AI Microscopes for Faster Cancer Diagnosis
Pushing the Boundaries of Medical Science: The ARM Technology's Role in Streamlining Diagnoses and Enhancing Pathologist Expertise
Google and the Department of Defense are building an AI-powered microscope to help doctors spot cancer
Google and the U.S. Department of Defense have teamed up to build a super-smart microscope that uses artificial intelligence (AI) to help doctors spot cancer. This high-tech microscope, called an Augmented Reality Microscope (ARM), not only shows docs where the cancer is but also how bad it is, all in real-time. The tool is still in the testing phase but has already proven useful. For example, Dr. Nadeem Zafar used it to settle a disagreement with a colleague about how severe a patient's prostate cancer was. The ARM backed up Zafar's hunch in seconds.
This microscope ain't your high school biology version; it's got AI brains connected to a computer tower. It’s designed to fit right into the workflow of pathologists, the doctors who diagnose diseases like cancer by looking at tissues and fluids. These docs are swamped these days, and any mistake can be costly, so a helping hand from AI could be a game-changer.
Microsoft AI Researchers Expose 38TB of Data, Including Keys, Passwords and Internal Messages
Oops! Microsoft goofed up big time, accidentally exposing a whopping 38 terabytes of sensitive data. This happened when they were updating some open-source AI training stuff on GitHub. What got spilled? Well, it's a laundry list: backups from two employee computers, company secrets, passwords, and a ton of internal messages between staff.
The blunder was caught by Wiz, a cloud security startup founded by former Microsoft engineers. They were doing a routine scan and found that Microsoft messed up the settings, making a whole lot of data accessible to anyone who had the link. Even worse, whoever accessed this could also delete or mess with the files.
This isn't just about leaked secrets; it's a full-blown security nightmare. If bad guys got their hands on this, they could've tampered with Microsoft's AI models. Anyone using those models could've been hit with malicious code.
Good news is, Microsoft fixed the access issue in two days after Wiz told them about it, and replaced the flawed link within a month. But man, that's one egg on their face they won't be wiping off anytime soon.
UK antitrust regulator lays out seven AI principles
The UK's main antitrust watchdog, the Competition and Markets Authority (CMA), has laid down seven rules for companies making AI. They're focusing on big AI systems that many other products rely on. The rules cover a range of stuff—like making sure companies are responsible for what their AI does, and keeping the tech open for everyone to use. They also want companies to be clear about the risks and limits of using AI.
The idea is to keep the playing field level, so no single company becomes too powerful. These guidelines came after chatting with big tech companies and experts. Other countries, like the EU and China, are also working on AI rules, while the US is still getting its act together.
Writer nets $100M for its enterprise-focused generative AI platform
Writer, a company making AI that can write text for businesses, just scored a big win: $100 million in new funding. This latest cash influx, led by ICONIQ Growth and others, pushes the company's total money raised to $126 million and sets its worth at up to $750 million. What's the plan? They're gonna develop specialized AI that's a whiz at creating industry-specific text.
They've got some edge, too. Unlike competitors, they use non-copyrighted business writing to train their AI and let customers peek under the hood at the code. Also, they make sure their AI plays by the company rules, whether that's legal stuff or brand voice. With a growing list of big-name clients and a 10x revenue bump in two years, it looks like Writer is making waves in the crowded AI writing market.
Anyscale optimizes open source AI deployments with Endpoints
Anyscale, the main company behind the popular open-source Ray framework for machine learning, is making it easier and cheaper to use large language models (LLMs). They've just rolled out Anyscale Endpoints, a service that lets businesses fine-tune and deploy these LLMs without the headache of doing it themselves. It's as simple as using an API. They've also teamed up with Nvidia to make their platform even better.
Anyscale's not just stopping at making deployment easier. They're also letting companies fine-tune these models to work better for specific tasks. So, you get a cheaper model that still does the job well. Plus, for companies worried about data privacy, Anyscale is offering a Private Endpoints service. It lets you keep all the data in-house while still getting all the benefits.
Oracle brings voice-activated AI to healthcare with Clinical Digital Assistant
Oracle's jumping into the healthcare game with a voice-activated AI assistant that helps doctors and nurses cut down on paperwork and focus on patients. Announced at a health conference in Vegas, this tool gets plugged into electronic health records and can handle routine stuff like pulling up MRI scans when a doc just asks for it. The tech aims to build trust in a sector that's been slow to adopt AI, mostly 'cause people are worried it might be biased or untrustworthy.
Not just for healthcare pros, this assistant will also help patients by scheduling appointments and even answering medical questions. Oracle says the tool will help with the big problem of not having enough healthcare workers; they expect we'll be short by 18 million by 2030. Some features are out now, but they're planning to roll out everything within a year.
Betterleap leverages AI to revolutionize recruitment, launches with impressive $13M in seed funding
Betterleap, a hot new startup, just jumped into the job-hunting game with a cool $13 million in initial funding. Big names like a16z and Peakstate Ventures are backing them. What's the deal? They've got a monster database of job seekers—over a billion profiles from places like LinkedIn and Indeed. Using fancy AI, Betterleap helps recruiters find the right people faster and cheaper.
The CEO, Khaled Hussein, says their tech is a game changer. They've got unique features like AI-driven personalized messaging and a special focus on reaching talent across different industries. They're also big on diversity, using AI to ignore stuff like names and race when picking candidates.
AI and machine learning can successfully diagnose polycystic ovary syndrome
New study says computers can help docs spot Polycystic Ovary Syndrome (PCOS), a common hormone problem in women. This condition can mess with your periods, cause acne, and even lead to other health issues like diabetes. Researchers from the National Institutes of Health looked at 25 years of data and found that AI—basically, really smart computer programs—can pick up on signs of PCOS with pretty high accuracy, between 80-90%.
This is good news because often PCOS goes unnoticed and misdiagnosed, leading to more health problems down the line. The study suggests that adding AI tools to electronic health records could make diagnosing PCOS easier and more reliable. So, not only could this make life better for women dealing with PCOS, it could also save some cash for the healthcare system.
Multi-AI collaboration helps reasoning and factual accuracy in large language models
MIT researchers found a cool way to make AI language models smarter and more accurate. Instead of just one model tackling a question, they got multiple AIs to chat it out and come up with a group answer. This back-and-forth between the AIs helps them correct each other's mistakes and settle on a better solution. It's like having a team pow-wow, but for computers. The best part? You don't need to mess with the inside techy stuff of each AI to make this work. Plus, it's not just about talking; this method can be used for other stuff like video and speech analysis.
This group debate among AIs even helps them avoid spitting out random nonsense. They tested this setup on math problems and saw a big improvement. But the method isn't perfect yet; it struggles with really complex questions and needs more work to fully act like a human group discussion. Still, it's a big step toward making AIs think more like us and could even let them improve themselves without human help.
What'd you think of today's edition?
What are MOST interested in learning about AI?
What stories or resources will be most interesting for you to hear about?