A new paper by security experts Metomic surveying more than 400 CISOs in the UK and the US found security breaches linked to generative AI worry almost three-quarters (72%) of the respondents.

But that’s not the only thing CISOs are worried about, when it comes to generative AI, the report warns, as they also fear people will use sensitive company data to train the Large Language Models (LLM) used to power these tools. Sharing the data this way is a security risk, as there is a theoretical possibility that a malicious third party might extract this information somehow.

Spotting malware

CISOs have every right to be worried, though. Data breaches and similar cybersecurity incidents have been rising quarter into quarter, year after year. Since the introduction of generative AI tools, these attacks have gotten even more sophisticated, some researchers said.

For example, poor writing, as well as grammar and typing errors, were the best way to spot a phishing attack. Today, most hacking groups use AI to write convincing phishing emails for them, not only making them harder to spot, but also significantly lowering the barrier for entry.

Another example is the writing of malicious code. Be it for a landing page, or for malware, hackers are constantly finding new ways to abuse the new tools. Generative AI developers are fighting back, putting limits in place that prevent the tools from being used this way, but threat actors have so far always managed to find a way around such roadblocks.

Good news is that AI can also be used in defense, and many organizations have already deployed advanced, AI-powered solutions.

This article first appeared on Techradar.com

The post CISOs are nervous Gen AI use could lead to more security breaches appeared first on CURT HULBERT.

So recently, when the Census Bureau released its new estimates for domestic migration, I thought we were finally going to see a reversal of the big-city exodus. But I was wrong. From mid-2022 to mid-2023, the bleeding in many big metropolitan areas continued. New York lost 238,000 more people than it gained. The numbers read like casualty reports: 155,000 in Los Angeles, 54,000 in San Francisco, 25,000 in Seattle. Granted, the urban flight isn’t as bad as the crisis-level hemorrhaging we saw in the first year of the pandemic. But every day, hundreds and hundreds of people continue to forsake America’s greatest cities for smaller, more affordable destinations.

We’ve heard a lot about how the mass migration has been bad for major cities, sending them into a “doom loop” of empty offices and shuttered storefronts. But a new paper coauthored by Enrico Moretti, one of the best thinkers on the geography of jobs, highlights the dangers the migration poses for the very professionals who are ditching big cities. Moving away from a major city, Moretti found, can be terrible for your career.

Moretti, an economist at the University of California at Berkeley, followed workers whose companies shut down between 2010 and 2017. How people fared after that depended on where they lived. Those who lived in small labor markets were less likely to find a new job within a year than those in large labor markets. To get back on their feet professionally, those in small markets were more likely to be forced to relocate for employment. They were also more likely to settle for a role that was misaligned with their college degree, or in an entirely different industry.

“The big takeaway is that market size matters,” Moretti says. “It’s clear that larger markets improve the quality of the match.”

That’s precisely why workers and the companies that employ them tend to cluster in the same cities. Economists call it agglomeration. Let’s say you’re a coder specializing in AI. You’re far more likely to find a job in San Francisco than you are pretty much anywhere else in the world, because there are a lot of AI-related companies there. And it’s because AI specialists flock to San Francisco that AI businesses set up shop there in the first place. That’s how cities become hubs for particular industries, like finance in New York and fashion in Paris. And that’s why people put up with all the downsides of cities — because it increases their odds of growing their careers. Moretti’s new paper confirms that when it comes to jobs, geography is destiny.

At first it seemed as though the pandemic had rewritten that rule. With the rise of remote work, professionals thought they could afford to leave their expensive cities without a risk to their careers. If you moved to Des Moines and wound up losing your job, you could just stay put and get another work-from-home gig. Your house might be in Iowa, but your job market was still back in California or New York.

But over the past year, more and more employers have stopped hiring for remote roles. The market for WFH jobs has cratered, putting everyone who moved away from big cities at risk. If they wind up getting laid off or they outgrow their current role, living in a smaller job market is going to severely limit their career options. As Moretti’s paper shows, they’ll either (1) wind up unemployed for a long stretch, (2) be forced to settle for a local job they’re overqualified for, or (3) have to make an abrupt and costly move back to the big city they abandoned.

Moretti characterizes being in a large labor market as “insurance” against future shocks. Living in a big city isn’t just about having a good job right now. It’s what sets you up for success to land your next job — and the job after that. Those who moved away from big cities effectively gave up their career insurance.

And that’s not all they gave up. When you live in an industry hub, you’re surrounded by professional peers, making it easier for you to accumulate knowledge and skills. That’s not just because you get to collaborate in person with your coworkers every day. In a big city, you also run into people who work for other companies in your industry — on the bus, at the bar, in line at the deli. Those serendipitous conversations not only expand your professional network, they also create what economists refer to as “knowledge spillovers,” helping you learn new stuff that’s relevant to your work. That’s why innovation, as measured by patents, is higher in large markets, and why businesses in big cities tend to have higher productivity.

These upsides to living in a big hub are less obvious than the low rents and nice homes that have lured so many professionals to smaller cities. They also take time to manifest — you have to lose your job before you realize how hard it will be to find another. So professionals haven’t come to grips yet with what they gave up by moving away. “The benefits of being a big city,” Moretti tells me, “have been underappreciated” during the pandemic.

Until recently, I was one of those underappreciators. In 2021, my remote job allowed me to move from San Francisco to Sacramento for my then-wife’s job. We could suddenly afford a spacious two-bedroom in an apartment complex with a pool, and I loved the slower pace of life. Even after we split up, my initial plan was to stay in Sacramento. But as I started driving into San Francisco more often to meet with sources in the tech industry, I realized that being in the city was helping me come up with better story ideas. So a few months ago, I moved back to the Bay Area. Being here makes me better at my job. And if I should lose my job at some point, I know my search will go much better here, where there are more journalism jobs, than it would if I had stayed in Sacramento.

Of course, lots of professionals who left big cities during the early days of remote work will stay where they are, even if they lose their jobs. After all, one of the other trends spurred by the pandemic was the realization that career isn’t everything. Plenty of people will be happy to settle for a lesser job if it means they don’t have to sit in soul-crushing traffic and can have a big backyard for their kids.

But Moretti thinks the exodus from big cities is nearing an end. As the outward migration slows, he predicts, new people looking for career opportunities will flood into urban areas, more than making up for the people who left. The big hubs will resume agglomerating, just as they did in the decades leading up to the pandemic. Cities, in short, will be cities again.

“It’s more a matter of when than if,” Moretti says. “I never thought this was going to be a permanent change in the geography of labor.”

Madison Hoff contributed reporting.

This article was posted on MSN.com: https://www.msn.com/en-us/lifestyle/career/remote-workers-are-about-to-get-a-rude-awakening/ar-BB1lDKzu

The post Remote workers are about to get a rude awakening appeared first on CURT HULBERT.

To date, most successful experiments have involved large amounts of human oversight. Explaining every computation inside models the size of GPT-4 and larger will almost certainly require more automation—perhaps even using AI models themselves.

Facilitating this timely endeavor, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a novel approach that uses AI models to conduct experiments on other systems and explain their behavior. Their method uses agents built from pretrained language models to produce intuitive explanations of computations inside trained networks.

Central to this strategy is the “automated interpretability agent” (AIA), designed to mimic a scientist’s experimental processes. Interpretability agents plan and perform tests on other computational systems, which can range in scale from individual neurons to entire models, in order to produce explanations of these systems in a variety of forms: language descriptions of what a system does and where it fails, and code that reproduces the system’s behavior.

Unlike existing interpretability procedures that passively classify or summarize examples, the AIA actively participates in hypothesis formation, experimental testing, and iterative learning, thereby refining its understanding of other systems in real time.

Complementing the AIA method is the new “function interpretation and description” (FIND) benchmark, a test bed of functions resembling computations inside trained networks, and accompanying descriptions of their behavior.

One key challenge in evaluating the quality of descriptions of real-world network components is that descriptions are only as good as their explanatory power: Researchers don’t have access to ground-truth labels of units or descriptions of learned computations. FIND addresses this long-standing issue in the field by providing a reliable standard for evaluating interpretability procedures: explanations of functions (e.g., produced by an AIA) can be evaluated against function descriptions in the benchmark.

For example, FIND contains synthetic neurons designed to mimic the behavior of real neurons inside language models, some of which are selective for individual concepts such as “ground transportation.” AIAs are given black-box access to synthetic neurons and design inputs (such as “tree,” “happiness,” and “car”) to test a neuron’s response. After noticing that a synthetic neuron produces higher response values for “car” than other inputs, an AIA might design more fine-grained tests to distinguish the neuron’s selectivity for cars from other forms of transportation, such as planes and boats.

When the AIA produces a description such as “this neuron is selective for road transportation, and not air or sea travel,” this description is evaluated against the ground-truth description of the synthetic neuron (“selective for ground transportation”) in FIND. The benchmark can then be used to compare the capabilities of AIAs to other methods in the literature.

Sarah Schwettmann, Ph.D., co-lead author of a paper on the new work and a research scientist at CSAIL, emphasizes the advantages of this approach. The paper is available on the arXiv preprint server.

“The AIAs’ capacity for autonomous hypothesis generation and testing may be able to surface behaviors that would otherwise be difficult for scientists to detect. It’s remarkable that language models, when equipped with tools for probing other systems, are capable of this type of experimental design,” says Schwettmann. “Clean, simple benchmarks with ground-truth answers have been a major driver of more general capabilities in language models, and we hope that FIND can play a similar role in interpretability research.”

Automating interpretability

Large language models are still holding their status as the in-demand celebrities of the tech world. The recent advancements in LLMs have highlighted their ability to perform complex reasoning tasks across diverse domains. The team at CSAIL recognized that given these capabilities, language models may be able to serve as backbones of generalized agents for automated interpretability.

“Interpretability has historically been a very multifaceted field,” says Schwettmann. “There is no one-size-fits-all approach; most procedures are very specific to individual questions we might have about a system, and to individual modalities like vision or language. Existing approaches to labeling individual neurons inside vision models have required training specialized models on human data, where these models perform only this single task.

“Interpretability agents built from language models could provide a general interface for explaining other systems—synthesizing results across experiments, integrating over different modalities, even discovering new experimental techniques at a very fundamental level.”

As we enter a regime where the models doing the explaining are black boxes themselves, external evaluations of interpretability methods are becoming increasingly vital. The team’s new benchmark addresses this need with a suite of functions, with known structure, that are modeled after behaviors observed in the wild. The functions inside FIND span a diversity of domains, from mathematical reasoning to symbolic operations on strings to synthetic neurons built from word-level tasks.

The dataset of interactive functions is procedurally constructed; real-world complexity is introduced to simple functions by adding noise, composing functions, and simulating biases. This allows for comparison of interpretability methods in a setting that translates to real-world performance.

In addition to the dataset of functions, the researchers introduced an innovative evaluation protocol to assess the effectiveness of AIAs and existing automated interpretability methods. This protocol involves two approaches. For tasks that require replicating the function in code, the evaluation directly compares the AI-generated estimations and the original, ground-truth functions. The evaluation becomes more intricate for tasks involving natural language descriptions of functions.

In these cases, accurately gauging the quality of these descriptions requires an automated understanding of their semantic content. To tackle this challenge, the researchers developed a specialized “third-party” language model. This model is specifically trained to evaluate the accuracy and coherence of the natural language descriptions provided by the AI systems, and compares it to the ground-truth function behavior.

FIND enables evaluation revealing that we are still far from fully automating interpretability; although AIAs outperform existing interpretability approaches, they still fail to accurately describe almost half of the functions in the benchmark.

Tamar Rott Shaham, co-lead author of the study and a postdoc in CSAIL, notes that “while this generation of AIAs is effective in describing high-level functionality, they still often overlook finer-grained details, particularly in function subdomains with noise or irregular behavior.

“This likely stems from insufficient sampling in these areas. One issue is that the AIAs’ effectiveness may be hampered by their initial exploratory data. To counter this, we tried guiding the AIAs’ exploration by initializing their search with specific, relevant inputs, which significantly enhanced interpretation accuracy.” This approach combines new AIA methods with previous techniques using pre-computed examples for initiating the interpretation process.

The researchers are also developing a toolkit to augment the AIAs’ ability to conduct more precise experiments on neural networks, both in black-box and white-box settings. This toolkit aims to equip AIAs with better tools for selecting inputs and refining hypothesis-testing capabilities for more nuanced and accurate neural network analysis.

The team is also tackling practical challenges in AI interpretability, focusing on determining the right questions to ask when analyzing models in real-world scenarios. Their goal is to develop automated interpretability procedures that could eventually help people audit systems—e.g., for autonomous driving or face recognition—to diagnose potential failure modes, hidden biases, or surprising behaviors before deployment.

Watching the watchers

The team envisions one day developing nearly autonomous AIAs that can audit other systems, with human scientists providing oversight and guidance. Advanced AIAs could develop new kinds of experiments and questions, potentially beyond human scientists’ initial considerations.

The focus is on expanding AI interpretability to include more complex behaviors, such as entire neural circuits or subnetworks, and predicting inputs that might lead to undesired behaviors. This development represents a significant step forward in AI research, aiming to make AI systems more understandable and reliable.

“A good benchmark is a power tool for tackling difficult challenges,” says Martin Wattenberg, computer science professor at Harvard University who was not involved in the study. “It’s wonderful to see this sophisticated benchmark for interpretability, one of the most important challenges in machine learning today. I’m particularly impressed with the automated interpretability agent the authors created. It’s a kind of interpretability jiu-jitsu, turning AI back on itself in order to help human understanding.”

Schwettmann, Rott Shaham, and their colleagues presented their work at NeurIPS 2023 in December. Additional MIT co-authors, all affiliates of the CSAIL and the Department of Electrical Engineering and Computer Science (EECS), include graduate student Joanna Materzynska, undergraduate student Neil Chowdhury, Shuang Li, Ph.D., Assistant Professor Jacob Andreas, and Professor Antonio Torralba. Northeastern University Assistant Professor David Bau is an additional co-author.

More information: Sarah Schwettmann et al, FIND: A Function Description Benchmark for Evaluating Interpretability Methods, arXiv (2023). DOI: 10.48550/arxiv.2309.03886

This story is republished courtesy of MIT News (web.mit.edu/newsoffice/), a popular site that covers news about MIT research, innovation and teaching.

Provided by Massachusetts Institute of Technology

This story was originally published on Tech Xplore.

The post AI agents help explain other AI systems appeared first on CURT HULBERT.

Nursing has faced numerous challenges in recent years, but some believe that artificial intelligence (AI) could save the day. Here’s how Nvidia (NASDAQ: NVDA) is getting involved with healthcare, and how the company plans to disrupt the nursing workforce.

Nursing shortages are a real problem

One of the biggest challenges facing hospitals right now is nurse shortages. According to a survey conducted by management consulting firm McKinsey, nearly one-third of nurses may leave their job within the next year.

The reason? Money. Per McKinsey’s report, nurses cited “inadequate compensation” as the top reason for considering leaving their job in patient care.

According to the Bureau of Labor Statistics, median pay for nurses is about $39 per hour, or $81,000 per year. Considering median household income is about $75,000, nursing might appear more appealing compared to other careers.

But nurses typically take on longer shifts compared to traditional office work, and are actively vulnerable to higher-risk environments given the exposure to sick patients.

Artificial intelligence (AI) could come to the rescue

Nvidia has been the poster child of the AI revolution. The company is best known for its high-performance graphics processing units (GPUs) and data center services. However, Nvidia is also making inroads in software and is quietly building an end-to-end AI solution.

And Nvidia is looking to apply AI to healthcare. The company recently revealed that it is working with a start-up called Hippocratic AI, which is developing virtual agents that are capable of understanding and processing basic tasks related to patient care.

For example, the company’s generative AI can help with appointment scheduling as well as follow-up activity after a patient is discharged. To help build its health-focused large language model (LLM), Hippocratic AI is leveraging Nvidia’s Avatar Cloud and NIM microservices technology.

Remember to think long term

Munjal Shah, co-founder and CEO of Hippocratic AI, stated that the end goal of these AI healthcare agents is to “help mitigate widespread staffing shortages and increase access to high-quality care.”

This is a noble mission, but one that should be understood on a deeper level. Patient care is extremely complex. This is why practicing medicine requires several years of education, and why there are so many different types of doctors and specialists.

To think that chatbots will replace human-provided care is almost otherworldly. And yet, here it is — sort of. In the grand scheme of things, Hippocratic AI’s agents are performing fairly basic functions. Time will tell if the technology becomes sophisticated enough to provide commensurate (or superior) value to real-life nurses and doctors.

I think AI has a multitude of applications in the healthcare arena, but I would not bet on robots replacing doctors. Instead, I think a more reasonable area where AI can disrupt medicine is in clinical trials.

For example, Novo Nordisk, which is the pharmaceutical machine behind Ozempic, Rybelsus, and Wegovy, is currently funding an Nvidia-powered supercomputer that it hopes will help in drug development.

Bringing accelerated computing capabilities to the clinical trial process can theoretically assist researchers process data more efficiently. Subsequently, pharmaceutical businesses could uncover meaningful insights including expanded indications for their medicines.

The bigger theme here is that artificial intelligence is being introduced into many facets of healthcare. My personal outlook is that AI will become an integral pillar augmenting research and development. While more personal use cases surrounding patient care could be on the horizon, I see these as lofty ambitions.

In either case, Nvidia is involved in one way or another. I think now is a great time to consider scooping up shares as the company aggressively pursues markets outside of its core chip business.

*Original Story From https://www.msn.com/en-us/health/other/nvidia-is-bringing-artificial-intelligence-ai-to-the-healthcare-space-here-s-how/ar-BB1l070F

The post Nvidia Is Bringing Artificial Intelligence (AI) to the Healthcare Space.* appeared first on CURT HULBERT.

Leading the pack in Nvidia and Broadcom

Nvidia Corporation (NVDA) takes up a front-row position in the AI chip market as it contains a wide gamut of product lines that involve graphics cards, enterprise workstations, and cloud-based products. The company’s gaining of massive market shares of AI-enabled chips for cloud and data centers is impressive evidence of its dominance. The market capitalization of $2.2 trillion, an impressive year-on-year stock growth of 224%, and investor return opportunities are some of the compelling factors when investing in Nvidia stock. The analysts’ consensus is clearly in favor of acquiring NVDA. The projected price at which the share could rise by as much as 4.3% serves as a sign that the market is sure that this market leader will stay that way.

Being one of the industry go-to leaders like Broadcom Inc. (AVGO), the company delivers a vast amount of semiconductor and infrastructure software solutions. With its market cap reaching $620.7 billion and its stock prices demonstrating a whopping 111.7% rise in the prior year, Broadcom’s success is noteworthy. In addition, the stock’s dividend yield of 1.47% combined with a “Strong Buy” recommendation from analysts, who forecast a 12.5% upside potential, give this stock high attractiveness to the investor.

Emerging “Junior SamurAI”

The “junior samurAIs” actress-coach Arya suggests investing in Marvell Technology, Inc. (MRVL), Micron Technology, Inc. (MU), and Advanced Micro Devices, Inc. (AMD). It is obvious that the major identified segment of AI networking, as well as the custom chip market, is data infrastructure, which has been ideal for Marvell’s focus. The company’s performance was boosted by an 83.5% stock price hike and analysts’ “Strong Buy” tag this year; Marvell looks forward to its entrance into the large-capacity accelerator industry.

Micro Technology has gained a reputation due to its advanced expertise in the field of memory and storage solutions. It is another company that could be considered a strong opponent. The corporation has recorded an amazing 113.9% surge in stock market performance over the recent 1-year period. Arya secures Micron’s investment and regards it to have the possibility of expanding its market share in the high-bandwidth memory subsector. Analysts have been the strongest supporters of MU with ‘Strong Buy’ recommendations and a continuing expectation of growth.

AMD has been very much overlooked because of its incredible placement, particularly in the CPU and GPU lines, where its Ryzen processors and Radon graphics are on the rise. Thanks to the shift of its path to AI, AMD can now involve more shares, which shows an 81% increase compared to the previous year. It is reputed that this company might have high placement in the AI accelerator market, with analysts assigning AMD “Strong Buy” status, and selling price growth of 12.72% is expected.

Strategic growth in the AI market

It is no overstatement that the semiconductor industry is at the forefront of the AI revolution, both in its operation and adoption. The 2 contrasting companies have located themselves in the middle of the demand for AI solutions. Three years down the line, the secondary markets are expected to grow bigger by over USD 200 billion. These firms deserve to capitalize on the new opportunities.

The Bank of America Securities’ study thereby accentuates the great prevalence of these AI chip stocks. Investments adopting AI could be some stocks capable of fetching more profits, which the investors may want to consider for their portfolios. The broad spectrum of well-known leaders like Nvidia and Broadcom, along with a handful of fast-blooming stars such as Marvell, Micron, and AMD, comes together to form a diversified way to invest in the AI-centric semiconductor industry.

The semiconductor industry has been involved in AI entailment as much as it has been in AI forwarding. Companies seek to blur boundaries in AI, and AI chip stocks provide a doorway for investors to participate in the industry’s expansion. Considering these companies’ large market positions and previous operations, which are mainly inventive in AI, these stocks are promising for good investment in the vigorous AI industry.

Original Story From https://www.barchart.com/story/news/25332919/5-top-ai-chip-stocks-to-buy-this-april

*Reposted Story From https://www.msn.com/en-us/money/topstocks/ai-chip-stocks-surge-as-demand-skyrockets/ar-BB1lhHkW

The post AI Chip Stocks Surge as Demand Skyrockets* appeared first on CURT HULBERT.