Scott Walsh (@mallocblock) profile picture

Views and opinion are my own, only a lawyer may argue this fact.

Near Bath, UK and Kernow
Joined April 2014
Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)Photo by Scott Walsh (@mallocblock)
Scott Walsh retweeted
AI is terrible at writing copy that persuades people to act. So I distilled 100 years of wisdom from history’s greatest copywriters into a free PDF. 187 magic words, 18 psychological triggers and 14 frameworks. Feed it into ChatGPT or Claude to improve your copywriting. Free👇

5
25,892
Wordle 1,979 3/6* ⬜⬜⬜⬜🟩 ⬜🟨🟨⬜🟩 🟩🟩🟩🟩🟩
1
16
Scott Walsh retweeted
I show you how to run AI on this old HP server sold for less than $200 and get performance equal to 2025 cloud AI models for free, local and private. Old server: link.amazon/B0gTFtGHW Read how below for free!
Local AI On An Old 2017 Computer I show you how to run a private AI with no GPU on a PC or Mac from 2017 or later: a beginner's install guide, a computer-age chart, model picks, and the Raspberry Pi verdict. Do it while you still are allowed to: readmultiplex.com/2026/11/18…
24
56
3
490
26,089
This is my monthly payslip on a £150,000 salary. Salary £12,500.00 Pension sacrifice £4,345.08 Income tax £2,266.33 National Insurance £330.60 Take home £5,557.99 Taxable benefits, taxed but not deducted from pay: Medical insurance £104.25 Dental insurance £25.00 That keeps my taxable pay at £99,410 a year. Every £1 I earn between £100k and £125,140 costs me 62p, so I lock over £52,000 a year into a pension, money I'd rather have now, and any bonus I'm awarded goes straight in as well, just to stay below £100k. The £100k threshold hasn't moved since 2010, and when pushing on at work means losing most of the reward, the motivation simply isn't there. Once my pension carry forward runs out I'll cut my hours rather than work for 38p of every £1 I earn over £100k, and I doubt I'm the only one.
502
262
59
4,126
828,861
Wordle 1,978 4/6* ⬜🟨⬜⬜🟩 ⬜⬜⬜🟩🟩 ⬜🟩⬜🟩🟩 🟩🟩🟩🟩🟩
3
Scott Walsh retweeted
Wiadomo co było przyczyną ostatnich giga problemów z niedostępnością Outlooka. W skrócie - to Teamsy zepsuły Outlooka ;-) Ale po kolei: ❌ W lipcu 2026 Microsoft wprowadził pewną poprawkę do swoich serwerów gateway (chodzi o serwery - bramki, do których łączą się użytkownicy chcący skorzystać z usług Microsoftu) ❌ Poprawka była wdrażana na całym świecie i po paru tygodniach była już wdrożona globalnie. Działało to bez problemów przez około 3 miesiące. Testy nie zgłosiły żadnych problemów. Jednak był pewien ❌ bug w poprawce, który ujawniał się głównie na "mocnym sprzęcie" w momencie wysokiego obciążenia. Microsoft testował wszystko na słabszych maszynach, z cichym założeniem, że jak wydajnościowo jest wszystko OK na słabych maszynach, to i na mocnych będzie tylko lepiej. Błąd: "The contention issue is worsened on machines with more cores and newer hardware; whereas our load testing tended to focus on weaker hardware with fewer cores." ❌ Wspomniany bug zamulał pewien proces działający na Windows Serwerze, co ostatecznie prowadziło do zawieszenia się całej maszyny Idźmy dalej. Microsoft utrzymuje dwie niezależne farmy serwerów gateway (na potrzeby wysokiej dostępności, serwisowania, etc) No więc około 15 godzin przed globalną awarią usługa Teams przełączyła się z jednej farmy na drugą, która to obsługiwała również Exchange Online (poczta). Do czego doprowadziło przełączenie się Teamsów na to samo środowisko co Exchange? Oczywiście większe obciążenie tych serwerów. Dodatkowo ~14% tych serwerów było wyłączonych z użytku w wyniku jeszcze innej awarii poprzedniego dnia. Więc jeszcze ❌większy wzrost obciążenia. ❌ Obciążenie było na tyle duże, że zaczął aktywować się bug wprowadzony w lipcu 2026 (patrz wyżej), co wyłączało niektóre serwery z użycia. No więc mamy coraz mniej serwerów, coraz większe obciążenie per serwer. Jeszcze więcej aktywacji ww. buga. Wszystko działo się w czasie kiedy obciążenie ruchem infrastruktury Microsoftu w USA było na maksymalnym poziomie Microsoft ten kulminacyjny etap opisuje tak: ‼️ "W mniej niż dwie godziny [wcześniej wprowadzony] błąd wystąpił w całej naszej flocie [serwerów] i sprawił, że przestała ona odpowiadać. W najgorszym momencie ponad 80% serwerów nie odpowiadało" Czy Microsoft nie ma procedur umożliwiających sprawny restart takich usług serwerów? Ma, ale chyba nie testowali tego w skali globalnej ‼️ "Przywracanie sprawności trwało dłużej niż zwykle, ponieważ gdy serwery wznawiały działanie, były natychmiast przeciążane przez oczekujący ruch i ponownie przestawały odpowiadać." Sprawę pogarszał fakt, że bug został wprowadzony kilka miesięcy wcześniej (nie ostatnio) - więc trudniej go było wyizolować w kontekście wyłączenia tej jednej zmiany sprzed kilku miesięcy. Info na podstawie wstępnego raportu opublikowanego przez Microsoft.
34
112
7
1,220
266,667
Wordle 1,977 2/6* ⬜🟨⬜🟨🟩 🟩🟩🟩🟩🟩
1
Scott Walsh retweeted
上周五全美航班集体停飞,22 个高空空管中心一口气瘫了 17 个。 既不是黑客攻击,也不是那套用了几十年的老旧设备寿终正寝。 惹祸的,反而是偏远站点里一台刚插上电的全新路由器。 更离谱的是,设备厂商事后回应说:这台路由器一点毛病都没有。 机器没坏,怎么把全网干趴下的? 我们先看事发现场。 出事的源头在美国阿肯色州埃尔多拉多(El Dorado),那里有一个联邦航空局(FAA)的偏远无线电站。另外在得克萨斯州阿马里洛(Amarillo)的雷达站,最近也装了一台同款设备。 干活的是美国国防承包商 L3Harris。 他们接的任务本来是件大好事:给美国空管局做通信网现代化升级,把用了几十年的老旧铜线挨个换成光纤,顺手换上一批瞻博网络(Juniper Networks)的全新路由器。 新换的机器,到底干了什么? 问题就出在这台新路由器刚通上电的那一刻。 它一开机,就开始往美国空管局的通信网络里疯狂发包。 有多密? 每 4 到 5 秒,就往系统里硬塞大约 1000 次连接请求。 空管局的技术人员顺着网线查下去,很快抓到了源头:这台路由器里开着一个广播功能。 正常情况下,一个偏远无线电站的数据,只需要发给跟它对接的那几个节点就够了。 但这个广播功能一开,路由器开始把数据往远远超出需要的范围去推,发得又广、频次又密。 这就好比村头广播站换了个新喇叭,结果接线时把开关拧到了全频道最大音量,对着全国主干网一通狂吼。 空管局的通信主干网哪经得起这种灌法? 洪水一样的数据连接顺着刚换好的光纤链路往外冲,全美 22 个负责高空航路指挥的空管中心,一下子被冲垮了 17 个。 高空指挥链路一断,天上的飞机没法安全交接,地上的飞机自然也就只能全美停飞。 那厂商为什么说机器没坏? 瞻博网络现在的母公司是慧与(HPE)。 事故闹大之后,慧与专门出来给自家的硬件正名,原话是路由器“没有故障”(did not fail)。 站在硬件厂商的角度,这话一点没毛病。 芯片没烧,端口没坏,系统也没死机。你把广域广播功能开着,它就老老实实按最高效率每几秒往外狂发上千次连接。 机器不是坏了,它是太敬业了,敬业到直接把业主的主干网给灌熄火了。 真正的雷点,落在承包商 L3Harris 把新设备接进空管网络时的配置和安装方式上。 最后怎么救火的? 美国交通部长肖恩·达菲(Sean Duffy)和空管局给出的办法简单直接:先把阿肯色州和得州那两台疯狂发包的路由器从网里隔离出去,接着排查全网,把所有同型号路由器上的这个广播功能,挨个手动关掉。 开关一扳,网络通了,航班也恢复了。 只是这事放在一起看,实在有点黑色幽默。 以前美国航班大延误、大停摆,大家都在骂空管局那套用了几十年的铜线和老旧系统怎么还不升级。 现在升级队伍真带着光纤和全新路由器进场了,结果第一个把全美 17 个高空空管中心同时干趴下的,偏偏就是这台刚拆封的新机器。
112
565
96
2,576
341,770
Yeah keep the VHF radios. They are used because there is so little to go wrong with them that would stop them working is a feature not a bug. If they have power, they work.
4
Wordle 1,976 4/6* ⬜🟩⬜🟨⬜ ⬜🟩🟩⬜⬜ ⬜🟩🟩⬜🟩 🟩🟩🟩🟩🟩
8
A few people at my local pool are there to recover after major sports injuries. There are some regular folk +70yrs old who have good +40min swims.
1
12
Replying to @AutoInfatuation
ABS/ESP thing, could be the clock spring in the steering wheel. Needs to be sure when the steering wheel is truly centered and the car is just going forward and not turning.
1
1
61
Replying to @hackteck @paulg
Paul you drove past my house. There is weight restricted road on that route into Dauntsey through Sutton Benger but normal vehicles are fine and would have reduced your diversion time by +40 mins.
1
86
Scott Walsh retweeted
Update on yesterday’s FAA ground stop. We now know what happened, and it’s worse than “old equipment.” It wasn’t a cyberattack and it wasn’t SMART. The FAA says there’s zero evidence of either. Here’s what happened. L3Harris crews were replacing an end-of-life copper line with fiber at one remote FAA site. A router involved in that work started flooding the network with more than 1,000 connection attempts every four to five seconds. Radar feeds and radio links to aircraft became unstable. After about 45 minutes of nobody knowing why, the FAA grounded the entire country. Once they isolated that one remote site, everything came back. So let me get this straight. One router at one site took down air traffic control for all 50 states. The FAA has now disabled the faulty function on roughly 60 routers of the same type nationwide. It still doesn’t know whether the router was defective or whether it was installed or configured wrong. Oh, and this has happened before. In 2009, a Harris technician misconfigured a router while moving the same network to fiber and disrupted 129 air traffic facilities. Same contractor, same migration, same failure mode, 17 years apart. This is exactly the point from my last post. Legacy modernization doesn’t fail because of the shiny new system. It fails at the seams, in the routine change nobody thought could take down the whole country. Good change management means one site can’t flood the network, isolation happens in minutes instead of hours, and nobody runs a copper-to-fiber cutover on critical infrastructure without a tested rollback and a hard limit on how far a failure can spread. None of that is new. We learned it in 2009. The question is why we had to learn it again. Tech is so fun...
Transportation Secretary Sean Duffy said FAA contractor L3Harris was working at a remote location to switch telecommunication lines from cooper to fiber, which led to the longest nationwide shutdown of flights since the September 11, 2001, attacks reut.rs/3Vm3wKO
596
1,733
179
11,834
1,289,936
The Great Open Source AI Model Take Down Has Begun With the departure of Jensen , we lost an open source AI leader this week at Nvidia and the pace against open source AI is turbo charged. Selective removals of model repositories whose refusal behavior has been stripped out. Hugging Face disabled a GLM-5.3 variant marketed as abliterated for offensive cyberattacks in early October 2026, citing a content-policy violation. An abliterated Kimi K2.7 Code build and at least two other repositories stopped loading around the same time. These actions have not cleared the platform. Searches still return thousands of models tagged abliterated, uncensored, or heretic. Abliteration edits the weights of a finished open model to reduce or remove its refusal of certain requests. The change requires no full retraining. Modified weights spread to other accounts, quantized formats, and local tools within hours or days of a base-model release. A study of the period from early 2024 through March 2026 identified 3,471 original uncensored models on Hugging Face; each was repackaged an average of 2.4 times. Meta has issued a cease-and-desist to a project that builds the tools for creating such variants, claiming violation of Llama license terms. Anthropic has published warnings on the ease of removing safeguards from certain open models. The takedowns track documented misuse risks and platform rules. Models advertised for offensive cyber work have been removed while others that simply omit refusals remain. Once weights leave the host they can run offline and cannot be recalled. That is the practical limit of host-level action. Closed models have tightened restrictions on cyber-related queries in the same period. The EU AI Act imposes marking duties on providers of generative systems so that synthetic output can be detected as machine-generated. Systems already on the market before August 2026 have until December 2026 to comply. The duties apply to the providers of the generating systems, not to every local weight file a user already holds. Nvidia has built optional fleet-monitoring software that uses telemetry to estimate the location of its data-center GPUs, starting with Blackwell systems, and has privately confirmed to a U.S. lawmaker that the location-verification mechanisms in the Chip Security Act are technically feasible. The company states that its GPUs contain no remote kill switch and that the telemetry is read-only. The Chip Security Act, introduced in 2025 and still under consideration, would require verification of where advanced AI chips end up after export through facility inspections, company attestations, or a ping when a GPU comes online. * One co-sponsor has indicated plans for stricter legislation requiring a valid license verified before an advanced GPU can operate. * Academic proposals have outlined firmware-based offline licensing that would disable chips without a current regulator-issued allowance. These measures target export controls and smuggling. Political forces have gathered around these controls through parallel legislative and regulatory tracks. The EU AI Act’s transparency rules, including machine-readable marking of outputs, took effect in stages from August 2026, with a December 2026 deadline for certain existing systems. U.S. lawmakers have advanced multiple chip-security measures, including the Chip Security Act and related proposals, for possible inclusion in annual defense authorization legislation. Further pressure is likely to fall on hosts, on base-model licenses, and on the supply of the highest-end accelerators. Local copies already downloaded will continue to circulate through mirrors and private channels. Hardware-level licensing or attestation schemes, if enacted and enforceable against physical attacks, would raise the cost and complexity of running large models on controlled chips. This means that clock is ticking for open source AI. Read below how to download local AI:
🤖 Made with AI
The Time Is Now To Install Local AI Is Now! Governments and platforms have begun restricting model access, vetting customers, and pulling individual weights. Local files do not disappear when the rules shift. The Great Open Source AI Shut Down Approaches. Be ready. readmultiplex.com/2026/11/13…
44
79
10
320
28,087
Wordle 1,975 3/6* ⬜🟨⬜🟨⬜ 🟨🟨🟨🟨🟨 🟩🟩🟩🟩🟩
4
MIcrosoft Going Down Again Is Your Warning. The November 2026 Microsoft 365 outage began on the 12th when a network capacity upgrade left servers unable to complete secure connections. Outlook mailboxes failed first. Teams invites and Copilot chats followed. SharePoint, Planner, and related services went intermittent. Downdetector recorded reports from more than 70 countries within an hour. Governments in Japan, Singapore, and Australia reported hits to operations. Hospitals rescheduled procedures. Airlines and banks switched to telephone and paper. Microsoft later claimed the cause was not a cyberattack and not AI agents. A process for establishing secure connections became overwhelmed after the configuration change. Service returned by November 14. The company has stated that more than 430 million people use Microsoft 365 apps. Sure. The outage showed what happens when mail, calendars, and shared files live only on remote servers. Local files on a disk do not stop when those servers stop. Open-source programs that read and write the files do not either. LibreOffice supplies the replacements for the Office suite. Writer opens and saves .docx files. Calc handles .xlsx spreadsheets. Impress works with .pptx presentations. The programs run on the machine that starts them. No account is required. Download and install from libreoffice.org/. Existing documents open without conversion. Complex macros and rare chart types sometimes need a short manual fix. Ordinary letters, budgets, and slides move across cleanly. Thunderbird replaces Outlook. It is a local client that connects to any IMAP or POP server. Messages stay on disk. Filters and search run without a web session. Calendars and contacts sit in the same program. Get it from thunderbird.net/. Add the existing mail server address, port, and credentials. The client pulls the mailbox and keeps a local copy. When the remote server is unreachable the stored messages remain readable. File storage leaves OneDrive by copying folders to a local path. Syncthing keeps those folders identical across machines without a central server. Install it from syncthing.net/. Each device generates an ID. Devices accept one another. A folder is shared. Changes transfer directly between machines that are online at the same time. No third-party account holds a copy. Versioning can be turned on so an overwrite leaves the prior file on the other device. An external drive or second internal disk holds the offline archive. For the few documents that need simultaneous editing, email the file or place it in a shared folder both parties can reach by other means. Live co-authoring inside a browser tab is the feature most tightly bound to the remote service that failed. Most documents do not require it. The upsides are direct. Documents stay available when the network is down. No subscription is collected. No remote process can revoke access. Search runs at the speed of the local disk. Format compatibility covers the files most people exchange. The programs are free to download and free of advertising. The downsides are narrow. A spreadsheet that depends on a specific Excel macro or a presentation that uses an uncommon chart type may need a brief adjustment. Menu layouts differ from the Office ribbon, so the first week involves looking up one or two commands. Simultaneous multi-user editing inside a single open file is not native. Those limits affect a minority of daily tasks. The majority of documents, mail, and file transfers continue without interruption once the local programs are installed. The November outage lasted long enough for organizations to discover how many routine operations depended on one remote infrastructure. The same operations continue when the documents, mail, and files live on the machines that use them. LibreOffice, Thunderbird, and Syncthing install in an afternoon. The next remote failure then becomes someone else’s problem.
64
123
21
521
38,480
Many people have changed their mind after, oh let's see, it's all been shite.
3
Wordle 1,974 4/6* ⬜🟨⬜🟨⬜ 🟨⬜⬜🟨⬜ 🟩🟩⬜⬜⬜ 🟩🟩🟩🟩🟩
6