Latest news with #LanguageModels
Geek Wire
a day ago
- Business
- Geek Wire
Amazon will offer OpenAI's open-weight models, sidestepping Microsoft via Apache 2.0 license
Image via Amazon OpenAI released its first open-weight AI models in more than five years, and Amazon quickly said they will be available on its Bedrock and SageMaker platforms — the first time the cloud giant has offered products from the ChatGPT maker. The surprise announcement comes at a pivotal moment in OpenAI's longstanding relationship with Microsoft, which has invested about $13.75 billion in the San Francisco-based company. Microsoft has exclusive rights to OpenAI's API on its Azure cloud platform. OpenAI says the new models, gpt-oss-120b and gpt-oss-20b, are state-of-the-art language models with advanced reasoning capabilities. Their status as open-weight models means that the underlying parameters, or weights, that determine their behavior are freely available for download. This lets developers and companies customize, fine-tune, and deploy the models for themselves. The new models are available under the Apache 2.0 free software license, which means they aren't subject to the exclusive provisions of OpenAI's Microsoft technology agreement. As a result, any cloud provider could offer them — although it's clear from the timing that Amazon got a heads up, at least. Amazon trumpeted the news, saying that the availability of the models on Amazon Web Services 'will put more powerful AI technologies into the hands of organizations and expand the impact of OpenAI's leading technology by making it available to the millions of customers on AWS, the world's most comprehensive and broadly adopted cloud.' OpenAI said as part of its announcement that Microsoft is bringing GPU-optimized versions of the smaller gpt-oss-20b model to Windows devices. We've contacted Microsoft for further comment on the news and its plans. Amazon Web Services has likely lost some business in the past due to the fact that it didn't have any OpenAI models to offer, said Patrick Moorhead, CEO and chief analyst at Moor Insights & Strategy. 'While these aren't the best OpenAI models, they are a start,' Moorhead said via email. 'Microsoft will continue to have the best Open AI models out there, but they now have some competition with different modalities.' Amazon's Bedrock platform offers customers a wide selection of AI models from various companies. In part as a response to Microsoft's OpenAI partnership, Amazon has invested $8 billion in its partner Anthropic, the maker of the Claude chatbot, and offers the startup's models on the platform. Microsoft and OpenAI are currently in talks to renegotiate their agreement in a deal that could give Microsoft ongoing access to critical OpenAI technology, even if the startup achieves its goal of building artificial general intelligence (AGI), according to a Bloomberg News report last week. The talks have reportedly expanded into a larger renegotiation. One key issue is the size of Microsoft's equity stake as OpenAI converts to a for-profit entity.
Asharq Al-Awsat
a day ago
- Business
- Asharq Al-Awsat
OpenAI Releases Open-Weight Reasoning Models Optimized for Running on Laptops
OpenAI said on Tuesday it has released two open-weight language models that excel in advanced reasoning and are optimized to run on laptops with performance levels similar to its smaller proprietary reasoning models. An open-weight language model's trained parameters or weights are publicly accessible, which can be used by developers to analyze and fine-tune the model for specific tasks without requiring original training data. "One of the things that is unique about open models is that people can run them locally. People can run them behind their own firewall, on their own infrastructure," OpenAI co-founder Greg Brockman said in a press briefing. Open-weight language models are different from open-source models, which provide access to the complete source code, training data and methodologies. The landscape of open-weight and open-source AI models has been highly contested this year. For a time, Meta's Llama models were considered the best, but that changed earlier this year when China's DeepSeek released a powerful and cost-effective reasoning model, while Meta struggled to deliver Llama 4. The two new OpenAI models are the first open models OpenAI has released since GPT-2, which was released in 2019. OpenAI's larger model, gpt-oss-120b, can run on a single GPU, and the second, gpt-oss-20b, is small enough to run directly on a personal computer, the company said. OpenAI said the models have similar performance to its proprietary reasoning models called o3-mini and o4-mini, and especially excel at coding, competition math and health-related queries. The models were trained on a text-only dataset which in addition to general knowledge, focused on science, math and coding knowledge. OpenAI did not release benchmarks comparing the open-weight models to competitors' models such as the DeepSeek-R1 model. Microsoft-backed OpenAI, currently valued at $300 billion, is currently raising up to $40 billion in a new funding round led by Softbank Group.
Reuters
a day ago
- Business
- Reuters
OpenAI releases open-weight reasoning models optimized for running on laptops
SAN FRANCISCO, Aug 5 (Reuters) - OpenAI said on Tuesday it has released two open-weight language models that excel in advanced reasoning and are optimized to run on laptops with performance levels similar to its smaller proprietary reasoning models. An open-weight language model's trained parameters or weights are publicly accessible, which can be used by developers to analyze and fine-tune the model for specific tasks without requiring original training data. "One of the things that is unique about open models is that people can run them locally. People can run them behind their own firewall, on their own infrastructure," OpenAI co-founder Greg Brockman said in a press briefing. Open-weight language models are different from open-source models, which provide access to the complete source code, training data and methodologies. The landscape of open-weight and open-source AI models has been highly contested this year. For a time, Meta's (META.O), opens new tab Llama models were considered the best, but that changed earlier this year when China's DeepSeek released a powerful and cost-effective reasoning model, while Meta struggled to deliver Llama 4. The two new OpenAI models are the first open models OpenAI has released since GPT-2, which was released in 2019. OpenAI's larger model, gpt-oss-120b, can run on a single GPU, and the second, gpt-oss-20b, is small enough to run directly on a personal computer, the company said. OpenAI said the models have similar performance to its proprietary reasoning models called o3-mini and o4-mini, and especially excel at coding, competition math and health-related queries. The models were trained on a text-only dataset which in addition to general knowledge, focused on science, math and coding knowledge. OpenAI did not release benchmarks comparing the open-weight models to competitors' models such as the DeepSeek-R1 model. Microsoft-backed OpenAI, currently valued at $300 billion, is currently raising up to $40 billion in a new funding round led by Softbank Group (9984.T), opens new tab.
RTÉ News
3 days ago
- Science
- RTÉ News
Is AI making our brains lazy?
Have you ever googled something but couldn't remember the answer a short time later? Well, you're probably not alone. Multiple studies have shown that "digital amnesia" or the "Google effect" is a consequence of having information readily available at our fingertips. It happens because we don't commit the information to memory in the knowledge that we can easily look it up again. According to The Decision Lab, this bias exists not only for things we look up on internet search engines but for most information that is easily accessible on our computers or phones. For example, most of us can't remember friends, family members or work colleagues' phone numbers by heart. So how can we help our brains to remember information? "Writing is thinking," said Professor Barry O'Sullivan from the School of Computer Science & IT at UCC. Professor O'Sullivan believes there are learning benefits associated with Large Language Models (LLMs), but his view is that we should be applying a precautionary principle to them as its such early days. "If you're not the one doing the writing then you're not the one doing the thinking, so if you're a student or you're the employee, the writing really does need to be yours," he said. LLMs with chatbots or virtual assistant chatbots such as Gemini or ChatGPT, have been trained on enormous amounts of text and data and their systems are capable of understanding and generating human language. With these rapid advances in AI certain tasks are now easier, and when used effectively can save time and money, in our personal and working lives. However, there are concerns about critical thinking, creativity and problem solving. Some AI companies claim their models are capable of genuine reasoning, but there's ongoing debate over whether their "chain of thought" or "reasoning" is trustworthy. According to Professor Barry O'Sullivan, these claims "just aren't true." "These large language models don't reason the same way as human beings," he said. "They have pretty weak abilities to reason mathematically, to reason logically, so it really is the big stepping stone, but it's always been the big stepping stone in AI." he added. He cautions people and workers to use these tools as an assistant, and as a sometimes "unreliable assistant". Professor O'Sullivan also warns AI generated answers could contain a biased view of the world, that is when human brain power is needed to apply sense, reasoning and logic to the data. The Google Search Engine was launched in 1998 and is considered a narrow form of AI. Since then, there have been many studies highlighting how the "Google Effect" is a real phenomenon and its impact on how we remember and learn. Launches such as Open AI's chatbot ChatGPT, and more recently Google's AI Overviews and AI Mode search are all relatively new, meaning there has been less time to study the effects. A new study by researchers at the media lab at Massachusetts institute of Technology (MIT) "Your Brain on ChatGPT", divided 54 people aged 18 to 39 into three groups to write essays. One group used Open AI's ChatGPT, one used Google's Search Engine, and the remaining group used no tools; they had to rely purely on brain power. While it was a very small study and in a very specific location (Boston), it found that of the three groups, ChatGPT users had the lowest brain engagement. While in the brain-only group, researchers found they were more engaged and curious, and claimed ownership and expressed higher satisfaction with their essays. The study suggests the use of LLMs, such as ChatGPT which is what they used but is similar to other LLMs, could actually harm learning, especially for younger users. "The MIT study is one source, but there isn't any definitive evidence of that, however the growing amount of evidence does seem to be tipping on that side of the argument that the more we rely on very sophisticated reasoning systems that can automate the process of writing, the less we are thinking," said Professor O'Sullivan. "People should remember that writing is thinking, so when you when you give up the writing to somebody else, you're not thinking anymore and that does have a consequence." Is digital dependence shaping our brains? Whether in education or in the workplace, the use of AI is becoming increasingly prevalent. In a nutshell it does appear to be shaping our brains, but the debate continues over whether it's happening in a negative or a positive way. As a professor in UCC, Mr O'Sullivan ponders over whether table quizzes are as popular as they used to be, and how young people view need-to-know information. "You often hear students saying, "Well I don't really need to know that because if I was out on the job I'd Google it, and wouldn't that be just fine?" "It is to some extent fine for some pieces of information, but it's also important to know why the information is that way, or what the origin is, or why things are that way," Professor O'Sullivan said. There is a skill shift happening with how we and our brains engage with new technology. This is why human judgement and decision making is more important than ever according to Claire Cogan, behavioural scientist and founder of Behaviour Wise. "There is an effect (from AI) on how the brain learns, so there's an impact on brain health. Some of that is relevant to employers, and its very relevant to individuals," said Ms Cogan. AI is useful in the workplace when it can automate mundane or time-consuming tasks such as generating content and extracting key points from large amounts of data. Ms Cogan noted the theory is when people talk about the pluses and minuses, AI should free up time to allow people to do other things. "So as long as that balance is there, it's a good thing. If that balance isn't there, then it's simply going to have a negative impact," she stated. Referring to the MIT study, she assessed it found evidence that using AI can slow attention and have an impact on the working memory. "The brain will go for shortcuts, if there's an easier way to get to something, that's the way the brain will choose," said Ms Cogan. "However, there are still areas where human intelligence far outweighs anything AI can do, particularly around judgment and decision making, and they're going to become more and more important," she stated. "That's the side driven by people, so there's a whole new skill where people are going to have to learn how to judge when to use AI, how to use it, how to direct it, and how to manage it," she said. Does reliance on AI impact on critical thinking in the workplace? Since the late 90s people have been using search engines to find facts. With the advances and sophistication of AI people are becoming more wary, with real concerns about misinformation, disinformation and deep fakes. So while we are relying on AI tools to help find information, it's more important than ever that we engage core human skills in terms of decision making. Ms Cogan believes in an ideal world teachers and lecturers would be almost preparing people for what is going to happen in five or more years. "It's a particular skill to know when and when not to use AI, just teaching the value in decision making because ultimately the overall goal or the aim is defined by the person. There is a skill to making a good decision, and in a work context to know how to arrive at the best decision, that in itself is a whole topic in behavioural science," she said. What's next for our brains? "For our own sake, we need to nurture our brains and we need to look after our brain health," said Ms Cogan. "The best way we can do that is by remaining actively involved in that whole learning process," she said. The authors of the MIT study urged media to avoid certain vocabulary when talking about the paper and impact of generative AI on the brain. These terms included "brain scans", "LLMs make you stop thinking", "impact negatively", "brain damage" and "terrifying findings". It is very early days when it comes to learning about how these technological advances will impact on us in the long term. Maybe in the near future, AI will be able to summarise and analyse data from upcoming studies to tell us if it is rewiring our brains or making them lazy.
Barnama
30-07-2025
- Business
- Barnama
Meltwater Debuts Industry-First GenAI Lens, Unlocking Enhanced Brand Insights
SAN FRANCISCO, July 30 (Bernama) -- Meltwater, a global leader in media, social, and consumer intelligence, today announces the launch of GenAI Lens, an industry-first, groundbreaking tool that gives companies unprecedented visibility into how their brand, competitors and industry are discussed across all major AI Assistants and LLMs (Large Language Models) including ChatGPT, Claude, Gemini, Perplexity, Grok, and Deepseek. Marketing, Comms, and PR professionals face a growing challenge with more channels to manage and even more to monitor. As generative AI becomes a driving force behind content creation and news cycles, teams need complete visibility into not just what is said, but how it's being shaped by LLMs.
