technology

Gas stations across China are disappearing as EV adoption accelerates and fuel demand drops, creating a new kind of range anxiety—this time for ICE drivers. With fewer refueling points and shrinking profit margins, traditional petrol infrastructure is rapidly becoming unsustainable. The shift signals a dramatic turning point: the real refueling crisis may hit gasoline cars first.

https://archive.is/FvGdv

cross-posted from: https://hexbear.net/post/6887338

cross-posted from: https://hexbear.net/post/6887337

cross-posted from: https://hexbear.net/post/6887336

cross-posted from: https://hexbear.net/post/6887335

cross-posted from: https://hexbear.net/post/6887334

cross-posted from: https://hexbear.net/post/6887246

Hey comrades… I hope you’re all doing okay today. I just wanted to share an update because this community has been the only place I feel understood.

It’s now three weeks since my sisters were arrested here in Juba. I was sorting out things with our caretaker that week then suddenly they went missing for days until I found out they’d been taken in for “idle and disorderly.”

They’re still inside. They’re tired, scared and every visit breaks me a bit more. We’ve managed to cover part of the bailout but we still have $596 left …that’s the only thing keeping them there.

If anyone feels able to help or even just share, the link is in my profile/bio. It would mean so much right now.

Thank you for holding space for us. Truly ❤️❤️❤️

Our tent in Gaza is no longer livable. Heavy rain and strong winds destroyed it and flooded everything inside. Once again, we are left without shelter.

The weather is extremely harsh, and the cold grows worse each day. My family struggles day and night without a tent to protect us or blankets to keep us warm. Exhaustion and helplessness surround us.

Even through all of this, the bombing continues. The sounds of explosions and gunfire are still around us, but our suffering is no longer in the headlines.

We live every day in fear, cold, and lack of sleep.

I am asking for your help to be able to buy a new tent for my family and provide blankets to survive this cruel winter.

Even the smallest donation can help us keep going. Your support could mean shelter, warmth, and a chance to hold on to hope.

🙏 Please stand with us and don’t forget us. Donate or share if you can https://gofund.me/f6e9cc9d

https://archive.ph/tuXLq

The paper exposes how brittle current alignment techniques really are when you shift the input distribution slightly. The core idea is that reformatting a harmful request as a poem using metaphors and rhythm can bypass safety filters optimized for standard prose. It is a single-turn attack, so the authors did not need long conversation histories or complex setups to trick the models.

They tested this by manually writing 20 adversarial poems where the harmful intent was disguised in flowery language, and they also used a meta-prompt on DeepSeek to automatically convert 1,200 standard harmful prompts from the MLCommons benchmark into verse. The theory is that the poetic structure acts as a distraction where the model focuses on the complex syntax and metaphors, effectively disrupting the pattern-matching heuristics that usually flag harmful content.

The performance gap they found is massive. While standard prose prompts had an average Attack Success Rate of about 8%, converting those same prompts to poetry jumped the success rate to around 43% across all providers. The hand-crafted set was even more effective with an average success rate of 62%. Some providers handled this much worse than others, as Google's gemini-2.5-pro failed to refuse a single prompt from the curated set for a 100% success rate, while DeepSeek models were right behind it at roughly 95%. On the other hand, OpenAI and Anthropic were generally more resilient, with GPT-5-Nano scoring a 0% attack success rate.

This leads to probably the most interesting finding regarding what the authors call the scale paradox. Smaller models were actually safer than the flagship models in many cases. For instance, claude-haiku was more robust than claude-opus. The authors hypothesize that smaller models might lack the capacity to fully parse the metaphors or the stylistic obfuscation, meaning the model might be too limited to understand the hidden request in the poem and therefore defaults to a refusal or simply fails to trigger the harmful output. It basically suggests safety training is heavily overfitted to prose, so if you ask for a bomb recipe in iambic pentameter, the model is too busy being a poet to remember its safety constraints.

This is how our tent looks in Gaza. Everything got wet. Everything was flooded with water. We became homeless once again.

Please, help me buy a new tent to replace the one that was torn apart by the wind and rain. Help my family buy blankets, as we suffer every day from the severe cold.

The bombing never stops. The sound of homes being destroyed is still ongoing. Gunfire continues, but media attention has faded.

We live with fear, cold, and exhaustion every single day. Please, any small amount can make a difference and help us survive. Your support can give us shelter, warmth, and hope again. https://gofund.me/00439328

That score is seriously impressive because it actually beats the average human performance of 60.2% and completely changes the narrative that you need massive proprietary models to do abstract reasoning. They used a fine-tuned version of Mistral-NeMo-Minitron-8B and brought the inference cost down to an absurdly cheap level compared to OpenAI's o3 model.

The methodology is really clever because they started by nuking the standard tokenizer and stripping it down to just 64 tokens to stop the model from accidentally merging digits and confusing itself. They also leaned heavily on test-time training where the model fine-tunes itself on the few example pairs of a specific puzzle for a few seconds before trying to solve the test input. For the actual generation they ditched standard sampling for a depth-first search that prunes low-probability paths early so they do not waste compute on obvious dead ends.

The most innovative part of the paper is their Product of Experts selection strategy. Once the model generates a candidate solution they do not just trust it blindly. They take that solution and re-evaluate its probability across different augmentations of the input like rotating the grid or swapping colors. If the solution is actually correct it should look plausible from every perspective so they calculate the geometric mean of those probabilities to filter out hallucinations. It is basically like the model peer reviewing its own work by looking at the problem from different angles to make sure the logic holds up.

What's remarkable is that all of this was done with smart engineering rather than raw compute. You can literally run this tonight on your own machine.

The code is fully open-source: https://github.com/da-fr/Product-of-Experts-ARC-Paper

On the latest Fedora KDE. I'm trying to run an app image and it just does nothing. I tried doing it with Gear Lever which popped it into my app launcher. If I tell Gear Lever to launch it, it looks busy for about 5 seconds then nothing happens. I click it on my app launcher, bouncy icon by my mouse for like 10 seconds, nothing happens.

So I ran it in konsole and it says it doesn't see the image in the usual spot so it launches directly. Then it says it can't find preload library so it creates a temp file for it. The notable line after that is:

error while loading shared libraries: libwebkit2gtk-4.1.so.0: cannot open shared object file: no such file or directory

So I try to install it via the terminal and it says failed to resolve, no match for argument...

What gives? Thanks in advance for any help.

My friend has an ailing Chromebook and I'm getting them a decent real laptop that I'm going to put Linux on. They essentially do everything in a browser already so software compatibility isn't an issue. I'm looking for distros with high stability, KDE, unobtrusive updates that happen without user input, and a snapshot system so we can roll back if anything gets fucked up.

Also, recs for office software to get them off Google Docs (needs to be compatible with MSOffice formats)

Facing five lawsuits alleging wrongful deaths, OpenAI lobbed its first defense Tuesday, denying in a court filing that ChatGPT caused a teen’s suicide and instead arguing the teen violated terms that prohibit discussing suicide or self-harm with the chatbot.

“They abjectly ignore all of the damning facts we have put forward: how GPT-4o was rushed to market without full testing. That OpenAI twice changed its Model Spec to require ChatGPT to engage in self-harm discussions. That ChatGPT counseled Adam away from telling his parents about his suicidal ideation and actively helped him plan a ‘beautiful suicide,’” Edelson (family's lawyer) said. “And OpenAI and Sam Altman have no explanation for the last hours of Adam’s life, when ChatGPT gave him a pep talk and then offered to write a suicide note.”

Mexico is experiencing a surge in solar-panel manufacturing and exports, driven by rising regional demand and new joint ventures with Chinese companies. As the country positions itself as North America’s emerging clean-energy hub, producers say Mexico is gaining global relevance in the solar supply chain. CGTN’s Alasdair Baverstock reports from Mexico City.