The AI push runs on setting venture capital money on fire and charging chatbot users way less than the bot costs to serve.
The idea is that the customers will get hooked on the business value. Then the AI vendors can gouge them. And the vendors can try not to haemorrhage quite so much cash.
So Microsoft, Anthropic, and OpenAI have all been moving their software customers off monthly subscriptions to token-based billing. Tokens have the added enterprise pricing advantage that the cost of anything is completely obscure.
One source told Axios that one of their clients had spent $500 million in a month on Claude Code. We don’t know who it was — but the best guesses are Amazon or Uber, both of which are known to have run up stupendous bills. [Axios]
It’s not just the coders — it’s the ordinary office workers! They’re being told to AI up everything and give it that veneer of slop. That’s professional now.
Walmart said in early June that it was rationing its internal AI tool “Code Puppy”, which did office workslop as well as code. Code Puppy used to be unlimited. Now it’s not. [Bloomberg]
This was really an excuse for layoffs — Accenture’s consulting business is badly down and its stock price has cratered over the past year.
But Accenture employees responded to the incentive — and now the bill’s coming due. 404 Media got a leak of a meeting at Accenture: [404, archive]
“We’re seeing from some of the data internally at least that it’s actually not our engineers that are driving the token consumption. It’s a lot of the non-engineers that are doing some of those behaviors […] you were talking about.”
… Stuart Henderson … jokes he hopes Kwak didn’t just convert a PDF into images and then into markdown files. “I’m learning that’s one of the big token chewers,” Henderson says. “Turning PDFs into markdown: is that right?”
If you order people to use a chatbot that doesn’t do anything useful, they’ll just point it at any old trash. Then the bill hits.
The customer backlash is bad enough that Sam Altman at OpenAI is openly talking about price cuts. [WSJ, archive]
But OpenAI can’t afford price cuts. They need revenue numbers to make their planned IPO look plausible. And Altman doesn’t have the sort of Elon magic that SpaceX ran their IPO on.
Elon Musk isn’t convincing the markets either. The SpaceX stock price shot up — then back down a few days later. It’s been steady since, just below the offering price.
The AI scam cannot possibly pay for itself from sales. AI turns out not to be critical for real business work.
The use case for AI is doing things that should not be done. And there’s only so much market for that.
Between February 2022 and September 2025, Bellingcat staff and volunteers collected, geolocated, and shared more than 2,500 incidents of civilian harm following Russia’s full-scale invasion of Ukraine.
As part of this effort, Bellingcat tested a new machine learning model intended to rank Telegram social media posts on their likelihood of containing incidents of civilian harm.
This novel methodology dramatically reduced the search and selection time required, freeing researchers to focus on verifying incidents of civilian harm – not just searching for them.
This piece documents our methodology, ethical considerations and lessons learned in the hope that others researching similar topics can benefit from our work.
Open source research into civilian harm is still a relatively new field and it presents many challenges – one of the biggest is organising and sorting through the huge volume of user generated content being produced to find what is relevant.
Machine learning, a form of artificial intelligence that uses algorithms to identify patterns from large amounts of data and make predictions, can make this task more efficient.
With ongoing conflicts involving large amounts of civilian harm occurring in Sudan, and much of the Middle East, this guide aims to offer those covering these conflicts an example of how machine learning can be used to help find and sort incidents. You can also access the Code Notebook for our model here.
We defined “civilian harm” not just as civilian deaths or injuries resulting from armed conflict, but also the broader and delayed effects on civilians from mental trauma, loss of livelihood, displacement, destruction of infrastructure and more. This definition was informed by the Protection of Civilians bookon civilian harm.
Initial Telegram Dataset
Each Telegram post containing civilian harm which had already been manually verified by researchers was used to build an initial dataset of confirmed cases of civilian harm, which data scientists call positive instances. We collected a total of 5,848 unique URLs for these Telegram posts. For our manual collection we reviewed posts on relevant Telegram channels, working through oldest to newest posts each day. Assuming that a given post made it to our geolocated incidents list, it meant the researcher who flagged it also looked at the posts that appeared before and after it on Telegram and did not flag those ones, so we selected the 10 posts surrounding the verified civilian harm post as our additional dataset of posts that did not contain civilian harm. After excluding any deleted or duplicate posts, we ended up with 48,545 non-civilian harm posts, our negative instances.
The choice to overrepresent negative instances aims at better reflecting the real world and increasing data available for model training.
We enriched each URL with metadata from the Telegram API, such as the time of publication, reactions or textual content. As some of these posts had been deleted, we completed the missing data points with previously preserved versions from our Auto Archiver database, only available for the positive instances.
Feature Engineering
Training a machine learning model requires numerical data, as these models compute a prediction score based on mathematical operations.
We built these by converting raw data from our initial dataset, such as keywords signalling potential civilian harm, into numerical scores (or “features”) that the model could interpret, with the aim of increasing the model’s ability to identify patterns. This process, known as feature engineering, can significantly improve model results because it allows data scientists to suggest explicit context knowledge.
A full list of features we used to train the model can be found in the code notebook accompanying this piece. Many features were directly inspired by researchers’ input from their experiences manually screening cases of civilian harm by sorting through a set number of Telegram channels and inspecting each post individually.
Several of the features used were directly built from the metadata contained in each Telegram post including media_type, day_of_week; or binary ones: forwarded, edited andreply_to.
Other features included engagement information: views, forwards, total_reactions, and even individual features for most used emojis including the reaction_crying_face to count emoji.
Converting Text to Numbers
To embed the experience from the manual collection process, researchers put together a list of keywords both in Ukrainian and Russian that, to them, signalled posts likely to show civilian harm. For instance, “Шахед” and “КАБ” translated to “Shahed” and “Guided aerial bomb” respectively. We created a numerical feature to count their frequency.
In addition, we included several generic English-language keywords which meaningfully signalled potential civilian harm, such as “injured”, “school affected” and “hospital affected” that were only used for generating semantic similarity scores.
A semantic similarity score is a calculation used to determine the proximity in meaning between different words and phrases. To get the semantic similarity between the post text and each of our keywords, we represented each in a list of numbers via a Sentence Transformer model, which converts words into numerical representations called vectors that a computer can understand.
We then calculated the level of similarity between each vector using cosine similarity, one of the most popular methods for measuring similarity between two pieces of text.
Due to how embeddings work, this calculation results in a figure on a scale from -1 (no semantic proximity) to 1 (same meaning). For example, the words “hurt” and “injured” would have a high similarity score, while “residential” and “injured” would have a negative score as the words are not semantically similar.
Finally, to enable the model to identify the relevance of each post to civilian harm in Ukraine, we used a multilingual text transformer from the BERT family of language models to represent the entire post’s text as a vector of 768 numerical values. This model can efficiently represent text from many languages in a way that captures meaning: the same sentence in different languages will generate similar embeddings, and trained machine learning models can detect patterns in the embeddings.
It is important to note that for this initial prototype of a civilian harm detection model, we did not include any features derived from media content such as photos and videos, although that would be a logical next step in attempting to improve model performance.
Selecting, Training and Evaluating Models
With 54,393 rows of 893 numerical features each, we selected four machine learning algorithms to train our predictive models.
We chose Logistic Regression as a baseline algorithm due to its simplicity. We also selected three other “best in class” models, Random Forest, XGBoost, and LightGBM. These choices centred on the interpretability of the models and their ability to work on tabular data of this size. For example, we avoided neural networks due to a lack of interpretability and because those models work best with a larger dataset.
To genuinely assess the performance of the trained models, we split our dataset into three parts:
A training set – the data the models were trained on (60 percent of the full dataset’s rows)
A validation set – used for an intermediary evaluation when tuning model parameters (20 percent of all rows)
A test set – hidden for the final performance assessment, so the models were evaluated on unseen data (remaining 20 percent of rows)
We used a stratified split to divide the dataset instead of a random split. This method ensured the proportion of positive instances (i.e. confirmed cases of civilian harm) remained consistent across all three sets at about 11 percent.
To measure the performance of machine learning models, we ran them through the test set and measured the number of correct and incorrect predictions. Models output a likelihood between 0 and 1 that each Telegram post contains civilian harm, and we tried to find a cut-off threshold that leads to a good balance between flagging almost every post (0.1) or flagging very few (0.9).
There are two main types of evaluation metrics to gauge a model’s prediction power. Recall asserts what fraction of positive instances (i.e. known civilian harm posts) were correctly flagged as such. Precision measures the fraction of posts flagged as civilian harm that are indeed civilian harm posts.
During the training phase, we tuned the models to maximise average precision (PR-AUC), a metric that summarises precision across all recall levels. While this method also accounts for precision, it prioritises recall, which is preferable for this use case as it steers model selection to reduce the number of civilian harm posts that are skipped.
The following table sorts models from best to worst PR-AUC against a baseline of a coin-flip predictor. ROC-AUC and F1 are two other evaluation metrics included as sanity checks. Simply put, ROC-AUC measures the probability of ranking two instances, one negative and one positive, correctly; F1 balances precision and recall equally and its best cut-off threshold value.
Model test scores comparison, XGBoost stands out in every relevant metric evaluated.
From these results, we selected XGBoost as our final model as it had the best scores when compared across all metrics.
Interpreting the Model
Because these models are interpretable, we can understand which features are the most useful when predicting whether a post includes civilian harm. The above table shows the top 10 features that most strongly signal the XGBoost model to make a decision:
semantic_keywords_similarity: the semantic proximity between the post text and manually selected keywords “casualties”, “damage” and “civilian harm”
bert: the model was able to discern meaning from the text with the same strength as some of the other features in this list – there are three cases of this in the top 10
reaction_crying_face: reactions with crying face emojis on the post
group_of_messages: whether a post contains multiple media files
keywords_in_text: the number of custom Ukrainian or Russian keywords in the post
These results generally tally with what you might expect when selecting Telegram posts for instances of civilian harm, including that posts that generate a lot of emotional engagement and posts using keywords about civilian harm were among those most likely to contain content related to this topic. Not all models had the same top features as XGBoost. In fact, for the Random Forest model the most important feature was the number of crying face emojis present in a post, a soft pattern highlighted by researchers when this methodology was first imagined.
LLM Results and Comparison
Retroactively, we decided to run a sample of the same test dataset through different large language models (LLMs) to gauge their ability to make these same predictions.
We aimed to include an LLM-generated score as an extra feature for our trained models, which would be captured as relevant if it correlated with the correct predictions.
To start, we selected two local models, the 1B and 4B variants of Gemma 3 from Google DeepMind, and two cloud-hosted models, Gemini 2.5 flash and Gemini 3.5 flash. With this selection, we hoped to compare results across a wide range of models’ expected performance.
We generated a 400-row stratified sample (preserving the same proportion of real civilian harm instances) from the test dataset used for the custom models. For each of the four LLM models, we ran two tests: one where only the Telegram post message was sent, and another including both the message and the engineered features (excluding the text embeddings, as the model had direct access to the text). In the prompt for each model, we asked for a score between 0 and 1. We then evaluated the results as we did for the custom models.
The above table shows that LLMs can indeed extract value from the engineered features. All four LLMs surpassed the baseline Logistic Regression model in our tests, yet none of them performed better than the other custom-trained models, and XGBoost remained the one with the highest PR-AUC.
Still, Gemini 2.5 Flash performed better than its newer version 3.5 and even achieved a slightly higher best F1 score than any other model. While this is a good result, for the flagging of civilian harm posts, the PR-AUC remains the crucial metric, as it captures the model’s ability to identify infrequent instances of civilian harm while minimising false positives.
Ethical Considerations
Introducing an instrument of automated decision-making into a process of detecting civilian harm brings inherent ethical questions. These include automation bias, or how humans tend to blindly place faith in machine-generated recommendations; algorithmic bias, or how the results of these models echo the same patterns present in the training data, including under- or over-representation of types of civilian harm.
The decision to test an automated methodology for this particular project came from the fact that there were limited resources for both steps in the process – the detection of potential civilian harm and its actual verification. Historically, we built an enormous backlog of unverified incidents because a lot of time had to be spent on monitoring the most recent events so that potential evidence would be captured and preserved as soon as possible.
The automation of this process also reduced the exposure of researchers to a significant amount of unpleasant and distressing visual and text content, reducing the burden of exposure to traumatic content.
For this project, we tried to ameliorate the ethical challenges with a number of strategies including randomly flagging posts not captured by any model, monitoring which features models relied on to make decisions, and by doing historical comparisons of patterns in data.
Additionally, as stated above, for this initial prototype of a civilian harm detection model we did not include any features derived from the media content itself. In the future, it would be a logical next step in attempting to improve the model performance, to include the media from the posts – but using AI to review actual media comes with additional ethical challenges such as model bias.
Because of the opaque ownership of many LLM companies and their generative nature, the use of LLMs for an extra feature presented additional ethical challenges including privacy and safety concerns considering the sensitive nature of the data. Our model did not rely on LLMs, though we retroactively ran a sample through it.
How the Model Fits into the Bigger Picture
After selecting this model, we created a user interface where researchers could view a list of Telegram posts sorted from most to least likely to contain indications of civilian harm. The user interface was designed for quick triage and integration, where a positive confirmation from researchers would instantly send the post to the Auto Archiver (Bellingcat’s tool for preserving digital content) and then transfer it to ATLOS (our internal collaborative verification platform). Bellingcat staff and volunteers could then manually verify incidents. Researcher input was constantly stored so that this data could be used to improve the model in the future.
Preliminary feedback indicated that the AI model was useful. Not only were we able to reduce time and harm from scouring through dozens of war reporting Telegram channels, researchers also reported that the stream of new posts being added to the verification backlog were capturing real and diverse cases of civilian harm.
Despite the focus on civilian harm and Telegram (highly popular in Ukraine and Russia), this pipeline is generic and can be adapted to other conflict monitoring tasks. How easily this can be done does depend on how open the social media platform is and whether it is possible to scrape posts from it. Apart from that, it is easy to incorporate new features and data, and cheap to automatically retrain, test and deploy models as the system receives more human input.
Looking forward, sorting through overwhelming amounts of data in a conflict will continue to be challenging. Hopefully, this methodology can help newsrooms, conflict monitoring organisations, and others find the balance between ethical considerations and resources in order to carry out open source investigations on civilian harm and human rights violations.
Bellingcat is a non-profit and the ability to carry out our work is dependent on the kind support of individual donors. If you would like to support our work, you can do so here. You can also subscribe to our Patreon channel here. Subscribe to our Newsletter and follow us on Bluesky here, Instagram here, Reddit here and YouTube here.
In February, police in Claremore, Oklahoma arrested farmer Darren Blanchard for speaking a little too long during a community meeting about data centers. The city charged Blanchard with criminal trespass, a crime with a $200 penalty, but he’s vowed to fight the charge. He recently shared video of the bodycam footage for the first time with 404 Media and answered our questions about the moment cops arrested him for going over his time at a February 17 community meeting of the Claremore City Council.
The plan in February was for the City Council to listen to the concerns citizens had about a planned data center called Project Mustang. The residents of Claremore don’t want the data center and largely feel like the construction project was approved without their input. City officials signed non-disclosure agreements on behalf of the project’s developers and haven’t been forthcoming with details about its construction.
Object permanence: Major AI breakthrough; Disney v Pooh tombstone; Vancouver riot kiss; Farage admits Brexit lies; Protecting the web from its founders; Sanders x Hillary; Surveillance pricing v your dollars.
It's not often that someone on a panel says something that makes my jaw drop, but that's what happened earlier this week when the moderator of a panel I was on in Toronto described jailbreaking an iPhone as "rampant theft of IP."
Some context: the panel was in Toronto, and the nominal subject was "digital sovereignty," though all the panelists (except me) interpreted that to mean "sovereign AI." All of their interventions were focused on how Canada could build and operate its own AI, which I found very weird, since there is no AI-related threat to Canadian sovereignty. If Donald Trump ordered OpenAI and Anthropic to turn off all of Canada's chatbots tomorrow, nothing would change: every firm, ministry and household would operate as per normal:
Now, that's not to say that Canada doesn't have a digital sovereignty problem – it really does! Donald Trump and US Big Tech have fused into a single entity and Trump now orders US tech giants to terminate the online accounts of foreign officials who displease him. When Microsoft turns off your Office365 account, you lose your working files, your calendar, your address book, your email archives, and the Outlook email address you use to log in to every online service:
So while turning off Canada's chatbots would not inflict any real harm on Canada, M365 terminations could paralyse any federal or provincial ministry, any structurally important firm, and most Canadian households.
The threat doesn't stop there: Trump can also order Apple and Google to brick any of Canada's iPhones or Android devices – terminating individual officials' mobile access, or terminating whole provinces. It's not just iPhones either – Trump can also brick any tractor in Canada:
This is the real digital sovereignty risk, and Canada needs to address it now. But Canada can't – our hands are tied…by us. In 2012, we passed a law, The Copyright Modernization Act, that criminalizes "jailbreaking," meaning that Canadian companies can't go into business figuring out how to install different app stores on phones and consoles, or change the firmware in tractors to enable independent repair, or reliably export their cloud data to rival Canadian services:
Why did we pass this law? Because the Americans promised us free trade and no tariffs on our exports if we agreed to it. That's a promise Trump tore up, but we're still holding up our end of the bargain. That's crazy. It means that American companies can use Canada's courts to destroy Canadian businesses that offer the Canadian people tools to help them escape Big Tech's sleazy ripoffs of their data and cash.
And boy do those US tech companies take in a lot of cash. The US ad-tech duopoly of Google/Meta rig the advertising market, taking 51% out of every ad dollar through an illegal, collusive arrangement called "Jedi Blue":
The US mobile tech duopoly takes 30 cents out of every dollar spent via an app, by forcing every app vendor to use their payment processors, which charge 1,000% more than any other payment processor in Canada. That means that every time a subscriber to a Canadian news site signs up through an app, 30% of the lifetime subscription revenue for that Canadian subscriber is funneled to one of two California companies.
The corollary, of course, is that if Canadian businesses were free to compete with US companies – if Canada stopped foolishly holding up its end of the bargain that Trump has dishonoured – then it would be as though every Canadian news outlet increased its subscriber base by 25% overnight! What's more, the Canadian companies that sell those jailbreaking tools would make billions out of US Big Tech's billions.
And that's where the moderator of this week's panel comes in. When I finished making this pitch, they turned to the rest of the panel and said something like, "Well, apart from rampant theft of IP, what else could Canada do to secure its digital sovereignty?"
That's when my jaw dropped. Making it possible for, say, a Canadian company to sell its own Canadian game to a Canadian customer, in Canada, without giving Apple or Xbox 30% of the purchase price, is not "theft of IP." It's not "theft of IP" for a rightsholder to sell their own products to their customers. It's not "theft of IP" for a Canadian owner of a device to decide for themselves which software they want to run on it. If buying software from the company that made it and installing it on a device you own is "theft of IP," then so is putting non-Nike shoelaces in your Air Jordans.
It's not "theft of IP." It's just good business. Moreover, it's the kind of good business that created America's tech giants in the first place. As Jeff Bezos tells his suppliers: "Your margin is my opportunity." US tech giants make whopping margins around the world, thanks to the anticircumvention laws that the US Trade Rep crammed down every US trading partner's throats, laws that allow US companies to use other countries' legal system to destroy their competitors.
I've been mulling this "rampant theft of IP" remark for a couple of days now, but it wasn't until a reader wrote to me to remind me about Apple's origin story that I realised what the punchline is. Apple founders Steve Jobs and Steve Wozniak financed their first product launch by selling "Blue Boxes" (devices that let you make free long distance calls by cheating the phone company) door to door in the UC Berkeley dorms:
Now, I'm not going to weep for the lost revenues that Jobs and Woz denied to AT&T. After all, AT&T was stealing that money from its customers, which is why, just a few years later, a federal court convicted AT&T of monopolistic practices and broke the company up:
But the legal term for what a Blue Box does is "toll theft," which is to say, Apple – a company literally founded on theft – now makes the majority of its profits by convincing people that making a competing product is literally stealing. A company whose founders got their seed capital by marketing illegal circumvention devices now markets products designed to make it a crime for a rightsholder to sell their own work to you.
I've long said that "every pirate wants to be an admiral":
But this is just a little too on the nose. When Apple went into business selling products to rip off the phone company, that wasn't progress. When Canadians go into business selling devices that let iPhone owners use their own property to do legal things – like buying copyrighted works directly from their creators – that is not piracy.
Canada has a real digital sovereignty problem, and it's not AI. Canada will not mitigate its digital sovereignty risk by successfully launching a Made in Canada version of the money-losingest venture in the history of the human species:
Canada's real digital sovereignty problem is its reliance on the apps, cloud services and devices that are tethered to the American cloud, access to which Donald Trump could – and does – terminate whenever he feels grumpy. Trump has repeatedly threatened to annex Canada and turn us into "the 51st state." He's trying to steal Alberta right now. Our digital sovereignty risk is the risk of Trump paralysing our country in order to steal Alberta – or the entire shop.
We can address that digital sovereignty risk – and make billions at the same time – by legalising jailbreaking and becoming the world's "disenshittification nation." Unlike a program to build Canadian AI, this will make billions, not lose them – and unlike Canadian AI, this will make our country more resilient and safer, by delivering products that Canadians – and the world – want to buy and will pay us a fortune for.
#15yrsago Rochester police use selective enforcement of parking laws to harass attendees at a meeting in support of Emily Good https://rochester.indymedia.org/node/7516
"Red Team Blues": "A grabby, compulsive thriller that will leave you knowing more about how the world works than you did before." Tor Books http://redteamblues.com.
"Chokepoint Capitalism: How to Beat Big Tech, Tame Big Content, and Get Artists Paid, with Rebecca Giblin", on how to unrig the markets for creative labor, Beacon Press/Scribe 2022 https://chokepointcapitalism.com
Currently writing: "The Post-American Internet," a sequel to "Enshittification," about the better world the rest of us get to have now that Trump has torched America. Fourth draft completed. Submitted to editor.
"The Reverse Centaur's Guide to AI," a short book for Farrar, Straus and Giroux about being an effective AI critic. LEGAL REVIEW AND COPYEDIT COMPLETE.
"The Post-American Internet," a short book about internet policy in the age of Trumpism. PLANNING.
A Little Brother short story about DIY insulin PLANNING
This work – excluding any serialized fiction – is licensed under a Creative Commons Attribution 4.0 license. That means you can use it any way you like, including commercially, provided that you attribute it to me, Cory Doctorow, and include a link to pluralistic.net.
Quotations and images are not included in this license; they are included either under a limitation or exception to copyright, or on the basis of a separate license. Please exercise caution.
"When life gives you SARS, you make sarsaparilla" -Joey "Accordion Guy" DeVilla
READ CAREFULLY: By reading this, you agree, on behalf of your employer, to release me from all obligations and waivers arising from any and all NON-NEGOTIATED agreements, licenses, terms-of-service, shrinkwrap, clickwrap, browsewrap, confidentiality, non-disclosure, non-compete and acceptable use policies ("BOGUS AGREEMENTS") that I have entered into with your employer, its partners, licensors, agents and assigns, in perpetuity, without prejudice to my ongoing rights and privileges. You further represent that you have the authority to release me from any BOGUS AGREEMENTS on behalf of your employer.
In a very short while the UK will be holding a referendum on the country's continued membership of the EU. At the moment, judged on the polls (which are of course often inaccurate, as with the last general election), things seem to be heading the way of a vote to leave.
I think it would be a great shame were this to happen. As a young scientist, I benefitted tremendously from the freedom of movement allowed within the EU. I'm not talking about my time within the European Space Agency, which is a non-EU organisation, although that experience certainly helped frame my views on European cooperation and integration. But having left ESA in 1994, I was immediately able to take up a two year postdoctoral position at a Dutch university, and I did so with the minimum of hassle and paperwork. Once again I was immersed in a pan-European working environment which I found stimulating and encouraging.
After my postdoctoral position expired in 1996, I found myself unemployed. There were a handful of possible job opportunities back in the UK, but I had grown fond of the Netherlands, and my partner at the time, who later became my wife, had a full-time job. She too had benefitted from freedom of movement within the EU. Disinclined to leave Holland, therefore, I signed on for unemployment benefit from the Dutch state, while continuing to look for work opportunities within the area where we lived. I applied for one job in Delft, working on satellite monitoring of the Earth's atmosphere, but was not made an offer.
Luck eventually intervened, in that I saw an advert in Nature for a newly founded business in Haarlem, which would revolve around developing scientific software for astronomical applications. It seemed right up my street, almost literally so, in that Haarlem was only a short train ride from where we lived near Leiden. I applied for the job and was suitably astonished to learn that the driving force behind the business was an old colleague of mine - and a Welshman, like me, who had settled in the Netherlands. We met for an interview, which went well. While there was a strong prospect of working for the company in the future, though, there was still going to be a few more months of unemployment. I therefore continued to sign on, while going through the motions of looking for work. It was an odd, unsettling time, but - in hindsight - a blessing, because it enabled me to dust off the abandoned manuscript of Revelation Space and finally give it the polish it needed prior to submission. That was early 1997, and the book sold two years later. Those months of unemployment were therefore literally life-changing, and I owe them to the Dutch state and EU regulations on worker's rights.
Many of the arguments for and against membership of the EU seem to revolve around economics, which seems to me to be an extremely narrow metric. Even if we are better off out of the EU, which we probably won't be, so what? This is already a wealthy country, and leaving the EU won't mend the widening inequality between the very rich and almost everyone else. More than that, though, look at what would be lost. Friendship, commonality, freedom of movement, a sense that national boundaries are (and should be) evaporating. When many countries (including the Netherlands) moved to the Euro, it was a joy not to have to pack Guilders, Belgian francs, Deutschmarks, for a simple drive to visit to family in Germany a few hours away. The eradication of visible borders did not lead to a smearing out of regional cultures, but instead it made it much more easy to sample those cultures and gain a deeper sense of European history. I never stopped feeling that living in the EU was a thing to be proud of, and more than ever I am content to think of myself as European before British. I therefore hope that the Remain vote will win the day.
Meta announced a fabulous new programme in April — the Model Capability Initiative! [BBC]
a new tool will run on Meta’s computers and internal apps, logging their activity to be used as training data for AI technology.
This collected every keystroke, mouse click, and screen image. Meta told the BBC it had:
safeguards in place to protect sensitive content.
Meta employees started a petition in May against the key logging. They asked about these safeguards. There weren’t any: [letter]
no completed privacy reviews were provided. The outlined privacy mitigations were vague, and leadership’s confidence in them appeared limited — evidenced by the selective opt-out afforded to executives.
Meta left potentially sensitive information collected from employee laptops accessible to anyone inside the company.
Meta paused the data collection. But it wants to start keylogging again as soon as possible. [Wired]
Meta is also sending expensive engineers off to become AI trainers in the Applied AI Engineering Unit! [Reuters]
Join or be fired. Applied AI head Maher Saba said:
AAI is one of the company’s highest priorities and we’re resourcing it by moving our strongest talent to address it. Therefore, the transfers aren’t optional.
The engineers are writing programs to generate training and tests for the AI. One draftee told Wired: [Wired]
You have zero purpose in life all of a sudden, you barely interact with anyone, you just have these tasks every week.
Log in
emoji.
