Klarna and StubHub are delaying plans to go public after President Donald Trump's sweeping tariff announcement sent shock waves through the U.S. markets.

The Fast Company Impact Council is an invitation-only membership community of leaders, experts, executives, and entrepreneurs who share their insights with our audience. Members pay annual dues for access to peer learning, thought leadership opportunities, events and more.The way we produce and consume food is changing. Not only is the current food system a threat to our health, its also a threat to our planet. As a food producer, the challenge is clear: How do we transition toward more nutrient-dense, environmentally responsible food choices without compromising taste or accessibility?Modern food production has often emphasized convenience, leading to highly processed products that lack substance and sustenance. However, traditional diets, such as the Mediterranean diet, offer a time-tested solution. Rooted in whole, plant-based ingredients, these diets highlight a variety of grains, legumes, and plant-based proteins that provide essential nutrients while reducing the environmental footprint of food production. By returning to these principles, we can create a more sustainable and nutritious future.Learn from traditional dietsFor centuries, Mediterranean communities have thrived on ingredients that not only support long-term health but also align with sustainable farming practices. Unlike modern industrial agriculture, which prioritizes monoculture crops and mass production, traditional food systems accentuate biodiversity and soil regeneration.Ancient grains like buckwheat are regaining popularity for their rich nutrient profiles and minimal environmental impact. Legumes, such as chickpeas and lentils, are packed with plant-based protein, fiber, and essential minerals while also playing a crucial role in the rise of sustainable practices in modern agriculture. We need to take inspiration from these traditional approaches and champion minimally processed ingredients for the sake of human health and ecological stability.Interestingly, many Americans report that while traveling in Italy, they experience fewer food intolerances and improved digestiondespite indulging more during their vacations. This may be attributed to Italian cuisines deep-rooted tradition and propensity toward whole, minimally processed foods and traditional dishes made with simple, high-quality ingredients.Innovation to address modern challengesTraditional food systems are not the sole answer to todays complex issues of food security, climate change, and health. We must also leverage technology, research, innovation, and a little bit of creativity.The use of spirulina is the perfect example of how innovation can build on traditional principles to create more sustainable and nutrient-dense food options. Called the food of the future by the Food and Agriculture Organization of the United Nations, spirulina is a blue-green algae with superfood benefits which has been consumed for centuries, providing a rich source of protein, vitamins, and minerals. What makes it especially promising for the future of food is its minimal environmental footprintit requires very little water, land, and energy to produce. It also contributes to the reduction of greenhouse gasesproduction can be carbon neutral or even carbon negative, as the growing cells sequester CO2.Many brands are now incorporating ingredients like spirulina into their product offerings, from snacks to beverages and even in more unexpected areas like pasta. For example, at Felicia we source Italian Apulia Kundi spirulina directly from our spirulina park at the Italian headquarters, a facility dedicated to growing these precious microalgae by using the water resulting from the pasta-making process. The water is purified and reused to make the pasta, fostering a virtuous circular economy.This type of sustainable production is a crucial step in our shared journey toward a healthier planet and population. But for innovation like this to really make an impact, it needs to happen at every stage, from food production to product design and consumer education. Creating variety in ones diet doesnt have to be complicatedit can be as simple as incorporating diverse grains, legumes, and vegetables into daily meals and building a colorful plate to delight the senses, without compromising on taste. As brands, we hold a responsibility to our consumers to create the products that make this possible.By embracing both traditional wisdom and modern innovation, with renewed passion, creativity and care, we can reach a thriving future.Carlo Stocco is managing director, North America at Felicia and Andriani.

Back in the day, when you needed a portable light source to carry with you around your house, the most practical one would be a flashlight or a small portable lamp. But in todays dynamic world, versatility and convenience are important and so were seeing a lot of different products entering the market. Wireless and portable are key words that you are probably looking for with this kind of product that you need around the house. Designer: FOSHAN ELECTRICAL AND LIGHTINGThe FSL Wireless Portable Desk Lamp perfectly embodies the qualities you may be looking for, as its name says. This lamp isnt just a source of light; its a stylish, adaptable companion for your workspace, bedside table, or even your outdoor patio. It has an elegant and modern design with a sleek appearance and a minimalist design. It has a cylindrical design with a flat head that incorporates the lamps innovative reflective technology for a uniform and wide-area lighting effect. The cordless and portable design as well as the built-in rechargeable battery liberates you from the constraints of power outlets. This freedom allows you to place the light exactly where you need it most, whether youre working on a computer, engrossed in a book, or enjoying a quiet evening outdoors. The lightweight design makes it easy to carry around and is a versatile lighting solution for various spaces, both inside and out. It also has a USB-C charging port for when the battery runs out, although you should be able to use it for several hours on a single charge. This portable lamp can be used for a lot of things, like as a reading light for wherever in your space you want to read or as an ambient lighting to create a cozy atmosphere in your living room or dining area. It can also be used outdoors to illuminate patios, balconies, or outdoor dining spaces. You can also use it as an emergency light during power outages, as a more elegant version of a flashlight. Whether you need focused task lighting or a gentle ambient glow, this portable lamp offers the freedom and flexibility to illuminate your world, your way. Its thoughtful design, coupled with its wireless portability and adaptable lighting options, makes it a versatile and indispensable tool for anyone seeking convenient and stylish illumination in various aspects of their daily life. Say goodbye to cumbersome cords and hello to the freedom of light, wherever you need it most.The post Wireless portable lamp illuminates your space when and where you need it first appeared on Yanko Design.

3D art of the week: Guochao Zho

MacworldApples AirTag does a fine job tracking stuff but it just cant fit into a wallet. Thats where this slim,credit card-sized trackercomes in. It works with Apples Find My app just like an AirTag, but without the bulky round shape poking out of your pocket.Its called theKeySmart SmartCard, and it offers a sleek alternative to Apples tracker, complete with Qi wireless charging, a built-in lanyard slot, and a higher waterproof rating. Right now, you can grab a 3-pack for $89.99 with free shipping $30 each and about the same price as the AirTag, but with a wallet-friendly design that may suit your needs better.This smart card slips right into your wallet and shows its location on Apples Find My network. Youll get helpful features like left behind alerts, audible location pings, and real-time tracking across millions of Apple devices. Bonus: While AirTags require additional purchases for keychains or straps, this tracker already has a slot for lanyards, badge reels, or anything else you want to clip it to.You can get three of theseAirTag alternativesfor $89.99 with free shipping. Keep one for yourself and share the others in Easter baskets, or use the others in your passport, luggage, pet carriers, or anywhere else you need tracking.KeySmart SmartCard Works with Apple Find My (3-Pack)See DealStackSocialprices subject to change.

Nintendo formally unveiled Mario Kart Worldat its Switch 2 Direct earlier in the week, and though the wait for aMario Kart 8successor has been a long one, its not going to drag on much longer.Mario Kart Worldwill release in June as a Switch 2 launch title, and ahead of its release, Nintendo isnt being stingy with the amount of gameplay footage it is showcasing.Hot on the heels of extensive gameplay showcased during Thursdays Treehouse livestream, Nintendo followed up today with more of the same. Close to 40 minutes of new gameplay footage have been revealed forMario Kart Worldin total, showing off more of its Grand Prix mode, some of the many tracks and characters that will be featured in the game, how it will incorporate GameChat features, and more. You can check out the gameplay below.Mario Kart Worldwill launch exclusively for the Nintendo Switch 2 on June 5. The game will be sold at a price of $79.99.

(Feed generated with FetchRSS)

President Donald Trumps additional 75 day delay to TikToks sale-or-ban deadline leaves service providers like Apple, Google, and Oracle on shaky ground, and, according to one influential Democrat, is straight-up against the law.After Trump announced the extension on Friday, 12 Republican members of the House Select Committee on China, including Chair John Moolenaar (R-MI), released a joint statement in response. The statement did not address legal concerns with the second extension, but it said that any resolution must ensure that U.S. law is followed, and that the Chinese Communist Party does not have access to American user data or the ability to manipulate the content consumed by Americans. The letter says signatories look forward to more details on a proposed deal.In a separate statement, three Republican members of the House Energy and Commerce Committee, including Chair Brett Guthrie (R-KY) struck a similar note, saying that, any deal must finally end Chinas ability to surveil and potentially manipulate the American people through this app.The whole thing is a sham if the algorithm doesnt move from out of Beijings handsSenate Intelligence Committee Vice Chair Mark Warner (D-VA) was more critical in a phone interview with The Verge. The whole thing is a sham if the algorithm doesnt move from out of Beijings hands, Warner said. And close to 80 percent of Republicans knew this was a national security threat will they find their voice now?Trump signed an executive order on his first day in office delaying enforcement of the TikTok divestiture law, a move legal experts already found questionable. Then, he failed to announce a deal before the new April 5th deadline amid chaos over new global tariffs. Letting the delay expire would have put US companies that serviced TikTok after the deadline at even greater risk of hefty penalties.The original Protecting Americans from Foreign Adversary Controlled Applications Act was passed with overwhelming bipartisan support to address what lawmakers insisted was a pressing national security threat, then upheld by the Supreme Court in January. TikTok has long denied that the Chinese government could access US user data or put its thumb on the scales of the recommendation feed through ByteDance, but many lawmakers have consistently doubted that defense. As the Trump administration has opted to effectively ignore the law, however, Congress has been relatively quiet.Trumps unilateral extension is illegal and forces tech companies to once again decide between risking ruinous legal liability or taking TikTok offlineA few Senate Democrats, including Ed Markey (D-MA), recently warned Trump that another extension would only introduce more legal uncertainty, and some expressed doubt that some of the reported deal scenarios could even resolve the apps legal concerns. In a statement after Trumps second extension, Markey says while hed like to see the deadline pushed, Trumps unilateral extension is illegal and forces tech companies to once again decide between risking ruinous legal liability or taking TikTok offline. He called the move unfair to those companies and unfair to TikToks users and creators. Instead, Trump should go through Congress to pass Markeys bill to the extend the deadline, he says. Rep. Ro Khanna (D-CA), a member of the China Committee whos criticized the law and warned it will harm free expression and creators livelihoods, also wants to see a solution go through Congress, but is seeking a full repeal of the law. Still, he called Trumps delay a good step.The new statements from China Committee and E&C Republicans appear to be the first coordinated moves to put a firm line in the sand on the topic. Some Republicans who support the divest-or-ban law have previously urged Trumps compliance in one-off statements or writings. Moolenaar previously warned in an op-ed that an adequate deal must fully break ties with ByteDance after reports that Trump was considering a deal with Oracle that would potentially leave some ties intact. Sen. Josh Hawley (R-MO) told reporters earlier this week that if a deal doesnt comply with the statute, he would advise the President against it. If he cant get a deal to sell the company in a way that fully complies, Hawley thinks Trump ought to enforce the statute and ban TikTok. This middle way, I dont think is viable.Warner maintains that lawmakers want a TikTok sale that keeps the app in the US, and he says the Biden administration should have been more aggressive in getting negotiations started. He remains concerned that TikToks ownership structure could allow a foreign adversary government to influence young Americans. During the negotiations, we saw the enormous bias in TikTok on things like the Uyghurs, the Hong Kong protests, the conflict in Gaza, says Warner. That was how we got 80 percent of the vote. Warner says he remains concerned about the security of US user data, but sees the potential for TikTok to be used to shape public opinion as the more serious threat. Still, lawmakers seem unlikely to do much beyond (maybe) trying to pass a new law should Trump continue to flout the existing one. Congress, says Hawley, we dont have an enforcement arm of our own.See More:

Sparse autoencoders are central tools in analyzing how large language models function internally. Translating complex internal states into interpretable components allows researchers to break down neural activations into parts that make sense to humans. These methods support tracing logic paths and identifying how particular tokens or phrases influence model behavior. Sparse autoencoders are especially valuable for interpretability applications, including circuit analysis, where understanding what each neuron contributes is crucial to ensuring trustworthy model behavior.A pressing issue with sparse autoencoder training lies in aligning training objectives with how performance is measured during model inference. Traditionally, training uses mean squared error (MSE) on precomputed model activations. However, this doesnt optimize for cross-entropy loss, which is used to judge performance when reconstructed activations replace the originals. This mismatch results in reconstructions that perform poorly in real inference settings. More direct methods that train on both MSE and KL divergence solve this issue, but they demand considerable computation, which limits their adoption in practice.Several approaches have attempted to improve sparse autoencoder training. Full end-to-end training combining KL divergence and MSE losses offers better reconstruction quality. Still, it comes with a high computational cost of up to 48 higher due to multiple forward passes and lack of activation amortization. An alternative involves using LoRA adapters to fine-tune the base language model around a fixed autoencoder. While efficient, this method modifies the model itself, which isnt ideal for applications that require analyzing the unaltered architecture.An independent researcher from Deepmind has introduced a new solution that applies a brief KL+MSE fine-tuning step at the tail end of the training, specifically for the final 25 million tokensjust 0.510% of the usual training data volume. The models come from the Gemma team and Pythia project. It avoids altering the model architecture and minimizes complexity while achieving performance similar to full end-to-end training. It also allows training time savings of up to 90% in scenarios with large models or amortized activation collection without requiring additional infrastructure or algorithmic changes.To implement this, the training begins with standard MSE on shuffled activations, followed by a short KL+MSE fine-tuning phase. This phase uses a dynamic balancing mechanism to adjust the weight of KL divergence relative to MSE loss. Instead of manually tuning a fixed parameter, the system recalculates the KL scaling factor per training batch. The formula ensures the total combined loss maintains the same scale as the original MSE loss. This dynamic control prevents the need for additional hyperparameters and simplifies transfer across model types. Fine-tuning is executed with a linear decay of the learning rate from 5e-5 to 0 over the 25M token window, aligning the process with practical compute budgets and preserving sparsity settings from earlier training.Performance evaluations show that this approach reduced the cross-entropy loss gap by 20% to 50%, depending on the sparsity setting. For example, on Pythia-160M with K=80, the KL+MSE fine-tuned model performed slightly better than a full end-to-end model, requiring 50% less wall-clock time. At higher sparsity (K=160), the fine-tuned MSE-only model achieved similar or marginally better outcomes than KL+MSE, possibly due to the simplicity of the objective. Tests with LoRA and linear adapters revealed that their benefits do not stack, as each method corrects a shared error source in MSE-trained autoencoders. Even very low-rank LoRA adapters (rank 2) captured over half the performance gains of full fine-tuning.Although cross-entropy results consistently favored the fine-tuned method, interpretability metrics showed mixed trends. On SAEBench, ReLU-based sparse autoencoders saw improvements in sparse probing and RAVEL metrics, while performance on spurious correlation and targeted probe tasks dropped. TopK-based models showed smaller, more inconsistent changes. These results suggest that fine-tuning may yield reconstructions better aligned with model predictions but may not always enhance interpretability, depending on the specific evaluation task or architecture type.This research underscores a meaningful advancement in sparse autoencoder training: a computationally light, technically simple method that improves reconstruction accuracy without modifying base models. It addresses key alignment issues in training objectives and delivers practical results across models and sparsity levels. While not uniformly superior in all interpretability metrics, it offers a favorable trade-off between performance and simplicity for tasks like circuit-level analysis.Check outthe Paper.All credit for this research goes to the researchers of this project. Also,feel free to follow us onTwitterand dont forget to join our85k+ ML SubReddit. NikhilNikhil is an intern consultant at Marktechpost. He is pursuing an integrated dual degree in Materials at the Indian Institute of Technology, Kharagpur. Nikhil is an AI/ML enthusiast who is always researching applications in fields like biomaterials and biomedical science. With a strong background in Material Science, he is exploring new advancements and creating opportunities to contribute.Nikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Introduces FASTCURL: A Curriculum Reinforcement Learning Framework with Context Extension for Efficient Training of R1-like Reasoning ModelsNikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Unveils a Reverse-Engineered Simulator Model for Modern NVIDIA GPUs: Enhancing Microarchitecture Accuracy and Performance PredictionNikhilhttps://www.marktechpost.com/author/nikhil0980/The Complete Beginners Guide to Terminal/Command PromptNikhilhttps://www.marktechpost.com/author/nikhil0980/How to Use Git and Git Bash Locally: A Comprehensive Guide