Tech Radar

A mysterious GitHub account recently unleashed a treasure trove of undisclosed 0-day vulnerabilities, and it’s sending ripples through the developer community. This isn’t just another security report; it raises some serious questions about ethics in the world of software vulnerabilities. Who’s behind this account? Are they doing the community a favor or just throwing a wrench into the works? The repo itself is a mixed bag. Some findings are more polished than others—looking at you, Ghidra—but there are definitely gems hidden in there that could shake things up. I went through the code, and while I automated my fuzzing workflow using AI, I still found myself wrestling with the results. It’s wild how much we rely on tools like 5-3-Codex-Spark to do the heavy lifting, but you can’t help but wonder if we’re losing our edge in the process. This release isn’t just a curiosity; it’s a call to action. I’m motivated to keep unearthing findings and sharing them with you. There’s a lot t...

We've spent the last two years obsessing over FLOPS and H100 clusters, but the real bottleneck for LLMs isn't actually compute. It's memory. The GPU spends most of its time just waiting for weights to move from memory to the cores. It's a massive waste of silicon. DSpark tries to fix this by flipping the script. Instead of blindly crunching the next token, it uses a tiny draft model to essentially guess what's coming next. If the guess is right, the big model just verifies it and moves on. It's a clever bit of speculation that treats the LLM more like a judge and less like a typewriter. The results are interesting, but it raises a question about the architecture we've settled on. If a tiny model can predict the output of a giant one with high accuracy, are we just over-provisioning our inference for the sake of a few edge cases? The Memory Bandwidth Bottleneck Autoregressive decoding is slow because it's a memory bandwidth problem, not a comp...

Most model updates are just a race to see who can brag about the highest benchmark score. GPT-5.6 Sol is different. Instead of chasing a higher number on a leaderboard, the focus here is on hardening the safety stack. OpenAI spent weeks trying to break their own system, pressure-testing it against real-world attacks and hunting for weaknesses in how it handles sensitive cyber requests. It's a pragmatic move. We've reached a point where "mostly safe" isn't good enough when a model is being used for high-risk activity or targeted misuse. I'm not saying the system is perfect, but the adversarial approach to this release is a shift in priority that we actually need to see. The Sol model is the flagship, but it's arriving with two siblings. Terra is meant for general daily work, and Luna is the budget option, which claims to be twice as cheap as the previous version. They're all hitting general availability in the next few weeks. The real question...

Om Malik didn't just report on the tech industry. He helped define how we actually analyze the intersection of business, software, and the web. He had a way of cutting through the noise that felt rare even back when the blogosphere was new. He passed away on June 24, 2026, at Stanford Hospital. He'd been dealing with heart issues for a long time, and he was surrounded by family and friends when it happened. It's a heavy loss for anyone who cares about the history of how we talk about technology. I'm still processing what his absence means for the way we cover this beat. There are a few things he did better than anyone else, and it's worth looking at why those things actually mattered. The Malik Method Om Malik's approach to tech blogging is a shift in how we consume industry news. Before he started, tech journalism was mostly a relay race of corporate press releases. Reporters took a company's claims at face value and published them. Malik stoppe...

Most developers treat their choice of ML framework like a religious conversion. They start with TensorFlow because it's what the tutorials use, move to PyTorch because the tensors actually make sense, and then eventually find themselves squinting at JAX because they want to do something truly weird with gradients. It's a predictable cycle of chasing the cutting edge, but the actual shift in momentum is harder to pin down than the anecdotes suggest. I decided to stop guessing and just look at the data. By overlaying search trends from Hacker News, you can see the exact moment the community's collective brain shifted. TensorFlow owned the early gold rush, but PyTorch didn't just nudge it aside, it completely gutted it in the research labs between 2019 and 2021. Now, JAX is the new darling for anyone doing heavy lifting in 2023. The most interesting part isn't the tools, though. It's the players. For a long time, OpenAI's dominance looked like a forego...

We're finally reading texts that have been physically inaccessible for two thousand years. The trick isn't unrolling the scrolls, but seeing through them. For centuries, the carbonized library of Herculaneum has held a cruel bargain. These scrolls survived the eruption of Mount Vesuvius, but they did so by becoming essentially charcoal. They're so fragile that trying to open one is basically an act of destruction. You can't just peel back the layers without turning the whole thing into dust. The latest preprint on the virtual unwrapping of these papyri is a strange victory for computer vision. By using high-resolution CT scans and machine learning to detect ink that's invisible to the human eye, we're bypassing the physical decay entirely. It's a clever workaround, though I suspect the actual translation of these fragments will be a much slower, more frustrating process than the tech makes it seem. The real question is whether the content of these ...