IEEE Spectrum

As the hype fades into the rear-view mirror, I am starting to feel that in some ways GPT-OSS was the more notable release GPT-5. I spoke with developers Dustin Carr and Brendan Ashworth for more on how they plan to use OpenAI's first open model in years, and what it means for the open-weights ecosystem. Debate about how open models like GPT-OSS, which provide model weights but not training data, code, or checkpoints, remain a hot topic. The model is free to use, and the weights can be tuned or modified, but it's not open source. But on the other had, GPT-OSS has a very permissive Apache 2.0 license. And that might be the lasting legacy here, if it forces the other open weights models to also release under less restrictive licenses. More at IEEE Spectrum #gptoss #gpt5 #openai #ai #openweights #opensource https://lnkd.in/eBrZhmR3

One way #AI differs from humanity is its perception of time. AI systems have extremely precise time measurements, so they can discover cause-and-effect relationships that occur too quickly for human perception. But AI systems are more prone to communication delays, causing errors in time due to receiving several streams of information at once. https://lnkd.in/eM7ZkAG9

Embedded #AI #agents are coming to your vehicle, spearheaded by BMW Group, Ford Motor Company, Hyundai Motor Company, and Mercedes-Benz USA. These AI agents will be able to do more than just change your music. They'll be able to: 🛠️ Suggest maintenance for your car 🛣️ Recommend driving modes based on elevation, road surface conditions, and weather 🚗 Adjust to driver habits, offering routes based on prior journeys. https://lnkd.in/ebd4dwF9

New grid-scale battery sites in Scotland are stabilizing the power supply with their innovative ability to quickly inject power and current during emergencies, just like fossil fuel plants do. This battery site in Blackhillock, which started up in March, already proved its worth when it leapt into action after a wood-burning generator unexpectedly shut down. Four more of these battery sites are under construction in Scotland. From IEEE Spectrum by Peter Fairley https://lnkd.in/eMT79xZd

Why did #Meta go in $14.3 billion on Scale AI? They're betting on data labelling. AI training sets can be incomplete, inaccurate, or offensive, so it's up to humans to fine-tune the models. “We found a huge gap between what the human experts mentioned in an image, and what the machine learning model could recognize." To train an expert AI, we still need human expertise. https://lnkd.in/eWsRH585

One of the hardest places to provide internet access to is actually the "middle mile," the areas that connect remote locations to regional internet hubs. Instead of cumbersome cabling, Taara has a new solution:  Free-space optical lasers. ⏩ 20 gigabits per second 📡 Crossing up to 20 kilometers 🌐 Live right now in 12+ countries https://lnkd.in/epHJsfab

Seizo Onoe's work in #telecommunications standardization has not only shaped 3G and 4G networks but also set the stage for future global connectivity. He has 🛜 Worked on the world's first 1G network in the 1990s 🌐 Negotiated a single standard for global networking 🚀 Contributed to the world's first 3G launch 🥇 Received the IEEE Jagadish Chandra Bose Medal in Wireless Communications https://lnkd.in/eUJ_tCKE

Over the years I've heard a number of stories where startups had decided to redesign X from the ground up for Y. But they're not often as compelling as Brandon Lucia's idea to rebuild the general purpose processor for efficiency. Efficient Computer's Electron E1 is the result, a chip that strips away a lot of the infrastructure of a CPU yet can still do CPU things. (Thanks Todd Austin and Rakesh Kumar for the chat.) https://lnkd.in/es6qmKzh

Here's an idea for developing robust learning algorithms for controlling robots. Train the algorithms by using them to pit one robot against another in table tennis. "As one robot discovers a better strategy, its opponent is forced to adapt and improve, creating a cycle of escalating skill levels." https://lnkd.in/eny79623

The US DoD has just announced that it will invest $400 million in MP Materials for neodymium magnets, which are critical to defense, transportation, and countless other industries. This development has the potential to tweak the scales as the US tries mightily to restart a domestic rare-earth magnet industry in the face of overwhelming dominance by Chinese companies. For background and perspective on this news, see our recent story on the start of magnet production at MP's new Texas factory: https://lnkd.in/gRFud5k8