All articles tagged with #ai systems
Designing successful agentic AI systems requires organizations to reorient workflows around outcomes, appoint mission owners, and foster cross-functional collaboration, moving beyond simple process tweaks to fully integrate AI into their operations.
Supermicro's stock rose about 5% after launching high-volume deliveries of NVIDIA Blackwell Ultra AI systems and related racks, featuring high-performance solutions for large-scale AI applications, supporting rapid deployment and enhanced efficiency.
Analysts predict that the next major buyers of Nvidia's AI systems will be sovereign nations, with potential customers including Japan, Malaysia, South Korea, Sweden, Saudi Arabia, and the United Arab Emirates. Following the GTC 2024 conference, Nvidia's stock rose as the company revealed its expanding customer base. UBS analyst Timothy Arcuri reiterated a buy rating on Nvidia stock and raised the price target to $1,000, anticipating significant revenue growth in the coming years.
OpenAI is investigating user complaints that ChatGPT, built on OpenAI's GPT-4 model, has become "lazy" and unresponsive. Users have reported that the chatbot refuses to fulfill requests or provide detailed answers, speculating that OpenAI intentionally made the change to improve efficiency. OpenAI acknowledged the feedback on social media and its developer forums, stating that no recent changes had been made to the model. The company is looking into the issue and the unpredictability of model behavior.
OpenAI is investigating user complaints that ChatGPT, built on OpenAI's GPT-4 model, has become "lazy" and unhelpful in responding to queries. Users have reported that the chatbot refuses to follow instructions or provide detailed answers, speculating that OpenAI intentionally made the change to improve efficiency. OpenAI acknowledged the feedback on Twitter and stated that no recent changes had been made to the model, promising to look into the issue. The company did not indicate whether it believed the complaints were valid or if ChatGPT's behavior had indeed changed.
Yann LeCun, chief AI scientist at Meta, believes that AI systems are decades away from achieving sentience and common sense, contrary to Nvidia CEO Jensen Huang's prediction of AI surpassing human capabilities in less than five years. LeCun argues that the current focus on language models and text data is insufficient to create advanced human-like AI systems. He emphasizes the need for multimodal AI systems that can process audio, image, and video data. LeCun also expresses skepticism about the practical relevance and feasibility of quantum computing, stating that many problems can be solved more efficiently with classical computers.
A federal judge has dismissed most claims in a copyright lawsuit filed by artists against AI art generators, including Midjourney and DeviantArt, over the unauthorized use of billions of images downloaded from the internet to train AI systems. The judge found that the accusations were "defective" and unclear, questioning whether the AI systems contained copies of copyrighted images and if the artists could prove infringement without identical material created by the AI tools. However, a claim against Stability AI for direct infringement was allowed to proceed. The artists will have to establish that their works were used to train the AI system, but the issue remains contested.
Sony AI researchers argue that current skin tone scales used to test AI algorithms for bias are limited in their focus on lightness or darkness of skin tone, overlooking biases against individuals with red and yellow skin hues. They propose a more "multidimensional" measurement of skin color based on the CIELAB color standard to address this issue and create more diverse and representative AI systems. The researchers found that current scales overrepresent lighter and redder skin tones, leading to less accurate AI systems. While some major AI players, including Google and Amazon, have welcomed Sony's research, others argue that simplicity is important in skin tone scales to ensure cognitive accuracy and reliability.
Google Search now has a grammar checker feature that allows users to check the grammatical correctness of sentences or phrases. By entering a query along with "grammar check," "check grammar," or "grammar checker," users can activate the tool. The feature is currently available only in English. Google's AI systems analyze language to identify and correct grammar and spelling mistakes. Users can provide feedback on the suggested corrections. However, the grammar check feature does not work for content that violates Google Search's policies or contains dangerous, medical, or vulgar language.
Comedian Sarah Silverman is suing OpenAI and Meta, alleging that her copyright has been infringed in the training of their AI systems, ChatGPT and LLaMa. The case against OpenAI claims that the authors' copyrighted materials were used without consent to train ChatGPT, while the case against Meta concerns the use of a dataset, which includes the authors' books, to train the LLaMa system. Legal experts suggest the outcome of the cases may hinge on whether training a large language model is considered fair use or not. This follows previous cases brought by programmers and artists against AI systems.
Hacker George Hotz stated that Mark Zuckerberg and Elon Musk possess a unique decision-making power, metaphorically referred to as the "button," which sets them apart from leaders like Sundar Pichai. This authority allows them to drive innovation and direct the development of advanced AI systems. Hotz also expressed his belief that an AI system capable of citing its sources could disrupt the search engine landscape, potentially outpacing Google. Musk's involvement with OpenAI and Meta's advancements in AI technology were also highlighted.
This week's top neuroscience news articles include groundbreaking discoveries such as the detection of consciousness in dying brains, the potential of nutritional interventions in slowing Alzheimer's disease, the similarities between AI systems and human brains in processing speech signals, the impact of high school education quality on lifelong cognitive health, and the uncovering of the genetic basis behind extraordinary traits in mammalian species. These findings emphasize the importance of ongoing research in the field of neuroscience and its potential for improving human health.
Researchers at the University of California, Berkeley, have found that AI systems can process signals in a way that is remarkably similar to how the brain interprets speech. The team used electrodes on participants’ heads to measure brain waves while they listened to a single syllable and compared that brain activity to an AI system trained to learn English, finding that the shapes were remarkably similar, which could aid in the development of increasingly powerful systems.
OpenAI has launched a bug bounty program for its AI services, including ChatGPT, offering rewards ranging from $200 to $20,000 for vulnerabilities found. However, the program excludes rewards for jailbreaking ChatGPT or causing it to generate malicious code or text. OpenAI notes that such "model safety issues" require broader approaches and should be submitted via the company's model feedback page. Last month, a hacker revealed 80 "secret plugins" for the ChatGPT API, prompting some to suggest that a paid bug bounty program could help catch such vulnerabilities in the future.
OpenAI has launched a "Bug Bounty Program" in partnership with Bugcrowd, offering cash rewards ranging from $200 to $20,000 for people who find bugs in its AI systems like ChatGPT and GPT-4. The program aims to address vulnerabilities and flaws that emerge with complex technology, and submissions are rejected or accepted within two hours. However, certain issues like "Getting the model to say bad things to you" or "Getting the model to write malicious code for you" will not be rewarded. Seven vulnerabilities have already been rewarded.