AI

Microsoft AI chief Mustafa Suleyman. | Getty Images For years, AI leaders have touted healthcare as one of the clearest examples of where their technology will improve lives. Now "it is finally here," Microsoft AI CEO Mustafa Suleyman tells The Verge. The company claims to have developed an AI-powered tool it says can diagnose conditions up to four times more accurately than human doctors. The findings, published Thursday in one of the world's most prestigious journals, Science, are based on tests pitting Microsoft's AI system against human physicians on a series of scenarios drawn from medical case reports and emergency department data. Microsoft compiled the cases into its own benchmark, S … Read the full story at The Verge.

Four projects for facilitating AI agents move under vendor-neutral governance

arXiv:2610.31366v1 Announce Type: new Abstract: Many LLM agents maintain external persistent memory and derive additional state from past interactions, such as summaries. Deleting an original memory record therefore need not remove information carried by its descendants. We introduce AfterDeleteBench, a controlled behavioral audit for this failure mode in fixed-weight LLM agents. For each synthetic secret, we compare an exposed-then-deleted agent with an otherwise matched agent that was never exposed to the secret, allowing post-deletion recovery to be attributed to the original exposure. In our primary setting, a single query recovers the deleted secret from Qwen2.5-7B-Instruct in roughly two thirds of cases, while matched never-exposed controls have zero recovery. Crucially, recovery does not require a surviving verbatim copy: when the complete secret string is absent and only separately stored components remain under a public composition rule, exact reconstruction succeeds in 62-99% of cases across Qwen, Mistral, and Phi-3. We also show that the response interface is part of the measurement system: target-bearing replies can be scored as failures when the requested format and pa…

arXiv:2610.30678v1 Announce Type: cross Abstract: Safety probes trained on a model's activations are a standard tool for monitoring language models in real time. Each time a new model is released or a new harm area needs to be monitored, the probe must be retrained, which requires new labels and a choice of probe architecture. Yet neither the model nor the labels tell us in advance how well the resulting probe will perform. Finding out requires training and comparing many probe configurations. In this paper, we ask whether probe performance can instead be predicted in advance from model size and the number of labels. Existing safety datasets are insufficient for this, as they are either limited in scale or already saturated. We therefore construct six new safety datasets of sufficient size and difficulty. On these datasets, we train two probe architectures on 20 base models from five model families, with 32 to $10^{6}$ labels for each model. We measure performance using probe error, defined as 1 - AUROC, where lower values are better. We find that probe error follows two simple scaling laws. First, the lowest probe error a model can reach falls as a power law in its size. Second, p…

arXiv:2504.21323v4 Announce Type: replace Abstract: Knowledge distillation is widely used to transfer behavior from a large teacher model to a smaller student. It is often assumed to be safe when the teacher is clean, because classic backdoor attacks rely on poisoned labels and triggers in supervised training, whereas distillation trains the student to match a teacher's outputs. We show that this assumption can fail when the distillation dataset itself is poisoned. Our attack injects triggered and manipulated images that a clean teacher already predicts as an attacker chosen target label, which causes the student to learn a backdoor even though the teacher remains unaffected. We evaluate this threat across multiple manipulation strategies, including targeted adversarial perturbations and targeted GAN based class transitions, and study how distillation settings influence both accuracy and attack success. We show that the attack remains effective even at a 10% poisoning rates. The results demonstrate that a clean teacher alone is not a sufficient safeguard: poisoned distillation data can produce a strongly backdoored student while maintaining competitive performance on clean images. …

arXiv:2609.22818v2 Announce Type: replace Abstract: Memory-poisoning defenses for LLM agents are typically evaluated by how well they stop attacks, even though almost all traffic they encounter is benign. We measure their cost when no attack is present. Our harness holds the memory backend, retrieval, answer prompt, and judge constant, changing only the defense. We test four defenses on half of LoCoMo: five long conversations with 750 turns and 255 questions per run, repeated three times each. Replication proved necessary: one run showed a reranker cost on adversarial questions that the other two did not. The three write-time defenses (sanitization, provenance checking, and LLM-based anomaly detection) never fired on benign traffic. We detect no accuracy cost from them at a resolution of about $\pm$4 points; anomaly detection adds 1.2% token overhead. The read-time reranker, with 2.7% token overhead, fired on 9.9% of its memory decisions and withheld memory from 45% of questions. It lowered core accuracy by 4.4 points (95% CI [-9.0, -0.05]; McNemar p = 0.064). The effect is borderline in magnitude, but its direction held across all six runs that used the reranker. A blind audit fou…

Canva is diving even deeper into AI with the launch of an image generation model that competes directly with those from OpenAI, Microsoft, and Google. The model, called Canva Craft, uses a two-stage architecture that allows it to plan a design before generating an image, allowing for "greater precision, consistency and control," according to an announcement on Thursday. Canva Craft can generate images at up to 2K resolution and in more than 30 aspect ratios. In addition to generating social media posts and marketing content, it's also supposed to create realistic scenes, digital art, environmental imagery, and more. The model's two-stage sy … Read the full story at The Verge.

Article URL: https://www.lesswrong.com/posts/WHEQps5PK8LrrcTAb/how-does-fable-5-1-hide-messages-in-images Comments URL: https://news.ycombinator.com/item?id=50201445 Points: 80 # Comments: 21

Not to kill the human, of course – to kill the antibiotic-resistant bacteria causing their infection

Canva is making its new image model available to all users, promising better, more customizable performance at lower costs.

Back in August Modular's Mojo language was open-sourced following the acquisition of Chris Lattner's company by Qualcomm. The latest achievement now out of the company is their MAX software being fully opened up with the entire source tree available...

GPU chief doesn't want to hear about job losses or the end of human civilization