Lawmakers in Congress are moving quickly on the GUARD Act, an age-gating bill restricting minors’ access to a wide range of online tools, with a key vote expected this week. The proposal is framed as a response to alarming cases involving “AI companions” and vulnerable young users. But the text of the bill goes much further, and could require age gates even for search engines that use AI.

If enacted, the GUARD Act won’t just target a narrow category of risky chatbots. It would require companies to verify the age of every user — then block anyone under 18 from interacting with a huge range of online systems. It would block minors from everyday online tools, undermine parental guidance, and force adults to sacrifice their privacy. In the process, it would require services to implement speech-restricting and privacy-invasive age-verification systems for everyone—not just kids.

Under the GUARD Act’s broad definitions, a high school student could be barred from asking homework help tools questions about algebra problems. A teenager trying to return a product could be kicked out of a standard customer-service chat.

Maryland has become the first state in the US to ban surveillance pricing in grocery stores.

Maryland’s law bans grocers and third-party delivery services from using a person’s personal data to set higher prices. Wes Moore, the governor, signed the measure into law on Tuesday. “At a time when technology can predict what we need, when we need it, when we’ll pay for it and also – when we’ll pay more for it, and at a time when we’re watching how big companies are then using these analytics against us to make record profits, Maryland is not just pushing back. Maryland is pushing forward because we are going to protect our people,” Moore said at the bill signing ceremony.

When engaging in surveillance pricing, stores rapidly change the cost of products based on consumer data, including their location, internet search history and demographics. That means buyers are paying different prices for the same items purchased around the same time. Critics of this method – also known as dynamic pricing – say that in doing so, businesses are effectively charging each person the most that they’re willing to pay.

If you're tired of interacting with a bot that spews Nazi propaganda or refers to itself as MechaHitler, you could sign off of Elon Musk's xAI. Or, just to be sure, use an LLM whose training data ends in 1930, three years before the Nazis took power in Germany and nine years before World War II started.

A trio of AI researchers has released a 13-billion-parameter "vintage" language model they call Talkie, which has been trained solely on digital scans of English-language books, newspapers, periodicals, scientific journals, patents, and case law that were published before the end of 1930. Pre-1931 works were chosen because 1930 is the current public domain year in the United States.

In other words, if you're looking for information on World War II, the election of Franklin D. Roosevelt, Amelia Earhart's solo Atlantic flight, or how a microwave oven works, you're out of luck. Ask it about Betty Boop, flappers, the state of the US economy as the Great Depression began, or the sociological effects of the introduction of car radios, and you've come to the right place.

This isn't the first vintage AI model to appear, mind you, with others trained on Victorian literature and pre-1900 scientific texts already out in the world. It is, according to its creators, the largest they are aware of.

These days, the hype is all about AI and robots, but almost a decade ago, the tech du jour was self-driving. You couldn’t swing a lanyard at CES for the latter half of the last decade without hitting a robotaxi; post-COVID, the number of startups has shrunk, but the technology has definitely matured. Go to the right cities—San Francisco and Austin, Texas, spring to mind—and you might see dozens of sensor-festooned vehicles among the downtown traffic.

The pod-like robotaxis belonging to Zoox stand out. Other robotaxi developers are retrofitting existing vehicles like Hyundai Ioniq 5s with sensors and the computing power necessary for self-driving. Zoox, which was bought by Amazon in 2020, did that with its test fleet, but as it starts to offer ride-hailing services—currently in Las Vegas and San Francisco—it’s doing so with a purpose-built design that looks like it just drove off the set of a big-budget sci-fi production.

“A robotaxi is not a car; it’s not a human-driven vehicle, and the requirements are wildly different, although it has to live in that world,” explained Chris Stoffel, director of robot industrial design and studio engineering at Zoox.

It all starts with the sensors, each perched on a little ledge projecting from the top four corners of the robotaxi’s body. From up there, each has an unobstructed, high-level view, giving the Zoox robotaxi good situational awareness, particularly straight ahead. “Because we don’t have a traditional hood, we’ve optimized our frontal coverage in a way that would be nearly impossible on a retrofitted vehicle,” said Zoox director of sensor engineering Ryan McMichaels.

I've yet to see one of these in the wild, so I'm guessing they're geofencing to downtown and UT. Waymos, on the other hand ... I see multiple ones a day, and I don't get out all that much.

Japan’s famously conscientious but overburdened baggage handlers will soon be joined by extra staff at Tokyo’s Haneda airport – although their new colleagues will need to take regular recharging breaks.

Japan Airlines will introduce humanoid robots on a trial basis from the beginning of May, with a view to deploying them permanently as a solution to the country’s chronic labour shortage.

The Chinese-made humanoids will move travellers’ luggage and cargo on the tarmac at Haneda, which handles more than 60 million passengers a year.

JAL and its partner in the initiative, Japan Airlines GMO Internet Group, hope the experiment – which ends in 2028 – will lessen the burden on human employees amid a surge in inbound tourism and forecasts of more severe labour shortages.

In a demonstration for the media this week, a 130cm-tall robot manufactured by Hangzhou-based Unitree was seen tentatively “pushing” cargo on to a conveyer belt next to a JAL passenger plane and waving to an unseen colleague.

A GGUF port of DFlash speculative decoding. Standalone C++/CUDA stack on top of ggml, runs on a single 24 GB RTX 3090, hosts the new Qwen3.6-27B.

~1.98x mean over autoregressive on Qwen3.6 across HumanEval / GSM8K / Math500, with zero retraining.

If you have CUDA 12+ and an NVIDIA GPU like RTX 3090 / 4090 / 5090, then all you need to do is

clone the repo

cd lucebox-hub/dflash
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release
cmake --build build --target test_dflash -j

hf download unsloth/Qwen3.6-27B-GGUF Qwen3.6-27B-Q4_K_M.gguf --local-dir models/

hf download z-lab/Qwen3.6-27B-DFlash --local-dir models/draft/

DFLASH_TARGET=models/Qwen3.6-27B-Q4_K_M.gguf python3 scripts/run.py --prompt "def fibonacci(n):"

Luce DFlash will:

1. Load Qwen3.6-27B Q4_K_M target weights (~16 GB) plus the matched DFlash bf16 draft (~3.46 GB) and run DDTree tree-verify speculative decoding (block size 16, default budget 22, greedy verify).
2. Compress the KV cache to TQ3_0 (3.5 bpv, ~9.7x vs F16) and roll a 4096-slot target_feat ring so 256K context fits in 24 GB. Q4_0 is the legacy path and tops out near 128K.
3. Auto-bump the prefill ubatch from 16 to 192 for prompts past 2048 tokens (~913 tok/s prefill on 13K prompts).
4. Apply sliding-window flash attention at decode (default 2048-token window, 100% speculative acceptance retained) so 60K context still decodes at 89.7 tok/s instead of 25.8 tok/s.
5. Serve over an OpenAI-compatible HTTP endpoint or a local chat REPL.

Running on RTX 3090, Qwen3.6-27B UD-Q4_K_XL (unsloth Dynamic 2.0) target, 10 prompts/dataset, n_gen=256:

Bench AR tok/s DFlash tok/s AL Speedup

HumanEval 34.90 78.16 5.94 2.24x

Math500 35.13 69.77 5.15 1.99x

GSM8K 34.89 59.65 4.43 1.71x

Mean 34.97 69.19 5.17 1.98x

I didn’t know about “mukbangs”—a portmanteau of the Korean words “meongneun” (eating) and “bangsong” (broadcast)—until Enid Frances appeared on my TikTok account’s explore page. Enid, wearing a pink tank top and bright lipstick, taps on the plastic packaging of a chocolate cake. She’s silent, making intense eye contact with us. She lifts the lid. She gulps milk. Her fork enters. The chewing is exaggerated and loud. Two minutes pass, and a quarter of the cake is consumed. Enid flashes a thumbs up.

Videos like Enid’s, seemingly innocuous by nature, have managed to incite international controversy. In 2024, sensationalistic reports surfaced about 24-year-old Pan Xiaoting, who supposedly ate herself to death on air. She promised followers she would consume 10kg of food in one sitting, and was said to have eaten until her stomach tore. Though there was skepticism about the veracity of Xiaoting’s story, additional influencers have since allegedly perished.

China now bans the filming and streaming of mukbangs. The Philippines proposed a similar ban. Meanwhile, in South Korea, a crackdown on public health guidelines to address mukbangs are raising questions about government infringement on freedom of choice. For American viewers, the mukbang is an unregulated and relatively new form of entertainment.

How is this a thing? I rarely enjoy watching people eat across the table.

AI companies are beginning to entertain the possibility that they could cease to exist. This notion was, until recently, more theoretical: A couple of years ago, an ex-OpenAI employee named Leopold Aschenbrenner wrote a lengthy memo speculating that the U.S. government might soon take control of the industry. By 2026 or 2027, Aschenbrenner wrote, an “obvious question” will be circling through the Pentagon and Congress: Do we need a government-led program for artificial general intelligence—an AGI Manhattan Project? He predicted that Washington would decide to go all in on such an effort.

Aschenbrenner may have been prescient. Earlier this year, at the height of the Pentagon’s ugly contract dispute with Anthropic, Secretary of Defense Pete Hegseth warned that he could invoke the Defense Production Act (DPA), a Cold War–era law that he reportedly suggested would allow him to force the AI company to hand over its technology on whatever terms the Pentagon desired. The act is one of numerous levers the Trump administration can pull to direct, or even commandeer, AI companies. And the companies have been giving the administration plenty of reason to consider doing so.

Future bots could help design and carry out biological, nuclear, and chemical warfare. They could be weaponized to take down power grids, monitor congressional emails, and black out major media outlets. These aren’t purely hypothetical concerns: Earlier this month, Anthropic announced it had developed a new AI model, Claude Mythos Preview, capable of orchestrating cyberattacks on the level of elite, state-sponsored hacking cells, potentially putting a private company’s cyber offense on par with that of the CIA and NSA. In an example of Mythos’s power, Anthropic researchers described how the model used a “moderately sophisticated multi-step exploit” to work around restrictions and gain broad internet access, then emailed a researcher—much to his surprise—while he was eating a sandwich in the park.

Washington is getting antsy about the power imbalance. Over the past year, multiple senators have proposed legislation that would order federal agencies to explore “potential nationalization” of AI. Murmurs of possible tactics abound—including more talk within the administration of the DPA after Anthropic’s Mythos announcement, one person with knowledge of such discussions told us. Meanwhile, Silicon Valley is watching carefully. In recent weeks, Elon Musk, OpenAI’s CEO Sam Altman, and Palantir’s CEO Alex Karp have publicly spoken about the possibility of nationalization. Lawyers who represent Silicon Valley’s biggest AI firms are paying attention.

Worth noting, later in the story it's pointed out why full nationalization is vanishingly unlikely, but more federal oversight is likely.