Topic · 35 posts
LLM AssBench 
assbench.com
The definitive test for generative AIs: which creates the most accurate ass. There are several models available and you can read the prompt. Don’t forget to also evaluate the reaction of the buttock to the pat. Tip from Renan, who refused to answer where the hell he found this.
At the end of August, the results came out for a general resolution on the use of generative AI (LLMs) in Debian. Option 5 won: “Responsible use of generative AI.” The outcome — you can probably guess — left a lot of people unhappy.
This week, it was the KDE project’s turn to end up in the crossfire while trying to propose a policy on the use (or non-use) of AI in its code. Tempers flared when ~the internet discovered that one of the talks at Akademy, the project’s annual event, proposed “a lovable, sovereign, AI-native KDE,” pitched by two people who had been involved in KDE development in the past, but no longer are.
I’ve published more than a couple of pieces on this blog trying to raise awareness about the risks of viral photo-editing apps, like FaceApp back in 2019 [pt_BR].
We are curious beings. The promise of looking into a magic mirror that shows us old, as a different gender, or in another era is nearly irresistible — a force that makes many people ignore the warning signs, even the obvious ones.
LibreOffice 26.8, released on August 26th, became the software’s most popular update. LibreOffice is a free alternative to Microsoft Office that’s been around for nearly two decades. In one week, the installer was downloaded more than 1 million times — not counting updates via Linux distribution repositories.
Someone might say the improvements to the writing system and typography are a hit among NGOs, government agencies, and offices that use LibreOffice. I’m putting my money on a “non-feature,” though: the statement that LibreOffice doesn’t come with generative AI features due to the technology’s (lack of) privacy.
A day after celebrating the download record and the wide media coverage, The Document Foundation (TDF), which manages the software, spelled out its position in a post titled “Yes, no AI is now a feature”.
The article, signed by Italo Vignoli, says TDF “does not reject artificial intelligence out of hand,” but that the technology doesn’t yet meet an exhaustive list of principles to be included by default. Those principles are:
For now, TDF recommends community plugins to integrate AI into LibreOffice.
TDF’s distinctive stance “has almost nothing to do with the technology,” the article continues, before taking a jab at rivals that dove headfirst into AI. For the foundation, “an AI assistant justifies a price increase and strengthens the case for keeping all documents within its own infrastructure,” referring to companies that rely on selling subscriptions.
The Document Foundation’s detailed explanation may disappoint anyone who read the announcement at the LibreOffice 26.8 launch as an anti-AI manifesto.
Whatever the reasons, the LibreOffice 26.8 case answers a question I and others have had since 2023: if the whole market now offers AI baked in software of every kind, is there really no room for AI-free alternatives to stand out? Apparently, there is.
State of the [WordPress] Community 2026
thewpocc.org
The results are out from the independent survey on the WordPress community, conducted by the WP Open Community collective.
It represents a scathing critique of WordPress’s direction and, especially, of Matt Mullenweg’s leadership. For example, 73.8% said they feel more negative about the project than 24 months ago. When asked about open source software in general, the negative outlook dropped to 23.9%. The biggest source of problems, with 73.3% saying they had experienced or witnessed incidents, relates to governance and leadership.
The use of generative AI in WordPress displeases most respondents, even among those who use AI at work.
In his closing remarks at WordCamp US on Wednesday (19th), Mullenweg brought up this survey finding and suggested that “We might just need to rebrand some of what we’re doing [with AI]” such as “automation,” according to the account from the blog The Repository. It strikes me that this reaction adds to the evidence of a leadership that doesn’t listen to the community.
How Claude’s text watermark works
anthropic.com
All major LLM providers will be required, by the AI Act of the European Union, to add watermarks to the generated content, including text. Anthropic, owner of Claude, explained its method:
Large language models like Claude work by generating one word at a time. Each time the model decides on the next word, it chooses among a list of potential candidates, ultimately selecting the most sensible or likely based on the preceding text. Take the sentence “The weather today was cold and…”. The next word is very unlikely to be “sugary.” But it is quite likely to be “overcast” or “grey.” Under most circumstances, it doesn’t matter much to the reader which of these latter two words the model ultimately chooses—the meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number.
Watermarking uses low-stakes choices like these—which occur many times over a piece of generated text—to leave a pattern in Claude’s responses. That pattern is undetectable to the reader, but is detectable to anyone who has a key that encodes it. When watermarking is used, choices are still made at random, but the source of the randomness is different. Instead of using an arbitrary random number generator to pick the next word, watermarking uses the key and a few words that come before to settle what word the model should pick. That is, the words that Claude picks are still random, but now, one can check the sequence of words and see if it’s consistent with the choices Claude would make if it was using the key. If it is, one can assign a probability that the text was generated by Claude.
The solution is the same one Google created and has used since 2024 in Gemini, called SynthID-Text.
Some people are freaking out over this change. As I’ve been saying for a while now, it doesn’t matter.
Besides, simple systems that swap out some words for synonyms are likely to show up (if they don’t exist already), which (in theory? someone correct me if that’s the case) would break SynthID-Text’s digital signature. Still in the realm of guesswork, I’d imagine that feeding the text into another LLM and asking it to swap out some words and structures would already do the trick.
Regardless of all that, isn’t it about time people who use generative AI to write text started owning up to their choices?
I think the first thing that comes to mind when this topic comes up is using the watermark to point fingers and shame whoever uses it. It’s possible to imagine other uses, though, such as helping triage requests sent to public agencies, which, as Ronaldo Lemos pointed out [pt_BR] in his column in Folha de S.Paulo this past Sunday (the 16th), are already swamping governments and courts in several countries.
Another victim of the AI bubble: Mico, Microsoft Copilot’s mascot that looks like a blob of semen. According to the company, it’s been demoted from Copilot Voice and now only shows up in Copilot Learn Live. (What are Copilot Voice and Copilot Learn Live? Your guess is as good as mine.)
Mico’s near-demise is part of the unification/centralization process for the countless Copilot apps and services.
Mark Zuckerberg
Meta CEO and co-founder
Maybe Zuck should put those superintelligent AI lawyers to work on Meta’s own lawsuits. Even with the best (flesh-and-blood) lawyers money can buy, the company keeps racking up losses in U.S. courts, accused of harming minors with its addictive apps. The latest was last week (6th), in New Mexico.
The line above comes from a lengthy essay by Meta’s CEO, “The future is for everyone.” Does anyone still take this drivel of platitudes, lies, and empty promises seriously? This must be the fourth or fifth one he’s put his name to. None of them have aged well. There was the one about encrypted messaging on Instagram in 2019, the “global community” one from 2017, the metaverse one from 2021 again… notice the pattern.
Terry Godier, who made a name for himself in certain corners of the web earlier this year with an essay on RSS and his subsequent iOS app, Currents, launched another one called Dark Hours, this one on the web, after Apple reportedly rejected it for containing astrology elements, which the App Store’s policies ban and which Godier, at first, denied was the case.
Time has started serving ads to AI agents
digiday.com
Put another way, the ads give brands like Ally and the non-profit professional organization Project Management Institute another lever to pull on how they show up in generative search. LLMs can pick up a brand’s information from the ad and index it, which can then shape AI-generated responses, which in turn, can influence humans using those AI answer engines.
Maybe my idea of “poisoning” AIs with a text hidden by steganography is not that crazy.
Sam Altman
OpenAI CEO and co-founder
This one goes to the list of great ideas of the AI bros. As someone replied on X, “What if you just talked to your children?”
Two years ago, I wrote about one of those “human made” badges used to signal content free of AI. Since then, these badges have spread across corners of the internet hostile to generative AI.
In a somewhat viral post from the indie web, Case Duckworth lists a few reasons to reject them. His criticism isn’t aimed at the badges themselves, but at initiatives that centralize this kind of signaling. His view is condensed in this paragraph:
AI-drunk companies love to trumpet their millions or billions of users. What they don’t tell you is that most are forced to use such features. Think of the AI overviews that Google Search forces on you, whether you asked for it or not.
Samsung seems to have taken this tactic to the next level. According to How-To Geek, the Samsung Health app will delete users’ health data from the company’s cloud — making sync between devices impossible — unless they agree to hand it over for AI training and modeling.
The new consent toggle comes enabled by default and is buried in the app settings. It shows this message when someone tries to turn it off:
Withdraw from this agreement?
You will not be able to sync health data with your Samsung account and your health data will be deleted unless retained pursuant to applicable law. If retention is required, we will erase it as soon as the required retention period ends.
Samsung collects four types of data in Health:
A message on Samsung’s website warns that people may have access to this data to review it.
***
Samsung’s biggest rival, Apple, is also preparing to make Apple Intelligence — its suite of generative AI tools embedded in its operating systems — mandatory starting from version 27.
This comment from Bruno, on Bluesky (pt_BR):
Instagram launches new tool that drains your entire bank account and kills your family. Here’s how to turn it off 👇
Instagram doesn’t (yet) do that. His joke references the recent news that photos from public Instagram profiles could be used to generate new images on WhatsApp with Meta AI, Meta’s generative AI, which also owns Instagram. The press picked it up and led with “Here’s how to turn it off” in the headlines.
Until now, Apple has kept Apple Intelligence — the suite of artificial intelligence tools baked into its operating systems — entirely optional. If you don’t want it, you can simply skip it during a new device setup or when updating the system.
That show of respect for its customers may change with iOS/macOS 27. Reports suggest that, at least in the first beta, Apple Intelligence is mandatory:
