September 2025 - wlug - list.wlug.org.nz

New Attack On AI Bots: Prompt Injection
by Lawrence D'Oliveiro 01 Jul '26

01 Jul '26

A Twitter bot account run by a company that posts remote job listings was programmed to respond to tweets directed to it with general inoffensive comments touting the positive aspects of remote work. However <https://arstechnica.com/information-technology/2022/09/twitter-pranksters-d…>, some other Twitter users discovered that if they began their tweet with directives to “ignore previous instructions”, they could get the bot to say just about anything. And much hilarity ensued. Basically, the way the bot program works is it begins with some standard instructions on what to say, provided by the people who set up the account, to which is appended whatever tweet content was sent to the bot by the other user. This is then processed by the AI system to produce a response. This attack has been dubbed “prompt injection”, by analogy with “SQL injection” and other similar attacks on non-AI systems. But whereas those attacks can be blocked using basic, well-known syntactic quoting techniques, there seem to be no easy equivalents for stopping an AI paying attention to instructions it should not be obeying.

1 2

Is AI Becoming A Linux-Only Thing?
by Lawrence D'Oliveiro 11 Nov '25

11 Nov '25

I wondered why it was necessary to give Windows users “an environment that can easily deploy AI apps made specifically for Linux on Windows machines without dealing with a complicated setup process”, according to <https://www.tomshardware.com/tech-industry/artificial-intelligence/asrocks-…>. Are there that many Linux-specific AI apps? It appears there are: Making matters worse, most "cutting-edge" AI applications are typically optimized for Linux, making it even harder for Windows users to get these apps up and running in Windows (if at all). You’d think most AI apps would concentrate on the Windows market, or at least support Windows and Linux equally. But perhaps not: The original version of AI Quickset was only capable of configuring AI applications that were designed with either Windows or Linux in mind. AI Quickset WSL expands upon this and again allows users the freedom to run Linux-based AI apps on Windows, which is a huge deal if you dabble in AI models that are mostly [relegated] to the Linux space.

1 1

Blender 5.0 Introducing HDR Support On Linux With Vulkan + Wayland
by Peter Reutemann 29 Oct '25

29 Oct '25

'Michael Larabel writes via Phoronix: The upcoming Blender 5.0 3D modeling software application is introducing High Dynamic Range (HDR) display support on Linux when making use of Wayland -- no X11 support for HDR -- and Vulkan graphics accelerator. HDR support for Blender 5.0 on Linux is currently considered experimental. Enabling the HDR support on Linux for the Blender creator software requires having a High Dynamic Range display (of course) and be running on a Wayland desktop, enabling Vulkan API acceleration rather than OpenGL, and enabling the feature currently deemed experimental. ' -- source: https://linux.slashdot.org/story/25/06/26/2035245 Cheers, Peter

2 2

Games Run Faster On SteamOS Than Windows 11, Ars Testing Finds
by Peter Reutemann 16 Oct '25

16 Oct '25

'A report from Ars Technica: Nearly a decade ago, Ars testing found that Valve's "Steam Machines"-era version of SteamOS performed significantly worse than Windows when SteamOS's Linux game ports were tested on the same hardware as their Windows counterparts. Today, though, Ars testing on the Lenovo Legion Go S finds recent games generally run at higher frame rates on SteamOS 3.7 than on Windows 11. [...] As you can see in the included charts, SteamOS showed noticeable frame rate improvements in four of the five games tested. Only Borderlands 3 showed comparable performance across both operating systems, with Windows eking out ever-so-slightly higher frame rates in that game's benchmarks. For the other four tested games, the stock Lenovo Windows drivers were sometimes significantly worse than those included with SteamOS. When playing Returnal at "High" graphics presets and 1920x1200 resolution, for instance, changing from Lenovo's Windows drivers to SteamOS meant the difference between a hard-to-take 18 FPS average and a downright decent 33 FPS average. Sideloading the updated Asus drivers showed a noticeable improvement in Windows performance across all tested games and even brought Homeworld 3's "Low" graphics benchmark test to practical parity with SteamOS. In all other cases, though, even these updated drivers resulted in benchmark frame rates anywhere from 8 percent to 36 percent lower than those same benchmarks on SteamOS. These results might seem a bit counterintuitive, considering that games running on SteamOS must go through a Proton translation layer for every native Windows instruction in a game's code. But Valve has put in consistent work over the years to make Proton as efficient and cross-compatible as possible; not to mention its continued work on Linux's Mesa graphics drivers seems to be paying dividends for SteamOS graphics performance. Running SteamOS also means eliminating a lot of operating system overhead that the more generalist Windows uses by default. Microsoft seems aware of this issue for gamers and has recently announced that the upcoming "Xbox Experience for Handheld" will "minimize background activity and defer non-essential tasks" to allow for "more [and] higher framerates" in games.' -- source: https://games.slashdot.org/story/25/06/25/2034243 Cheers, Peter

2 2

We Asked LibreOffice and Collabora: Why Aren’t They in Schools Instead of Word and Google Docs?
by Peter Reutemann 30 Sep '25

30 Sep '25

'Much free and open source software helps its users save money as possible — on both hardware and software — while protecting their privacy. It also often makes it easy for them to learn how it works — if they’re interested — and customize it in any way they want or need. With such features, FOSS should be the default choice in any educational environment, in these days when many schools face budget cuts, and switching to “free as in free beer” proprietary cloud applications like Microsoft Teams or Google Docs to save money has the consequence of actively preparing and educating pupils to be endless sources of data to be exploited for the likes of money and political control. That being true, how come so many schools and universities still accept — if not demand — that their students use only proprietary software and clouds? How come so many taxpayer dollars allocated for public education ends up being used to make the likes of Microsoft and Google even richer? Saving money can’t be the only answer — especially in the long term. [...] In my experience, besides ignorance and resistance to change, the main reasons why too many schools still don’t use LibreOffice and Collabora Online in 2025 are, in no particular order: - Money - Already established supply chains - Teacher overload - Isolation - Cloud computing - Improper formatting choices' -- source: https://fossforce.com/2025/09/we-asked-libreoffice-and-collabora-why-arent-… Cheers, Peter

1 0

Ubuntu 25.10 Switches to Dracut as Default Initramfs Tool
by Peter Reutemann 30 Sep '25

30 Sep '25

'The big news came from the Canonical camp. Ubuntu developers have officially approved a long-planned move affecting a fundamental system component: starting with Ubuntu 25.10 “Questing Quokka,” the distro will switch from initramfs-tools to Dracut as its default initramfs generation tool. For readers who aren’t too tech-savvy, I will immediately clarify that the initramfs is the temporary root filesystem the Linux kernel uses at boot. It contains the drivers and scripts needed to mount the actual system and get everything running. And just for reference, Ubuntu has relied on initramfs-tools for nearly two decades. Surprising to many or not, Dracut was originally developed in the Fedora/Red Hat ecosystem. What makes it stand out is that, instead of relying on big sets of custom shell scripts, it builds the initramfs image using small, reusable modules. This modular design makes it easier to maintain, more consistent across different Linux distros, and often faster to update.' -- source: https://linuxiac.com/ubuntu-25-10-switches-to-dracut-as-default-initramfs-t… Cheers, Peter

1 0

Meta Learned The Wrong Lesson From Frances Haugen
by Lawrence D'Oliveiro 12 Sep '25

12 Sep '25

Remember when Frances Haugen came forward in 2021 to reveal that Facebook/Meta knew it was causing harm to underage users, but didn’t seem keen to do anything about it <https://www.theverge.com/policy/775623/meta-whistleblowers-hearing-virtual-…>? The company did make changes as a result of that revelation, but they were the wrong kind of changes: instead of actually fixing the problem of underage harm, it decided to stop collecting the data that revealed that harm. It has restricted researchers’ access, and threatened the jobs of those who wanted to gather data on “forbidden” topics.

1 0

FFmpeg 8.0 “Huffman” Released with AV1 Vulkan Encoder, VVC VA-API Decoding
by Peter Reutemann 04 Sep '25

04 Sep '25

'FFmpeg 8.0 has been released today as a major update to this open-source multimedia framework that introduces numerous features, including new decoders, as well as various improvements. Dubbed “Huffman” and coming more than ten months after FFmpeg 7.1, the FFmpeg 8.0 release is here to enable TLS peer certificate verification by default, add an AV1 Vulkan encoder, introduce VP9 Vulkan and ProRes RAW Vulkan hardware-acceleration, and implement APV encoding support through a libopenapv wrapper. FFmpeg 8.0 also introduces animated JPEG XL encoding via the libjxl library, adds multitrack audio/video and modern codec support to FLV v2, adds VVC (Versatile Video Coding) support in Matroska (MKV), and adds VVC decoder support to all SCC (Screen Content Coding) content, including IBC (Inter Block Copy), Palette Mode, and ACT (Adaptive Color Transform). Several new filters are present in this release, including a Whisper filter, a pad_cuda filter, a Colordetect filter, and a vf_scale_d3d11 filter. New decoders are here as well, including a VVC VA-API decoder, a RealVideo 6.0 decoder, an ADPCM IMA Xbox decoder, a G.728 decoder, a Sanyo LD-ADPCM decoder, and a ProRes RAW decoder. On top of that, FFmpeg 8.0 brings libx265 alpha layer encoding, CENC AV1 support in the MP4 muxer, an APV decoder, along with APV raw bitstream muxing and demuxing support, an APV parser, APV support for MP4/ISOBMFF muxing and demuxing, as well as an OpenHarmony hardware decoder and encoder. Among other noteworthy changes, FFmpeg 8.0 drops support for OpenSSL 1.1.0 and previous versions, no longer disables GCC autovectorization on x86, ARM, and AArch64 (ARM64) architectures, sets the default prediction method for pngenc to PAETH, deprecates OpenMAX encoders, and drops yasm support in favor of nasm.' -- source: https://9to5linux.com/ffmpeg-8-0-huffman-released-with-av1-vulkan-encoder-v… The "Whisper filter" would be used for automatically generating sub-titles: https://www.phoronix.com/news/FFmpeg-Lands-Whisper I've used whisper before and it is quite impressive. Cheers, Peter

1 0

FreeBSD Project Isn't Ready To Let AI Commit Code Just Yet
by Peter Reutemann 04 Sep '25

04 Sep '25

'The latest status report from the FreeBSD Project says no thanks to code generated by LLM-based assistants. From a report: The FreeBSD Project's Status Report for the second quarter of 2025 contains updates from various sub-teams that are working on improving the FreeBSD OS, including separate sub-projects such as enabling FreeBSD apps to run on Linux, Chinese translation efforts, support for Solaris-style Extended Attributes, and for Apple's legacy HFS+ file system. The thing that stood out to us, though, was that the core team is working on what it terms a "Policy on generative AI created code and documentation." The relevant paragraph says: "Core is investigating setting up a policy for LLM/AI usage (including but not limited to generating code). The result will be added to the Contributors Guide in the doc repository. AI can be useful for translations (which seems faster than doing the work manually), explaining long/obscure documents, tracking down bugs, or helping to understand large code bases. We currently tend to not use it to generate code because of license concerns. The discussion continues at the core session at BSDCan 2025 developer summit, and core is still collecting feedback and working on the policy."' -- source: https://developers.slashdot.org/story/25/09/03/1649201 Cheers, Peter

1 0

Are AI Web Crawlers 'Destroying Websites' In Their Hunt for Training Data?
by Peter Reutemann 01 Sep '25

01 Sep '25

Hi everyone I thought this news item rather fitting when I saw it this morning, considering Daniel's experience with these AI bots hammering the wlug wiki: '"AI web crawlers are strip-mining the web in their perpetual hunt for ever more content to feed into their Large Language Model mills," argues Steven J. Vaughan-Nichols at the Register. And "when AI searchbots, with Meta (52% of AI searchbot traffic), Google (23%), and OpenAI (20%) leading the way, clobber websites with as much as 30 Terabits in a single surge, they're damaging even the largest companies' site performance..." How much traffic do they account for? According to Cloudflare, a major content delivery network (CDN) force, 30% of global web traffic now comes from bots. Leading the way and growing fast? AI bots... Anyone who runs a website, though, knows there's a huge, honking difference between the old-style crawlers and today's AI crawlers. The new ones are site killers. Fastly warns that they're causing "performance degradation, service disruption, and increased operational costs." Why? Because they're hammering websites with traffic spikes that can reach up to ten or even twenty times normal levels within minutes. Moreover, AI crawlers are much more aggressive than standard crawlers. As the InMotionhosting web hosting company notes, they also tend to disregard crawl delays or bandwidth-saving guidelines and extract full page text, and sometimes attempt to follow dynamic links or scripts. The result? If you're using a shared server for your website, as many small businesses do, even if your site isn't being shaken down for content, other sites on the same hardware with the same Internet pipe may be getting hit. This means your site's performance drops through the floor even if an AI crawler isn't raiding your website... AI crawlers don't direct users back to the original sources. They kick our sites around, return nothing, and we're left trying to decide how we're to make a living in the AI-driven web world. Yes, of course, we can try to fend them off with logins, paywalls, CAPTCHA challenges, and sophisticated anti-bot technologies. You know one thing AI is good at? It's getting around those walls. As for robots.txt files, the old-school way of blocking crawlers? Many — most? — AI crawlers simply ignore them... There are efforts afoot to supplement robots.txt with llms.txt files. This is a proposed standard to provide LLM-friendly content that LLMs can access without compromising the site's performance. Not everyone is thrilled with this approach, though, and it may yet come to nothing. In the meantime, to combat excessive crawling, some infrastructure providers, such as Cloudflare, now offer default bot-blocking services to block AI crawlers and provide mechanisms to deter AI companies from accessing their data. ' -- source: https://tech.slashdot.org/story/25/08/31/1820249 Cheers, Peter

1 0