Connect with us

AI Research

Worried AI could teach people to build bioweapons? Don’t teach it how, say researchers

Published

on


Welcome to Eye on AI! In this edition…teaching Deep Ignorance…Cohere’s big funding and new hire…AI deskilling…Anthropic acquires Humanloop cofounders…ChatGPT market share.

What if stopping AI from helping someone build a biological weapon was as simple as never teaching it how?

That question had long intrigued Stella Biderman, executive director of the grassroots nonprofit research lab Eleuther AI. In collaboration with the British government’s AI Security Institute, and lead authors Kyle O’Brien and Stephen Casper, Biderman set out to find the answer — something that had never been explored in public before.

In a new paper, Deep Ignorance, the researchers found that filtering risky information out of an AI model’s training data from the start can “bake in” safeguards that are harder to tamper with—even in open-source models that anyone can download and adapt. Crucially, these protections didn’t noticeably hurt the model’s overall performance.

To test the approach, the team trained versions of an open-source AI model on datasets scrubbed of certain “proxy” information—safe stand-ins for dangerous content, such as material related to bioweapons. The models trained on cleaner data were less able to produce harmful information, while performing just as well on most other tasks.

In an X thread about the project, Casper said the goal was to make LLMs “not only safe off the shelf, but also resist harmful tampering.” That’s difficult because most safety efforts so far have focused on post-training tweaks—changes made after a model is built. Those fixes, such as fine-tuning a model’s responses to avoid dangerous outputs, can work in the short term but are easier to undo and can sometimes weaken the model in unintended ways. Pre-training filters aim to bake in safety from the start, so the model stays safe even if someone tries to tamper with it later.

Biderman noted that this kind of work is rare in public research because it’s expensive and time-consuming—a barrier for most academic and nonprofit groups. Private AI companies like OpenAI and Anthropic have the resources, she said, but avoid revealing details of their pretraining processes for competitive reasons and out of concern over copyright risks.

“They could absolutely do this, and who knows if they do it,” she said. “They are incredibly secretive, and don’t really tell you anything.” She pointed to OpenAI’s own hints that it uses some filtering in both its recently released open-weights model and in its proprietary GPT-4o.

In the company’s model card for the open-weights model, OpenAI writes: “To improve the safety of the model, we filtered the data for harmful content in pre-training, especially around hazardous biosecurity knowledge, by reusing the CBRN pre-training filters from GPT-4o.” In other words, the company applied the same screening process used in GPT-4o to weed out potentially dangerous chemical, biological, radiological, and nuclear information before training.

For Biderman, Deep Ignorance is meant to go beyond what tech companies are willing to say publicly. “Having this out in public enables more people to do better,” she said. She added that she was motivated in part by the tech industry’s refrain that its massive datasets can’t be documented or scrutinized. “There’s a story that OpenAI especially really likes to tell about how data is unfathomably large, how could we possibly know what’s in our data,” she said. “That is something that has pissed me off for a long time. I think demonstrating repeatedly that this is wrong is important.”

With that, here’s the rest of the AI news.

Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman

This story was originally featured on Fortune.com



Source link

AI Research

School Cheating: Research Shows AI Has Not Increased Its Scale

Published

on


Changes in Learning: Cheating and Artificial Intelligence

When reading the news, one gets the impression that all students use artificial intelligence to cheat in their studies. Headlines in newspapers such as The Wall Street Journal or the New York Times often mention ‘cheating’ and ‘AI’. Many stories, similar to a publication in New York Magazine, describe students who openly testify about using generative AI to complete assignments.

With the rise of such headlines, it seems that education is under threat: traditional exams, readings, and essays are filled with cheating through AI. In the worst cases, students use tools like ChatGPT to write complete works.

This seems frustrating, but such a thought is only part of the story.

Cheating has always existed. As an educational researcher studying cheating with AI, I can assert that preliminary data indicate that AI has changed the methods of cheating, but not its volumes.

Our early data suggest that AI has changed the method, but not necessarily the scale of cheating that was already taking place.

This does not mean that cheating using AI is not a serious problem. Important questions are raised: Will cheating increase in the future due to AI? Is the use of AI in education cheating? How should parents and schools respond to prepare children for a life that is significantly different from our experience?

The Pervasiveness of Cheating

Cheating has existed for a very long Time — probably since the creation of educational institutions. In the 1990s and 2000s, Don McCabe, a business school professor at Rutgers University, recorded high levels of cheating among students. One of his studies showed that up to 96% of business students admitted to engaging in ‘cheating behavior’.

McCabe used anonymous surveys where students had to indicate how often they engaged in cheating. This allowed for high cheating rates, which varied from 61.3% to 82.7% before the pandemic.

Cheating in the AI Era

Has cheating using AI increased? Analyzing data from over 1900 students from three schools before and after the introduction of ChatGPT, we found no significant changes in cheating behavior. In particular, 11% of students used AI to write their papers.

Our diligent work showed that AI is becoming a popular tool for cheating, but many questions remain to be explored. For example, in 2024 and 2025, we studied the behavior of another 28000-39000 students, where 15% admitted to using AI to create their work.

Challenges of Using AI

Students are accustomed to using AI but understand that there are boundaries between acceptable and unacceptable use. Reports indicate that many use AI to avoid doing homework or to gain ideas for creative work.

Students feel that their teachers use AI, and many consider it unfair when they are punished for using AI in education.

What Will AI Use Mean for Schools?

The modern education system was not designed with generative AI in mind. Traditionally, educational tasks are seen as the result of intensive work, but now this work is increasingly blurred.

It is important to understand what the main reasons for cheating are, how it relates to stress, time management, and the curriculum. Protecting students from cheating is important, but ways of teaching and the use of AI in classrooms also need to be rethought.

Four Future Questions

AI has not caused cheating in educational institutions but has only opened new possibilities. Here are questions worth considering:

  • Why do students resort to cheating? The stress of studying may lead them to seek easier solutions.
  • Do teachers adhere to their rules? Hypocrisy in demands on students can shape false perceptions of AI use in education.
  • Are the rules concerning AI clearly stated? Determining the acceptability of AI use in education may be vague.
  • What is important for students to know in a future rich in AI? Educational methods must be timely adapted to the new reality.

The future of education in the age of AI requires an open dialogue between teachers and students. This will allow for the development of new skills and knowledge necessary for successful learning.



Source link

Continue Reading

AI Research

Artificial intelligence helps break barriers for Hispanic homeownership | National News

Published

on


























Artificial intelligence helps break barriers for Hispanic homeownership | National News | ottumwacourier.com

We recognize you are attempting to access this website from a country belonging to the European Economic Area (EEA) including the EU which
enforces the General Data Protection Regulation (GDPR) and therefore access cannot be granted at this time.

For any issues, call (641) 684-4611.



Source link

Continue Reading

AI Research

Billionaire Ken Griffin Is Loading Up on These 2 Artificial Intelligence (AI) Stocks That Have Increased 88,780% or More

Published

on


These longtime market leaders still have something left in the tank.

Billionaire Ken Griffin, CEO of hedge fund Citadel Advisors, was busy during the second quarter. He and his team went shopping and substantially increased the firm’s stake in some stocks, while also buying new ones.

Some of the biggest names on Wall Street, including Microsoft (MSFT -0.02%) and Apple (AAPL 3.62%), were among the companies whose shares Griffin bought during the period.

These are two of the largest companies in the world by market cap that have generated life-changing returns over the long run. Both have also made moves in the fast-growing artificial intelligence (AI) market. But are these tech leaders still attractive to long-term investors with market caps above $3 trillion?

Let’s find out.

MSFT Total Return Level data by YCharts

1. Microsoft

During the second quarter, Citadel Advisors bought 1.87 million shares of Microsoft, increasing its stake in the company by 1,635.75%.

Griffin and his team aren’t the only ones who have been loading up on the tech leader. There is a reason why Microsoft has crushed broader equities this year and is up 32% since January. Microsoft’s financial results back that up. The company’s revenue and earnings have been growing at a good clip.

In the fourth quarter of its fiscal year 2025, ended on June 30, Microsoft’s revenue jumped by 18% year over year to $76.4 billion. Operating income grew even faster, reaching $34.3 billion, a 23% increase compared to the year-ago period. Net income climbed 24% year over year to $27.2 billion. In other words, Microsoft is capitalizing on growth opportunities while keeping costs under control.

Person sitting at a desk working on a tablet.

Image source: Getty Images.

The tech giant’s most important business is currently its cloud unit, a segment that also offers a host of AI-related services and is growing sales faster than the rest of its business. Microsoft is gaining ground on Amazon, the leader in cloud computing. Although Amazon was first to market, Microsoft has been offering its Office 365 productivity tools (and other services) to businesses for a long time. It’s hardly a leap for these same companies to opt for a provider they already know and trust for their cloud needs.

And the best news is that this is still the early innings of cloud adoption, and for that matter, the AI revolution. As Andy Jassy, Amazon’s CEO, said, “85% of the global IT spend is still on-premises.”

Despite its massive size, Microsoft is poised for excellent long-term opportunities in cloud computing and AI. Add that to the company’s moat from switching costs, its excellent dividend program, and significant cash flow, and Microsoft looks like a no-brainer stock to buy right now.

2. Apple

Citadel Advisors’ stake in Apple increased by a whopping 10,715.95% during the second quarter. That seems like an odd move at first glance.

Apple has faced significant challenges this year, particularly the threat of tariffs. The company manufactures its products abroad, especially in China. With the Trump administration seeking to impose heavy tariffs on imported goods, the market has been concerned about what this will mean for Apple’s business.

Apple recently announced that it would increase its domestic investment in manufacturing to $600 billion over the next decade, in an attempt to appease the current administration and avoid tariffs.

However, Apple has other issues beyond that. The company’s Apple Intelligence — a suite of AI features and services it has released for its latest devices — has failed to impress consumers and investors. So, the iPhone maker is behind in this promising industry.

It’s due to all these factors (and others) that Apple’s shares have declined by 5% this year. However, Griffin and his team clearly saw this as an opportunity to load up on the company’s shares.

In my view, although Apple may struggle for the next few years, the stock remains a solid long-term option. For one, the company’s business is still highly profitable. Apple’s revenue in the third quarter of its fiscal year 2025, ended June 28, increased by 10% year over year to $94 billion. The company’s earnings per share came in at $1.57, representing a 12% increase compared to the year-ago period.

Notably, Apple generates a substantial amount of cash. The company’s trailing-12-month free cash flow may be down 11.6% year over year, but it remains a considerable $96.2 billion.

AAPL Free Cash Flow Chart

AAPL Free Cash Flow data by YCharts

Apple can invest a substantial amount of money in R&D efforts that will ultimately yield results, including advancements in AI. The company has been late to market several times, only to create an innovative version of an already existing product and find massive success. That’s what it did with the iPhone and several products after that, including its AirPods. The difference is that Apple now has a more valuable brand name than it did then.

Apple has an army of loyal customers, an installed base of billions of devices, and a services segment with more than 1 billion paid subscriptions. Even a single highly successful device can have a significant impact on the company’s results.

Lastly, Apple could find ways to fend off the tariff threat. CEO Tim Cook did so during President Donald Trump’s first term. And there is no guarantee that Trump’s aggressive trade plans will survive his administration.

For all these reasons, the stock remains attractive, particularly for investors willing to hold it over the long term.



Source link

Continue Reading

Trending