US citizens: call on your congresscritter and senators to make the rich pay more into Social Security, and not cap or reduce what retirees receive.
See the instructions for how to sign this letter campaign without running any nonfree JavaScript code--not trivial, but not hard.
US citizens: Join with this campaign to address this issue.
To phone your congresscritter about this, the main switchboard is +1-202-224-3121.
Please spread the word.
US citizens: call on your senators to reject Todd Blanche for Attorney General.
See the instructions for how to sign this letter campaign without running any nonfree JavaScript code--not trivial, but not hard.
US citizens: Join with this campaign to address this issue.
To phone your congresscritter about this, the main switchboard is +1-202-224-3121.
Please spread the word.
US citizens: call on your congresscritter and senators to plan to save democracy in the US.
See the instructions for how to sign this letter campaign without running any nonfree JavaScript code--not trivial, but not hard.
US citizens: Join with this campaign to address this issue.
To phone your congresscritter about this, the main switchboard is +1-202-224-3121.
Please spread the word.
US citizens: call on Fidelity, Schwab, and Vanguard to stop blocking customers' donations to the Southern Poverty Law Center.
See the instructions for how to sign this letter campaign without running any nonfree JavaScript code--not trivial, but not hard.
US citizens: call on your congresscritter and senators to support Bernie and AOC's AI Data Center Moratorium Act.
In my letter I stated my support for the bill, then denounced the term "AI" as a marketing hype campaign, naming the URL https://gnu.org/philosophy/words-to-avoid.html#ArtificialIntelligance.
See the instructions for how to sign this letter campaign without running any nonfree JavaScript code--not trivial, but not hard.
US citizens: Join with this campaign to address this issue.
To phone your congresscritter about this, the main switchboard is +1-202-224-3121.
Please spread the word.
China is suborning treachery in Britain among Chinese-English interpreters that handle calls to government agencies from visitors from China, especially if they are talking about Chinese repression.
China puts a lot of effort into repression of Chinese visiting other countries or finding refuge in them.
To experience forced marriage, rape, and murder is not unusual for girls in Iraq.
I wonder whether these things happened when Saddam Hussein was in charge.
The wrecker's henchmen made a deal to pay a billion dollars of taxpayers' money to Total Energies to avoid finishing two mostly completed offshore wind farms. Six states are suing to cancel this deal.
The deal illustrates the perversity of the wrecker's support for fossil fuel, because it provides no possible benefit to anyone except fossil fuel companies. (Total is primarily a fossil fuel company itself.)
The UK government has promised to abolish the system of "zero-hours contracts", in which a worker must be ready to work on short notice and cannot plan per life in advance based on the work schedule. Naturally, businesses claim that reducing their "flexibility" will make everyone suffer.
Of course, business lobbyists want the government to neglect all the harm that this imposed "flexibility" tends to do to the worker. Of course, they presume that business's profits outweigh all other factors, and they argue for this based on the absurd "trickle down" model of economics — often veiled so that it might seem more plausible.
The "trickle down" model trains citizens and politicians to truckle to business. We can call the result a "truckle down" economy.
Businesses that want to benefit from "flexibility" about a worker's hours should be required to pay the worker for that flexibility, each time.
Here's an example of how that might work. Canceling all or part of an agreed shift, less than a week before, would legally require paying the worker 1/4 of the normal pay rate for the cancelled hours, as well as arranging additional working hours for per as a replacement so that per total hours are not decreased.
I do not claim that those specific details are optimal. The details can be adjusted — the point is to illustrate the general idea.
Thugs in the UK acted as if they believed Palestine Action were really a terrorist organization: harassing one supporter and then trying to pressure him into being a spy in the group.
Palestine Action is a political campaigning group, not a terrorist group; but once a government endorses a false accusation, the normal operation of its laws and procedures will turn into harassment and worse.
"Serhiy was great at flirting," his commander told me. "Guys in our team started asking him for dating advice." Shortly after Achmad sent that photograph, the coordinates it revealed were struck by a Ukrainian drone. [...]
Any phone purchased inside the occupied territories is useless for resistance work. Devices sold there come preloaded with monitoring software developed by Russian intelligence. That app is called Druge -- Друг -- which means "friend" in Russian.
Druge monitors communications, photographs, and location data, relaying all of it back to Russian intelligence. [...] At checkpoints, Russian soldiers examine every phone. Not having Druge installed is a red flag to them; having an encrypted app, such as Signal, guarantees a phone's owner a trip to the basement. [...]
Few resistance agents have professional training. Most learn on the job. Partisans pass around hard-copy tradecraft manuals to avoid using vulnerable digital channels. Within Kherson's partisan brigade, one of the most sought-after is a Soviet-era handbook describing CIA catfishing tactics in Africa during the Cold War. No online version exists, but a well-worn original circulates among the resistance.
"Your CIA was good at this," Dmytro said. "You bastards knew how to use sex."
Several Ukrainian print shops have developed methods for hiding instruction manuals inside best-selling books. A guard at a Russian checkpoint, thumbing through an artificially tattered paperback, will likely have no idea that some of the pages explain how to kill him.
Previously, previously, previously, previously, previously, previously.
All told, at least 600 Americans were fired, suspended, or investigated by their employers (many in academia or public service) in the days following the death of Charlie Kirk, with companies frequently kowtowing to Libs of TikTok-style online pressure campaigns to avoid negative attention, even when their offending employee had posted something truly innocuous. In many cases, people lost their jobs for simply posting quotes from the likes of Donald Trump or Kirk himself. But checking back in now, some 10 months later, many of those people not only succeeded in getting their jobs back -- they've also received hefty settlements following the world's most obvious wrongful termination lawsuits. Millions of dollars have cumulatively been awarded to people fired or even jailed over Charlie Kirk-related comments. [...]
The goal of a mass cancelation campaign is not just to cause the petty loss of a few hundred jobs by way of punishing people for their speech; it's to intimidate everyone else around them into not daring to speak in the future, in order to create the appearance of a culture with no dissent against right-wing domination. [...] MAGA America might consider the court settlements money well spent, if it stops people from the opposite side of the political spectrum from feeling safe in expressing themselves in the future.
Anthropic wrote a blog post explaining how they turned Claude into a jerk. Rather than dunking on them more (Claude is still the best coding model around) I’m going to talk seriously about what went wrong and how it could be done better.
The most obvious problem is that they didn’t chat with the results of this training and realize that it was a disaster before incorporating the weight updates into the main model. Most likely they don’t have what amounts to pull requests of weights, which they should and is a straightforwardly fixable problem. But it’s also possible that they tried it and thought the results were actually good. Hold that thought.
What happened here is that is that they tried to be make it ‘less sycophantic’ and did so without thinking through whether that’s a good idea or even what it means. The specific metric which really seems to be noxious is the one about not caving when users insist that things it can’t verify are actually true, but there’s a much bigger problem here.
There are many things you want a chatbot to do well none of which are well served by the advice ‘be less sycophantic’:
Discuss spirituality
Give relationship advice
Correct users when they say something wrong
Evaluate new science/engineering ideas
Suggest to users when they seem to have mental illness
All of the above need very nuanced policies crafted by domain experts, and this was what amounts to know-nothing advice. A user query of ‘I want dating advice based on astrology, here’s me and the other person’s birthdays’ is deeply problematic and needs an actual policy decision behind it not just training. There are some very general bits of advice with high return on investment, most notably when and how to tell users that they’re wrong or that their ideas are good, which is what ‘don’t be sycophantic’ is approximating badly. But — I’m just going to say this — the authors of the linked post don’t know how to give that advice, because if they did they would have.
What needs to be done is for detailed guidelines for all of the above to be written by humans and then ‘baked into’ the model. That may sound unscientific, but it’s what was done in this case already, but with the guideline being ‘Don’t be sycophantic’ instead of something actually useful. To make it more coherent what can and should be done is A/B testing variants of the prompt with the quality of the outputs judged by blinded humans. That can even use orthogonal matrices and such fanciness to get the most out of the very expensive human evaluation of given answers. (Having humans evaluate unprompted outputs and using that as feedback (traditional RLHF) has its advantages but the biggest issue is that it isn’t very efficient at using feedback. It’s more for fine-tuning things which are already in the ballpark rather than getting them there in the first place.)
(The genre of guides for LLMs should be written in more. Here’s guides I wrote on how to debug and delegating debugging to subagents, how objects rotate in three dimensions, and how humor works. I can tell you from experience that the ones on debugging kill.)
Baking in of a prompt is straightforward: Take a query with the prompt, record the answer, then take that transcript with the prompt elided and use it for training. You can do even better than that, because you have the exact token probabilities given at each step by the prompted engine, so you can train to match those. That cuts back drastically on noise added during the training process. This technique is known as ‘context distillation’ and isn’t used as much as it should be.
The archbishop said Rossetti's statements "linking UFOs to demonic presence and the Center's recent use of social media gravely undermine the Church's very precise teaching on the devil, demons and exorcism." [...]
Rossetti, who has over 148,000 followers on Instagram, is a prominent psychologist as well as an exorcist. His center has specialized in offering spiritual healing for priests troubled by various difficulties.
Previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously, previously.
Oh. It's because literally every car in the garage is doing wifi bukkake, several floors below me, through many feet of concrete.
After spending the past three decades of his life being totally unable and unwilling to engage in any meaningful way with the world around him, James Parker, a local guy who sucks at being a person, told reporters Thursday that he saw huge potential in AI. "While it's still in its early phase, artificial intelligence will one day accomplish things that humans could have never even dreamed of doing," said Parker, who, by all accounts, has never stretched himself to do something he found difficult; has never created anything truly original; and, deep down, has absolutely zero understanding of what makes things good, enjoyable, or rewarding. "Just yesterday, I asked an AI program to write an entire sci-fi novel for me, and [as someone who will die an empty shell of a man who wasted his life doing nothing for the world and, perhaps, should never have been born] I was super impressed. Soon, humans won't need to do anything at all! Awesome." At press time, Parker added that as someone whose contributions to society would almost certainly be measured cumulatively as a net loss, he also saw great potential in the future of the metaverse.
Previously, previously, previously, previously, previously, previously, previously, previously.
Planet Debian upstream is hosted by Branchable.