The AI job cuts are here - or are they? Amazon's move this week to slash thousands of corporate jobs fed into a longstanding anxiety: that Artificial Intelligence is starting to replace workers. The tech giant joined a growing list of companies in the US that have pointed to AI technology as a reason behind layoffs. But some question whether AI is fully to blame - and have voiced scepticism that recent high-profile layoffs are a telling sign of the technology's effect on employment. Chegg, the online education firm, cited the new realities of AI as it announced a 45% reduction in workforce on Monday.

LLM agents have begun to appear as personal assistants, customer service bots, and clinical aides. While these applications deliver substantial operational benefits, they also require continuous access to sensitive data, which increases the likelihood of unauthorized disclosures. Moreover, these disclosures go beyond mere explicit disclosure, leaving open avenues for gradual manipulation or sidechannel information leakage. This study proposes an auditing framework for conversational privacy that quantifies an agent's susceptibility to these risks. The proposed Conversational Manipulation for Privacy Leakage (CMPL) framework is designed to stress-test agents that enforce strict privacy directives against an iterative probing strategy. Rather than focusing solely on a single disclosure event or purely explicit leakage, CMPL simulates realistic multi-turn interactions to systematically uncover latent vulnerabilities. Our evaluation on diverse domains, data modalities, and safety configurations demonstrates the auditing framework's ability to reveal privacy risks that are not deterred by existing single-turn defenses, along with an in-depth longitudinal study of the temporal dynamics of leakage, strategies adopted by adaptive adversaries, and the evolution of adversarial beliefs about sensitive targets. In addition to introducing CMPL as a diagnostic tool, the paper delivers (1) an auditing procedure grounded in quantifiable risk metrics and (2) an open benchmark for evaluation of conversational privacy across agent implementations.

The toy company uses a further classification algorithm specifically for toys that fall under Class A to break down those toys into additional classes. They use a regression algorithm to match categories of toys with customer profiles to produce probabilities of certain types of customers purchasing certain types of toys. Historical customer transaction data is used to determine the types of toys a customer previously purchased, how frequently those types of toys were purchased and how much, on average, the customer spent on those types of toys. For example, the algorithm determines that a customer who had previously purchased action figure toys five times over the past six months and who spent an average of $80 on each purchase is 71% likely to purchase a new action toy of the same type that is priced at $40. The purchase probability of other toys is also assessed and compared, and the promotional email sent to that customer is customized to highlight the new toys with at least a 60% purchase probability at the toy's current price.

Speed: GPUs have a high number of cores, which makes them well-suited for parallel processing tasks such as matrix operations. This makes them faster for certain types of machine learning tasks, such as training deep neural networks. Cost-effectiveness: Training large machine learning models can require a lot of computational resources, and using GPUs can be more cost-effective than using CPUs for these tasks, as they can process large amounts of data much faster. Large-scale training: Training deep neural networks requires a lot of data and computational power, which makes GPUs ideal for this type of work. By using GPUs, researchers and practitioners can train much larger and more complex models than they would be able to with CPUs alone.

Between 60–75% of users reported they feel more fulfilled with their job, feel less frustrated when coding, and are able to focus on more satisfying work when using GitHub Copilot. That's a win for developers feeling good about what they do! Developers reported that GitHub Copilot helped them stay in the flow (73%) and preserve mental effort during repetitive tasks (87%). That's developer happiness right there, since we know from previous research that context switches and interruptions can ruin a developer's day, and that certain types of work are draining [8, 9]. Between 60–75% of users reported they feel more fulfilled with their job, feel less frustrated when coding, and are able to focus on more satisfying work when using GitHub Copilot.

This note shares a final recap of my series of notes on the different topics from Andrew Ng's Machine Learning course. I hope that these notes help to make machine learning more accessible and create greater collective intuition around machine learning. What are the types of Machine Learning? ML is when a computer program is able to learn without being programmed explicitly, which is often illustrated as getting better at a task as it gains more experience according to a measure of performance. ML can be bucketed as either (1) "supervised learning" where it learns from the "right answers" (e.g., regression or classification), or (2) "unsupervised learning" where there are no "right answers" given, but the model finds structure or patterns in the data (e.g., clustering).

Every day, humans produce hundreds of millions of pieces of content -- countless pictures, comments, blog posts, videos, new social media channels, every second of every day. It's a never-ending process fueled by algorithms and human interaction, and the Internet is only getting more saturated. This has led many content creators to begin asking themselves: should they use artificial intelligence to create content for them? How can they possibly keep up and compete with the flood of new, interesting, SEO-optimized content their audience sees every time they open their phone? Wouldn't it be better for a publisher's site to use the infinite scaling of AI instead of the publisher's own finite time and energy?

In 2018, the European Union introduced the General Data Protection Regulation (GDPR) which has clauses that impact AI – notably text indicating a "right to explanation" – an area that affects AI algorithms and has been the subject of much debate since its introduction. Elsewhere, local regulations have been attempted, ranging from bans on the use of certain types of AI (such as facial recognition), to committees to examine the fairness of algorithms used in resource allocation. The exact criterion and specifics of the law are still being debated, with exceptions and loopholes having been identified by a number of institutions. There are many regulations in development, and to make things even more complicated, they differ in their geographical or industry scope and in their targets. Forming a cohesive practice will make it easier to see these regulations as connected entities that are addressed together.

In this talk, Aparna Dhinakaran, Co-Founder and CPO of Arize AI, covered the challenges organizations face in checking for model fairness, such as the lack of access to protected class information to check for bias and diffuse organizational responsibility of ensuring model fairness. Aparna also dived into the approaches organizations can take to start addressing ML fairness head-on with a technical overview of fairness definitions and how practical tools such as ML Observability can help build ML fairness checks into the ML workflow. If you've heard of Michelangelo, Aparna built part of the model store, which was eventually kind of integrated into Michelangelo. After that, Aparna actually went to a Ph.D. program in Computer Vision, at which time she started thinking about things like ML fairness, how does bias get introduced into our models, especially models like facial recognition? At that point, as a researcher, Aparna was realising that she couldn't even answer basic questions about model performance or model service metrics.

I am often asked about the state of data science and where we sit now from a maturity perspective. The answer is pretty interesting, especially now that it's been more than a year since COVID-19 rendered most data science models useless -- at least for a time. COVID forced companies to make a full model jump to match the dramatic shift in daily life. Models had to be rapidly retrained and redeployed to try to make sense of a world that changed overnight. Many organizations ran into a wall, but others were able to create new data science processes that could be put into production much faster and easier than what they had before.