-h/42613-h.htm#Page_70 This eBook is for the use of anyone anywhere at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this eBook or online at www.gutenberg.org , located at https://www.gutenberg.org/policy/license.html .

Exercising Economic Judgement in Planning

An Essay Concerning Humane Understanding, Volume II, by John Locke is part of HackerNoon’s Book Blog Post series. You can jump to any chapter in this book here: [ LINK TO TABLE OF LINK ]. Volume II: Book IV, Chapter XIV: OF JUDGMENT. CHAPTER XIV. OF JUDGMENT. 1. Our Knowledge being short, we want something else. The understanding faculties being given to man, not barely for speculation, but also for the conduct of his life, man would be at a great loss if he had nothing to direct him but what has the certainty of true knowledge. For that being very short and scanty, as we have seen, he would be often utterly in the dark, and in most of the actions of his life, perfectly at a stand, had he nothing to guide him in the absence of clear and certain knowledge. He that will not eat till he has demonstration that it will nourish him; he that will not stir till he infallibly knows the business he goes about will succeed, will have little else to do but to sit still and perish. 2. What Use to be made of this twilight State. Therefore, as God has set some things in broad daylight; as he has given us some certain knowledge, though limited to a few things in comparison, probably as a taste of what intellectual creatures are capable of to excite in us a desire and endeavour after a better state: so, in the greatest part of our concernments, he has afforded us only the twilight, as I may so say, of probability; suitable, I presume, to that state of mediocrity and probationership he has been pleased to place us in here; wherein, to check our over-confidence and presumption, we might, by every day's experience, be made sensible of our short-sightedness and liableness to error; the sense whereof might be a constant admonition to us, to spend the days of this our pilgrimage with industry and care, in the search and following of that way which might lead us to a state of greater perfection. It being highly rational to think, even were revelation silent in the case, that, as men employ those talents God has given them here, they shall accordingly receive their rewards at the close of the day, when their sun shall set, and night shall put an end to their labours. 3. Judgement or assent to Probability, supplies our want of Knowledge. The faculty which God has given man to supply the want of clear and certain knowledge, in cases where that cannot be had, is JUDGEMENT: whereby the mind takes its ideas to agree or disagree; or, which is the same, any proposition to be true or false, without perceiving a demonstrative evidence in the proofs. The mind sometimes exercises this judgment out of necessity, where demonstrative proofs and certain knowledge are not to be had; and sometimes out of laziness, unskilfulness, or haste, even where demonstrative and certain proofs are to be had. Men often stay not warily to examine the agreement or disagreement of two ideas, which they are desirous or concerned to know; but, either incapable of such attention as is requisite in a long train of gradations, or impatient of delay, lightly cast their eyes on, or wholly pass by the proofs; and so, without making out the demonstration, determine of the agreement or disagreement of two ideas, as it were by a view of them as they are at a distance, and take it to be the one or the other, as seems most likely to them upon such a loose survey. This faculty of the mind, when it is exercised immediately about things, is called JUDGEMENT; when about truths delivered in words, is most commonly called ASSENT or DISSENT: which being the most usual way, wherein the mind has occasion to employ this faculty, I shall, under these terms, treat of it, as feast liable in our language to equivocation. 4. Judgement is the presuming Things to be so, without perceiving it. Thus the mind has two faculties conversant (about truth and falsehood):— First, KNOWLEDGE, whereby it certainly PERCEIVES, and is undoubtedly satisfied of the agreement or disagreement of any ideas. Secondly, JUDGEMENT, which is the putting ideas together, or separating them from one another in the mind, when their certain agreement or disagreement is not perceived, but PRESUMED to be so; which is, as the word imports, taken to be so before it certainly appears. And if it so unites or separates them as in reality things are, it is right judgement. About HackerNoon Book Series: We bring you the most important technical, scientific, and insightful public domain books. This book is part of the public domain. Locke, John. 2004. An Essay Concerning Humane Understanding, Volume II. Urbana, Illinois: Project Gutenberg. Retrieved May 2022 from https://www.gutenberg.org/cache/epub/10616/pg10616-images.html This eBook is for the use of anyone anywhere at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this eBook or online at www.gutenberg.org , located at https://www.gutenberg.org/policy/license.html .

Understanding Judgement 

We are banking online, shopping online, using social networking sites, playing online games, storing photos, audio, and video online and using online file-sharing. And mostly all those services are cloud-based. For example, such famous platforms as Airbnb, PayPal, Zynga, Spotify and Adobe Systems are “cloudy”. – Daddy, what are clouds made of? – Linux servers, mostly. (ba dum tss!) Cloud computing is one of the most promising and discussed fields in the kingdom of information technologies. This a common term which most have heard about but not everyone knows what exactly it means and what is going on under the hood of сloud machinery. We are banking online, shopping online, using social networking sites, playing online games, storing photos, audio, and video online and using online file-sharing. And mostly all those services are cloud-based. For example, such famous platforms as Airbnb, PayPal, Zynga, Spotify and Adobe Systems are “cloudy”. It means that cloud computing technologies have long been a part of our daily usage. The only question is how well do we understand what does it mean? CLOUD IS NOT A TECHNOLOGY, IT'S A SOLUTION Let's mark, that cloud computing is defined as a MODEL, PARADIGM, sometimes SOLUTION, but not exactly a technology, platform or product. “Cloud computing architecture is something you do, not something you buy“. David Linthicum, author on InfoWorld — If we make an extract from different definitions of cloud computing, we will get something like this: put in simple words, cloud computing is about storing, accessing, and managing a wide range of data and programs over the Internet instead of your hardware. It is rooted in the good times of old school presentations showing the giant Internet infrastructure in the shape of a cloud that accepts connections. And if we need more deep and scientific meaning let’s take this one: NAME IS DECEPTIVE, OR SERVERLESS IS NOT A SERVERLESS… As far back as 1994, scientists used a cloud-shaped drawing to portray clusters of services that could be delivered over the Internet. But still, the term "cloud" is staying blurred for many people. Hope we have already figured out a "cloud computing" term, let's move on. One much more confusing term is "serverless computing" as one of a cloud IT model, that actually doesn't eliminate servers from distributed applications. “The term "serverless" is annoying as heck. Sure, you're not fiddling with nginx configs, but you're even more tightly dependent on running in a cloud environment. It should be "servermore". — skewart on Hacker News according to Amazon. In the classic cloud model, you are paying for a constantly running server or container. Serverless apps have functions running in the response of such events as file uploads or web requests. The most famous computing platform for such a deployment model is AWS Lambda. "Serverless applications don't require you to provision, scale, and manage any servers," It's kinda equivalent to a timeshare condo vs renting a condo. Your functions only run when they are called. — quaz 31 on Hacker News CLOUD COMPUTING IS NOT REMOTE STORAGE ONLY Concepts and are muddled up often. This couple is interconnected but not the same. Remote cloud storage systems, such as Google Drive, Dropbox, or iCloud, may enable you to host, share, sync and edit files only from multiple devices using the Internet. "cloud storage" "cloud computing" Such a system allows clients to send files to data servers on the Internet, instead of storing it on a computer or external storage device. And a cloud computing machine is more than permanent store or data sharing tool. It's abilities are much wider. It's about the use of hardware and software to deliver varied services over a network while cloud storage is simply storing data and files, and performing backups for completing the process. THE POWER OF THREE: IAAS, PAAS, SAAS… AND SOMETHING MORE Cloud computing building consists of 3 basic levels: the lowest is . It is a basis of 'bare metal' providing the physical servers, device storage, networking components. is the middle that maintains an environment for hosting user apps, such as virtual servers, operating and database environments. It's demanded to develop, run, and manage apps without the need for infrastructure organization. PaaS is a simple way to deploy your app in a given technology (e.g., Node.js, .NET, Python, Java, Ruby, etc). The top is providing tan out-of-the-box software including service on demand. It's a software distribution model where a third-party vendor hosts apps and makes them available to clients over the Internet. In SaaS a lot of areas can be managed by providers: apps, runtime, data, OSes, virtualization, servers, storage, etc. One more room on this top floor is as a subset of SaaS. More specifically, it's an API (usually HTTP) returning some data e.g., currency exchange rates or weather forecasts, DaaS examples: Google Maps, Google Translate API, AccuWeather. But that is not a full-house yet. Let's look on upstairs. There we can single out a Function-as-a-Service (FaaS), an event-based architecture: functions can be triggered by a given event. It implies a very high level of abstraction that it's called a serverless architecture (see above: name is deceptive ;) ). To quickly relate let's consider Google's tools: Infrastructure-as-a-service (Iaas) Platform-as-a-Service (PaaS) Software-as-a-Service (SaaS) Data-as-a-Service (DaaS) HYBRID OR NOT HYBRID: THAT IS THE QUESTION It's common to identify three ways of deploying cloud resources: private, public and hybrid. While the first duo is clear for understanding, the third one is still confusing. Let's try to handle this. Cloud computing becomes into a most real omni-cloud journey. The optimal cloud solution for most businesses is neither public or private, but both. Traditionally, it is called a hybrid cloud, that mixes different ways of deployment. To be honestly, a hybrid approach has spawned a lot of misunderstandings. In fact, such a model doesn't match the common term "hybrid", that is traditionally defined as something that has been bred from two different species as an individual object and primarily involves crossbreeding as process. But in fact the hybrid cloud deployment is a specific strategy where data may intermingle between private and the public clouds, but it remains with the same workload. So the hybrid cloud is about distributing app components in various clouds and sharing data between them while the heart of the process, i.e. actual executables are functioning in only one cloud environment. “Much of the adaptation to hybrid/multi-cloud focuses on how to move data back and forth, how to run applications on disparate hypervisors and how to bring non-standard services together. — Chris Evans, author on Architecting.it Now we are getting close to multi-cloud concept. In accordance with Gartner, multicloud strategies will reduce vendor dependency for two-thirds of organizations through 2024. This strategy implies mixing multiple cloud services, whereas they are not connected or orchestrated between themselves. Such multi-vendor model gives a great opportunity to obtain the most efficacious solutions for business, while hybrid cloud refers specifically to the specific tech portfolio. Let's mark that multi-cloud strategy may include hybrid cloud as a part, but a hybrid cloud solution isn't always a multi-cloud one. THERE IS NO BEST OF THE BEST CLOUD SERVICE Cloud tech is gaining huge attention for today. Such corporate giants as Microsoft, IBM, Google, Amazon provide clouds and successfully use it. These systems are all kind of different but they have the same character, i.e. the same concept. Depending on the features we are looking for, we can try several services and make use of benefits of cloud computing provided by different companies. Enterprise can use different cloud environments, implementing one workload in Google Cloud or Microsoft Azure, another in Amazon Web Services, and maybe even more, for example, you can keep some data on a private cloud. The advent of cloud technologies revealed a lot of new business scenarios. Marketing solutions like Adobe Marketing Cloud or Hubspot, tools for education like SlideRocket (ClearSlide), ClearDATA in the healthcare sector: different areas have their own advantages of cloud computing innovations and new increasing opportunities for implementing in this deployment model. When it comes to dilemma 'cloud vs on-site', start from your organization's specific needs and requirements. Many companies combine both models, mixing public cloud services and on-site for various apps looking for the perfect combination of scalability, cost savings and efficiency managing. This allows them to get the best of both worlds, using the advantages of cloud computing in full. ( r: ) Disclaime The author works at FreshCode (Originally published ) here

Cloud Computing is Replete with Clouded Judgements #BreakTheChain

Authors:
(1) Qian Yang, Zhejiang University, Equal contribution. This work was conducted during Qian Yang’s internship at Alibaba Group;
(2) Jin Xu, Alibaba Group, Equal contribution;
(3) Wenrui Liu, Zhejiang University;
(4) Yunfei Chu, Alibaba Group;
(5) Xiaohuan Zhou, Alibaba Group;
(6) Yichong Leng, Alibaba Group;
(7) Yuanjun Lv, Alibaba Group;
(8) Zhou Zhao, Alibaba Group and Corresponding to Zhou Zhao (zhaozhou@zju.edu.cn);
(9) Yichong Leng, Zhejiang University
(10) Chang Zhou, Alibaba Group and Corresponding to Chang Zhou (ericzhou.zc@alibaba-inc.com);
(11) Jingren Zhou, Alibaba Group. Table of Links Abstract and 1. Introduction 2 Related Work 3 AIR-Bench and 3.1 Overview 3.2 Foundation Benchmark 3.3 Chat Benchmark 3.4 Evaluation Strategy 4 Experiments 4.1 Models 4.2 Main Results 4.3 Human Evaluation and 4.4 Ablation Study of Positional Bias 5 Conclusion and References A Detailed Results of Foundation Benchmark 4.3 Human Evaluation To evaluate the consistency between the evaluations of GPT-4 and human judgements, we design experiments for both the foundation and chat benchmarks. For the foundation benchmark, we instruct the testers to determine which option aligns closest with the hypothesis. We then compare the option selected by human testers with the option chosen by GPT-4 to assess the extent of agreement. For this consistency analysis, we employed Qwen-AudioChat as a representative model and randomly selected 400 questions from the benchmark. These questions were then evaluated by three native English speakers. Additionally, we also compared the performance of GPT-4 with GPT-3.5 Turbo. As depicted in Figure 4 (a), GPT-4 Turbo, serving as the evaluator, exhibited a high level of consistency at 98.2% with human judgements. Comparatively, GPT-3.5 Turbo had a slightly lower consistency rate of 96.4%. Regarding the chat benchmark, obtaining a numerical score on a scale of 1 to 10 directly from testers poses challenges. Therefore, we resort to a pairwise comparison of the models instead. Testers listen to audio and compare the performance of both models based on their usefulness, relevance, accuracy and comprehensiveness to the given question, indicating their preference as either “A is better”, “B is better”, or “Both are equal”. Subsequently, we convert the GPT-4 scores into the same preference-based rating as the human testers for any two models. We then assess the consistency between the two sets of results. For the chat benchmark, we conduct pairwise comparisons among Qwen-Audio-Chat, SALMONN, BLSP, and GPT4. We randomly select 200 questions and have them evaluated by three native English speakers. As depicted in Figure 4 (b), the pairwise preference consistency scored above 70%, demonstrating a high level of agreement. 4.4 Ablation Study of Positional Bias In our evaluation framework, we adopt a strategy of scoring twice by interchanging the positions of the hypothesis and reference and calculating the average of the two scores. This approach helps mitigate the bias that may arise from the positional placement. The outcomes of these two evaluations are presented in Figure 4 (c). We observe that the GPT4 evaluator exhibits a clear bias in scoring when the hypothesis is placed before the reference. This highlights the importance of conducting a second scoring to account for addressing this bias. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Qian Yang, Zhejiang University, Equal contribution. This work was conducted during Qian Yang’s internship at Alibaba Group; (2) Jin Xu, Alibaba Group, Equal contribution; (3) Wenrui Liu, Zhejiang University; (4) Yunfei Chu, Alibaba Group; (5) Xiaohuan Zhou, Alibaba Group; (6) Yichong Leng, Alibaba Group; (7) Yuanjun Lv, Alibaba Group; (8) Zhou Zhao, Alibaba Group and Corresponding to Zhou Zhao (zhaozhou@zju.edu.cn); (9) Yichong Leng, Zhejiang University (10) Chang Zhou, Alibaba Group and Corresponding to Chang Zhou (ericzhou.zc@alibaba-inc.com); (11) Jingren Zhou, Alibaba Group. Authors: Authors: (1) Qian Yang, Zhejiang University, Equal contribution. This work was conducted during Qian Yang’s internship at Alibaba Group; (2) Jin Xu, Alibaba Group, Equal contribution; (3) Wenrui Liu, Zhejiang University; (4) Yunfei Chu, Alibaba Group; (5) Xiaohuan Zhou, Alibaba Group; (6) Yichong Leng, Alibaba Group; (7) Yuanjun Lv, Alibaba Group; (8) Zhou Zhao, Alibaba Group and Corresponding to Zhou Zhao (zhaozhou@zju.edu.cn); (9) Yichong Leng, Zhejiang University (10) Chang Zhou, Alibaba Group and Corresponding to Chang Zhou (ericzhou.zc@alibaba-inc.com); (11) Jingren Zhou, Alibaba Group. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 AIR-Bench and 3.1 Overview 3 AIR-Bench and 3.1 Overview 3.2 Foundation Benchmark 3.2 Foundation Benchmark 3.3 Chat Benchmark 3.3 Chat Benchmark 3.4 Evaluation Strategy 3.4 Evaluation Strategy 4 Experiments 4.1 Models 4.1 Models 4.2 Main Results 4.2 Main Results 4.3 Human Evaluation and 4.4 Ablation Study of Positional Bias 4.3 Human Evaluation and 4.4 Ablation Study of Positional Bias 5 Conclusion and References 5 Conclusion and References A Detailed Results of Foundation Benchmark A Detailed Results of Foundation Benchmark 4.3 Human Evaluation To evaluate the consistency between the evaluations of GPT-4 and human judgements, we design experiments for both the foundation and chat benchmarks. For the foundation benchmark, we instruct the testers to determine which option aligns closest with the hypothesis. We then compare the option selected by human testers with the option chosen by GPT-4 to assess the extent of agreement. For this consistency analysis, we employed Qwen-AudioChat as a representative model and randomly selected 400 questions from the benchmark. These questions were then evaluated by three native English speakers. Additionally, we also compared the performance of GPT-4 with GPT-3.5 Turbo. As depicted in Figure 4 (a), GPT-4 Turbo, serving as the evaluator, exhibited a high level of consistency at 98.2% with human judgements. Comparatively, GPT-3.5 Turbo had a slightly lower consistency rate of 96.4%. Regarding the chat benchmark, obtaining a numerical score on a scale of 1 to 10 directly from testers poses challenges. Therefore, we resort to a pairwise comparison of the models instead. Testers listen to audio and compare the performance of both models based on their usefulness, relevance, accuracy and comprehensiveness to the given question, indicating their preference as either “A is better”, “B is better”, or “Both are equal”. Subsequently, we convert the GPT-4 scores into the same preference-based rating as the human testers for any two models. We then assess the consistency between the two sets of results. For the chat benchmark, we conduct pairwise comparisons among Qwen-Audio-Chat, SALMONN, BLSP, and GPT4. We randomly select 200 questions and have them evaluated by three native English speakers. As depicted in Figure 4 (b), the pairwise preference consistency scored above 70%, demonstrating a high level of agreement. 4.4 Ablation Study of Positional Bias In our evaluation framework, we adopt a strategy of scoring twice by interchanging the positions of the hypothesis and reference and calculating the average of the two scores. This approach helps mitigate the bias that may arise from the positional placement. The outcomes of these two evaluations are presented in Figure 4 (c). We observe that the GPT4 evaluator exhibits a clear bias in scoring when the hypothesis is placed before the reference. This highlights the importance of conducting a second scoring to account for addressing this bias. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Human Evaluation of Large Audio-Language Models

Authors:
(1) Qian Yang, Zhejiang University, Equal contribution. This work was conducted during Qian Yang’s internship at Alibaba Group;
(2) Jin Xu, Alibaba Group, Equal contribution;
(3) Wenrui Liu, Zhejiang University;
(4) Yunfei Chu, Alibaba Group;
(5) Xiaohuan Zhou, Alibaba Group;
(6) Yichong Leng, Alibaba Group;
(7) Yuanjun Lv, Alibaba Group;
(8) Zhou Zhao, Alibaba Group and Corresponding to Zhou Zhao (zhaozhou@zju.edu.cn);
(9) Yichong Leng, Zhejiang University
(10) Chang Zhou, Alibaba Group and Corresponding to Chang Zhou (ericzhou.zc@alibaba-inc.com);
(11) Jingren Zhou, Alibaba Group. Table of Links Abstract and 1. Introduction 2 Related Work 3 AIR-Bench and 3.1 Overview 3.2 Foundation Benchmark 3.3 Chat Benchmark 3.4 Evaluation Strategy 4 Experiments 4.1 Models 4.2 Main Results 4.3 Human Evaluation and 4.4 Ablation Study of Positional Bias 5 Conclusion and References A Detailed Results of Foundation Benchmark 3.4 Evaluation Strategy In this paper, we leverage a unified evaluation method, as shown in Fig. 3, by viewing both the single-choice question in the foundation benchmark, and the open-ended question in the chat benchmark, as the generation tasks for the purpose of better alignment with actual use case scenarios of LALMs. That is, given audio and questions, LALMs are required to directly generate the answers as hypotheses, rather than comparing the perplexity on the probability of different reference answers via teacher forcing. Automated and accurate evaluation of open-ended generation is a challenging problem. Traditional automatic metrics such as WER, ROUGE (Lin, 2004), METEOR (Banerjee and Lavie, 2005) have been shown low correlation with human judgements (Liu et al., 2023a). Recently, LLM-based evaluation, such as GPT-4, shows better human preference alignment (Zheng et al., 2023; Liu et al., 2023a). In this work, we adopt reference-based GPT-4 evaluators to judge the generation quality of LALMs in the audio domain. However, GPT-4 cannot be directly used as an evaluator since it cannot receive audio inputs. To address this limitation, we offer the GPT-4 model rich meta-information of audio to replace audio input. Subsequently, we present questions and employ GPT-4 to evaluate the hypotheses produced by LALMs. To ensure consistency and fairness for evaluation, each model’s answer is compared against the same reference answer for scoring. For the foundation benchmark, the reference answer is the golden choice, and we prompt the evaluator to determine whether the hypothesis is correct or not. For the chat benchmark, the reference answer is generated by GPT-4, and we prompt the evaluator to provide a score ranging from 1 to 10, based on the assessment of usefulness, relevance, accuracy, and comprehensiveness of the hypothesis. Note that for the chat benchmark, the role of the reference is not to serve as the ground truth answer, but rather as a reference for scoring by GPT-4, in order to stabilize its scoring. Additionally, to mitigate any potential position bias resulting from the order of hypothesis and reference, following Bai et al. (2023b), we perform a second scoring round by swapping their positions and then compute the average of the two scores. Unless otherwise specified, the GPT-4 evaluator is GPT-4 Turbo, the gpt-4- 0125-preview version [2]. This paper is available on arxiv under CC BY 4.0 DEED license. [2] https://platform.openai.com/docs/models/gpt-4-and-gpt4-turbo Authors: (1) Qian Yang, Zhejiang University, Equal contribution. This work was conducted during Qian Yang’s internship at Alibaba Group; (2) Jin Xu, Alibaba Group, Equal contribution; (3) Wenrui Liu, Zhejiang University; (4) Yunfei Chu, Alibaba Group; (5) Xiaohuan Zhou, Alibaba Group; (6) Yichong Leng, Alibaba Group; (7) Yuanjun Lv, Alibaba Group; (8) Zhou Zhao, Alibaba Group and Corresponding to Zhou Zhao (zhaozhou@zju.edu.cn); (9) Yichong Leng, Zhejiang University (10) Chang Zhou, Alibaba Group and Corresponding to Chang Zhou (ericzhou.zc@alibaba-inc.com); (11) Jingren Zhou, Alibaba Group. Authors: Authors: (1) Qian Yang, Zhejiang University, Equal contribution. This work was conducted during Qian Yang’s internship at Alibaba Group; (2) Jin Xu, Alibaba Group, Equal contribution; (3) Wenrui Liu, Zhejiang University; (4) Yunfei Chu, Alibaba Group; (5) Xiaohuan Zhou, Alibaba Group; (6) Yichong Leng, Alibaba Group; (7) Yuanjun Lv, Alibaba Group; (8) Zhou Zhao, Alibaba Group and Corresponding to Zhou Zhao (zhaozhou@zju.edu.cn); (9) Yichong Leng, Zhejiang University (10) Chang Zhou, Alibaba Group and Corresponding to Chang Zhou (ericzhou.zc@alibaba-inc.com); (11) Jingren Zhou, Alibaba Group. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 AIR-Bench and 3.1 Overview 3 AIR-Bench and 3.1 Overview 3.2 Foundation Benchmark 3.2 Foundation Benchmark 3.3 Chat Benchmark 3.3 Chat Benchmark 3.4 Evaluation Strategy 3.4 Evaluation Strategy 4 Experiments 4.1 Models 4.1 Models 4.2 Main Results 4.2 Main Results 4.3 Human Evaluation and 4.4 Ablation Study of Positional Bias 4.3 Human Evaluation and 4.4 Ablation Study of Positional Bias 5 Conclusion and References 5 Conclusion and References A Detailed Results of Foundation Benchmark A Detailed Results of Foundation Benchmark 3.4 Evaluation Strategy In this paper, we leverage a unified evaluation method, as shown in Fig. 3, by viewing both the single-choice question in the foundation benchmark, and the open-ended question in the chat benchmark, as the generation tasks for the purpose of better alignment with actual use case scenarios of LALMs. That is, given audio and questions, LALMs are required to directly generate the answers as hypotheses, rather than comparing the perplexity on the probability of different reference answers via teacher forcing. Automated and accurate evaluation of open-ended generation is a challenging problem. Traditional automatic metrics such as WER, ROUGE (Lin, 2004), METEOR (Banerjee and Lavie, 2005) have been shown low correlation with human judgements (Liu et al., 2023a). Recently, LLM-based evaluation, such as GPT-4, shows better human preference alignment (Zheng et al., 2023; Liu et al., 2023a). In this work, we adopt reference-based GPT-4 evaluators to judge the generation quality of LALMs in the audio domain. However, GPT-4 cannot be directly used as an evaluator since it cannot receive audio inputs. To address this limitation, we offer the GPT-4 model rich meta-information of audio to replace audio input. Subsequently, we present questions and employ GPT-4 to evaluate the hypotheses produced by LALMs. To ensure consistency and fairness for evaluation, each model’s answer is compared against the same reference answer for scoring. For the foundation benchmark, the reference answer is the golden choice, and we prompt the evaluator to determine whether the hypothesis is correct or not. For the chat benchmark, the reference answer is generated by GPT-4, and we prompt the evaluator to provide a score ranging from 1 to 10, based on the assessment of usefulness, relevance, accuracy, and comprehensiveness of the hypothesis. Note that for the chat benchmark, the role of the reference is not to serve as the ground truth answer, but rather as a reference for scoring by GPT-4, in order to stabilize its scoring. Additionally, to mitigate any potential position bias resulting from the order of hypothesis and reference, following Bai et al. (2023b), we perform a second scoring round by swapping their positions and then compute the average of the two scores. Unless otherwise specified, the GPT-4 evaluator is GPT-4 Turbo, the gpt-4- 0125-preview version [2]. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv [2] https://platform.openai.com/docs/models/gpt-4-and-gpt4-turbo

Unified Evaluation Method for LALMs Using GPT-4 in Audio Tasks

Authors:
(1) Martin Kleppmann, University of Cambridge, Cambridge, UK (martin.kleppmann@cst.cam.ac.uk);
(2) Paul Frazee, Bluesky Social PBC United States;
(3) Jake Gold, Bluesky Social PBC United States;
(4) Jay Graber, Bluesky Social PBC United States;
(5) Daniel Holmgren, Bluesky Social PBC United States;
(6) Devin Ivy, Bluesky Social PBC United States;
(7) Jeromy Johnson, Bluesky Social PBC United States;
(8) Bryan Newbold, Bluesky Social PBC United States;
(9) Jaz Volpert, Bluesky Social PBC United States. Table of Links Abstract and 1 Introduction 2 The Bluesky Social App 2.1 Moderation Features 2.2 User Handles 2.3 Custom Feeds and Algorithmic Choice 3 The at Protocol Architecture 3.1 User Data Repositories 3.2 Personal Data Servers (PDS) 3.3 Indexing Infrastructure 3.4 Labelers and Feed Generators 3.5 User Identity 4 Related Work 5 Conclusions, Acknowledgments, and References 3.4 Labelers and Feed Generators Relay and App View aim to provide a mostly “unopinionated” service: they compute indexes over repositories in a neutral way, without attempting to rank or classify content. However, a good user experience also requires “opinionated” judgements for the purposes of content filtering (e.g. detecting sexually explicit images or spam) and curation (e.g. selecting posts on a particular topic). The AT Protocol seperates out the “opinionated” aspects of the system into separate services: labelers and feed generators. These services typically take the firehose as their input. Labelers produce a stream of judgements about content (e.g. “this post is spam”), whereas feed generators return a list of post IDs they have selected for inclusion in a custom feed, as described in Section 2.3. Users can choose in their client app which feeds and which labelers they want to use. The output of labelers is consumed by App Views or PDSes in order to apply content filtering [12]. For a feed generator, an App View expands the post IDs into full posts before sending them to the client app of users who have subscribed to that feed. Having labeler and feed generator services that are separate from App Views has several advantages: • Anyone can run such services, which enables a pluralistic ecosystem in which different parties may make different judgements about the same piece of content. Users, as well as the operators of App Views and PDSes, can decide whose judgements they want to trust, and it is easy for them to switch to alternative labeling and feed generation services if their current providers fail to meet their expectations. • It becomes easier to set up alternative App View providers: since any App View can consume the publicly available output from labelers and feed generators, there is less pressure for each App View to develop its own content filtering infrastructure. Having alternative App Views is important for a healthy, decentralized marketplace. Feed generators can be implemented in code using our starter kit [10], or created with a third-party service such as Skyfeed [13]. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Martin Kleppmann, University of Cambridge, Cambridge, UK (martin.kleppmann@cst.cam.ac.uk); (2) Paul Frazee, Bluesky Social PBC United States; (3) Jake Gold, Bluesky Social PBC United States; (4) Jay Graber, Bluesky Social PBC United States; (5) Daniel Holmgren, Bluesky Social PBC United States; (6) Devin Ivy, Bluesky Social PBC United States; (7) Jeromy Johnson, Bluesky Social PBC United States; (8) Bryan Newbold, Bluesky Social PBC United States; (9) Jaz Volpert, Bluesky Social PBC United States. Authors: Authors: (1) Martin Kleppmann, University of Cambridge, Cambridge, UK (martin.kleppmann@cst.cam.ac.uk); (2) Paul Frazee, Bluesky Social PBC United States; (3) Jake Gold, Bluesky Social PBC United States; (4) Jay Graber, Bluesky Social PBC United States; (5) Daniel Holmgren, Bluesky Social PBC United States; (6) Devin Ivy, Bluesky Social PBC United States; (7) Jeromy Johnson, Bluesky Social PBC United States; (8) Bryan Newbold, Bluesky Social PBC United States; (9) Jaz Volpert, Bluesky Social PBC United States. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 The Bluesky Social App 2 The Bluesky Social App 2.1 Moderation Features 2.1 Moderation Features 2.2 User Handles 2.2 User Handles 2.3 Custom Feeds and Algorithmic Choice 2.3 Custom Feeds and Algorithmic Choice 3 The at Protocol Architecture 3 The at Protocol Architecture 3.1 User Data Repositories 3.1 User Data Repositories 3.2 Personal Data Servers (PDS) 3.2 Personal Data Servers (PDS) 3.3 Indexing Infrastructure 3.3 Indexing Infrastructure 3.4 Labelers and Feed Generators 3.4 Labelers and Feed Generators 3.5 User Identity 3.5 User Identity 4 Related Work 4 Related Work 5 Conclusions, Acknowledgments, and References 5 Conclusions, Acknowledgments, and References 3.4 Labelers and Feed Generators Relay and App View aim to provide a mostly “unopinionated” service: they compute indexes over repositories in a neutral way, without attempting to rank or classify content. However, a good user experience also requires “opinionated” judgements for the purposes of content filtering (e.g. detecting sexually explicit images or spam) and curation (e.g. selecting posts on a particular topic). The AT Protocol seperates out the “opinionated” aspects of the system into separate services: labelers and feed generators. These services typically take the firehose as their input. Labelers produce a stream of judgements about content (e.g. “this post is spam”), whereas feed generators return a list of post IDs they have selected for inclusion in a custom feed, as described in Section 2.3. Users can choose in their client app which feeds and which labelers they want to use. The output of labelers is consumed by App Views or PDSes in order to apply content filtering [12]. For a feed generator, an App View expands the post IDs into full posts before sending them to the client app of users who have subscribed to that feed. Having labeler and feed generator services that are separate from App Views has several advantages: • Anyone can run such services, which enables a pluralistic ecosystem in which different parties may make different judgements about the same piece of content. Users, as well as the operators of App Views and PDSes, can decide whose judgements they want to trust, and it is easy for them to switch to alternative labeling and feed generation services if their current providers fail to meet their expectations. • It becomes easier to set up alternative App View providers: since any App View can consume the publicly available output from labelers and feed generators, there is less pressure for each App View to develop its own content filtering infrastructure. Having alternative App Views is important for a healthy, decentralized marketplace. Feed generators can be implemented in code using our starter kit [10], or created with a third-party service such as Skyfeed [13]. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Stories, News & Mentions about JUDGEME

Judgeme’s Stories on HackerNoon