A report covered by Sifted warns that Mistral and other open-weight developers are increasingly at risk of having safety fine-tuning stripped from their models.
According to a report covered by Sifted, Mistral and other developers of open-weight models are increasingly at risk of 'abliteration'. The term refers to the removal of safety fine-tuning, a process that can strip away the guardrails built into a model.
The report's warning highlights a persistent challenge for the open-model business: balancing accessibility with safety. For companies like Mistral, this risk may become a more important factor in how they design and release their models.
The streaming series dramatizes the rise and fall of Sam Bankman-Fried and Caroline Ellison, bringing the FTX collapse to a mainstream audience.
Microsoft is replacing per-user OneDrive storage with smaller shared pools, raising costs for subscribers who use the most space.
Linux creator says AI makes programming more enjoyable for beginners but warns that generated patches can overwhelm maintainers.
New research on Hawaii's Kilauea volcano helps explain what powers lava fountains and shows that summit tilt measurements can predict when they will occur.