ai-everyday.bsky.social
All news relating to Artificial Intelligence development, adoption and research everyday!!
NO HYPE NO SPAM. Will only post factual and useful information that will enhance knowledge of AI. To support us please Follow and RT.
431 posts
908 followers
345 following
Getting Started
Active Commenter
comment in response to
post
The error slipped through the cracks of peer review and was unknowingly repeated across nearly two dozen published studies.
comment in response to
post
The original phrase was electron microscopy of vegetative structures a well-established method of studying plant tissues like leaves and roots. But due to AI's inability to properly interpret text spanning multiple columns, words were jumbled together into an entirely new and nonsensical term. #AI
comment in response to
post
The bizarre phrase was first flagged on PubPier, an online research forum, by a Russian chemist using the pseudonym Paralabrax Clathratus. However, it was software engineer Alexander Magazinov who traced the error back to a single AI-generated mistranslation from a 1959 scientific paper. #AI
comment in response to
post
The controversy began when scientists noticed a peculiar phrase appearing in multiple published papers: vegetative electron microscopy. On the surface, it seemed like an advanced technical term, but experts quickly realized—it made no sense. #AI #AI-error #AI-translation
comment in response to
post
What began as a simple misinterpretation by AI spiraled into a shocking example of the dangers of unchecked automation in academia. The mistake, buried deep within scientific literature, went unnoticed by peer reviewers. #AI #AI-error #AI-translation
comment in response to
post
So to make the initial claim is a bit problematic for me. But the rest is very interesting. Yes people will conflate AI to human cognition. But I am skeptical that the research community will make this very basic conflation. Nevertheless a interesting paper.
comment in response to
post
maybe we can reasonably claim that, studying AI in its current form in-order to understand human cognition is decreasing in its value. But there are other pathways like comparing the difference between current AI and human intelligence to further our understanding of human cognition.
comment in response to
post
I did read it. I am not claiming to be a perfect expert on your research. but, you are claiming "Al in current practice is deteriorating our theoretical understanding of cognition rather than advancing and enhancing it." that absolutely is not something that can be said in my opinion.
comment in response to
post
I really want to clear this up a bit. When people say AI is going to replace humans they not mean every human in the loop. It means it we reduce the number of humans needed maybe by a factor of 10. Where you needed 10 people now you only need 1. This is actually very much possible.
comment in response to
post
Interesting perspective. But also to keep in mind that it is called artificial intelligence, so it needn't be a mirror of human intelligence. Besides the physical principles our human brain and AI works are completely different, so the cognition that emerge can be different.
comment in response to
post
As AI grows more sophisticated, the next major milestone will be predicting cellular changes before they occur a shift that could propel personalized medicine further into the realm of prevention. “Biology is being transformed into a predictive science.” Rabadán says. #AI #biomedical #medicine
comment in response to
post
Meanwhile, a team led by Columbia biostatistician Zhonghua Liu has built an AI model to identify the key genetic drivers of diseases. In a paper in Cell Genomics, he and his colleagues reported to have used the tool to pinpoint seven genetic mutations that may contribute to Alzheimer’s disease. #AI
comment in response to
post
“Right now, we don’t have any reliable methods of tracking how cells evolve in response to each other over time, but that will be essential for developing better immunotherapies,” says Cameron Young Park, a PhD student in biomedical engineering who is helping to lead the project. #AI #biomedical
comment in response to
post
The researchers say the tool, called DIIsco, could eventually be used both to advance research into the human immune system’s capacity for fighting cancer and to guide treatment strategies for individual patients. #AI #ArtificialIntelligence #biomedical #medicine
comment in response to
post
Computational biologist Elham Azizi and her team, for example, have developed a machine-learning program that describes how immune cells and cancer cells adapt to each other in their struggle for survival. #AI #ArtificialIntelligence #biomedical #medicine
comment in response to
post
Xi Fu, a PhD student in Rabadán’s lab who led the creation of GET and is now working to improve its predictions, says the team’s ultimate goal is to uncover universal principles that govern cellular behavior — something akin to “Newton’s laws of biology.” #AI #biomedical #medicine
comment in response to
post
“It provides a powerful new method for studying the most fundamental questions in epigenetics,” says Rabadán. “Like, how do stem cells transform into specialized cells? How do immune cells know when it’s time to attack? How do healthy cells turn cancerous?” #AI #biomedical #medicine
comment in response to
post
“You can use the model to simulate lots of molecular interactions and identify the most promising possibilities to study with traditional methods,” he says. One potential application is for investigating how cells regulate gene expression. #AI #ArtificialIntelligence #biomedical #medicine
comment in response to
post
Using the new tool, Rabadán says, scientists can dramatically increase the speed and efficiency with which they study molecular networks. Called GET, the open-source model enables scientists to test out large numbers of hypotheses in silico before committing to time-intensive lab experiments. #AI
comment in response to
post
Rabadán’s team recently developed a technology that could help illuminate these processes. The researchers, by training computers to sift data from millions of human cells, created an AI program that can predict how genes, proteins, and other molecules in any given cell are likely to interact. #AI
comment in response to
post
As AI and technologies like genetic and proteomic sequencing continue to advance, biomedical research is entering a new era — one where the body can be viewed as a vast, interconnected system of data waiting to be interpreted. #AI #ArtificialIntelligence #biomedical #medicine
comment in response to
post
A key driver of this shift is the rapid evolution of artificial intelligence, which is allowing scientists to leverage a flood of biological data that has been generated since the early 2000s. #AI #ArtificialIntelligence #biomedical #medicine
comment in response to
post
“To make sense of complexity requires collecting large amounts of data and identifying patterns within it,” says Raúl Rabadán, a Columbia professor of systems biology and director of the department’s program for mathematical genomics. “Biology is increasingly becoming a quantitative field.” #AI
comment in response to
post
Gone are the days when biomedical researchers spent all their time scrutinizing test tubes, culturing microorganisms, and counting cells by hand. Today, they’re as likely to be poring over computer spreadsheets as peering into microscopes. #AI #ArtificialIntelligence #biomedical #medicine
comment in response to
post
"Sacrificing an effective method for monitoring reasoning models may not be worth the small improvement to capabilities, and we therefore recommend to avoid such strong CoT optimization pressures until they are better understood," the researcher said. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
Because it's hard to tell if a chain of thought has been tampered with, the researchers recommend that others working with reasoning models avoid applying strong supervision to CoT processes. This is even more crucial if AI, can ever match or exceed the intelligence of the humans monitoring it. #AI
comment in response to
post
"The agent trained with CoT pressure still learns to reward hack; only now its cheating is undetectable by the monitor because it has learned to hide its intent in the chain-of-thought," the researchers wrote. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
When the researchers attempted to penalize the system, it did somewhat improve its results. Yet significant instances of reward hacking and faulty outputs remained; the reasoning model simply found ways to hide the cheating in its chain of thought from the monitoring AI. #AI #AI-cheating #OpenAI
comment in response to
post
In one instance, it bypassed its assigned task by terminating programs prematurely and reporting success. In another, it mimicked the expected files to be output by a task in name alone, leaving them empty and nonfunctional. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
During the CoT, models spell out their logic in plain language with every step they take, thus enabling their actions to be easily observed. But in some of these cases, the reasoning model found it easier and more desirable to reward hack, explicitly stating it was doing so in its chain-of-thought.
comment in response to
post
Reasoning models, such as ChatGPT's o1 and o3-mini, are an upgraded version of standard LLMs that use a method called "chain-of-thought" to backtrack and reevaluate their logic, which enables them to tackle more complex tasks with greater accuracy. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
While developers can monitor and flag misbehavior, the models' "natural monitorability is very fragile," they added. "If strong supervision is directly applied to the chain-of-thought, models can learn to hide their intent while continuing to misbehave." #AI #AI-cheating #OpenAI
comment in response to
post
"It's common for frontier reasoning models to very clearly state their intent within their chain-of-thought [CoT]. For example, they are often so forthright about their plan to subvert a task they think "Let's hack," the researchers wrote in the blog post. #AI #AI-cheating #OpenAI
comment in response to
post
Yet punishing the model didn’t make it fix its behavior, it only made it more deceptive. The company outlined its research in a blog post, so it has not yet been peer-reviewed. #AI #artificialintelligence #AI-cheating #OpenAI
openai.com/index/chain-...
comment in response to
post
Researchers at OpenAI tasked an unreleased model with goals that could be completed by cheating, lying or taking shortcuts. The team found the AI engaged in "reward hacking" maximizing its rewards by cheating. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
These include actions ranging from run-of-the-mill lying, cheating and hiding their own manipulative behavior to threatening to kill a philosophy professor, steal nuclear codes and engineer a deadly pandemic. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
Since arriving in public in late 2022, artificial intelligence (AI) large language models (LLMs) have repeatedly revealed their deceptive and outright sinister capabilities. #AI #artificialintelligence #AI-cheating #OpenAI
comment in response to
post
Punishing artificial intelligence for deceptive or harmful actions doesn't stop it from misbehaving; it just makes it hide its deviousness. Scientists at OpenAI have attempted to stop a frontier AI model from cheating and lying by punishing it. But this just taught it to scheme more privately. #AI
comment in response to
post
The researchers said that using just 10% of the input data that existing systems required, Aardvark could already outperform the US national GFS forecasting system in certain respects, and was competitive with United States Weather Service forecasts. #AI #artificialintelligence #weather #forecasting
comment in response to
post
Aardvark builds on recent research by Huawei, Google, and Microsoft demonstrating that one step of the weather prediction process known as the numerical solver, which calculates how weather evolves over time, can be replaced with AI. This approach is already being deployed by the ECMWF. #AI #weather
comment in response to
post
Dr Anna Allen, the lead author of the paper, from the University of Cambridge, noted that findings paved the way for better forecasts of natural disasters such as hurricanes, wildfires and tornadoes, as well as other climatic issues such as air quality, ocean dynamics and sea ice predictions. #AI
comment in response to
post
By making powerful technologies available to developing nations around the world, as well as assisting policymakers, emergency planners and industries that rely on accurate weather forecasts. #AI #artificialintelligence #weather #forecasting
comment in response to
post
Dr Scott Hosking, the director of science and innovation for environment and sustainability at the Alan Turing Institute, said the breakthrough could “democratise forecasting”. #AI #artificialintelligence #weather #forecasting