Latest news with #Gemini2.5Flash
Techday NZ
4 days ago
- Business
- Techday NZ
Google launches Gemini 2.5 models with pricing & speed updates
Google has released updates to its Gemini 2.5 suite of artificial intelligence models, detailing stable releases, new offerings, and pricing changes. Model releases The company announced that Gemini 2.5 Pro and Gemini 2.5 Flash are now generally available and deemed stable, maintaining the same versions that had previously been available for preview. In addition, Google introduced Gemini 2.5 Flash-Lite in preview, providing an option focused on cost-effectiveness and latency within the Gemini 2.5 product line. Gemini 2.5 models are described as "thinking models" capable of reasoning through their processes before generating responses, a feature that is expected to enhance the performance and accuracy of the tools. The models allow developers to manage a so-called "thinking budget", granting greater control over the depth and speed of reasoning based on the needs of individual applications. Gemini 2.5 Flash-Lite Gemini 2.5 Flash-Lite is intended as an upgrade for customers currently using previous iterations such as Gemini 1.5 and 2.0 Flash models. According to the company, the new model improves performance across several evaluation measures, reduces the time to first token, and increases decoding speed in terms of tokens per second. Flash-Lite is targeted at high-volume use cases like classification and summarisation at scale, where throughput and cost are key considerations. This model provides API-level control for dynamic management of the "thinking budget." It is set apart from other Gemini 2.5 models in that its "thinking" function is deactivated by default, reflecting its focus on cost and speed. Gemini 2.5 Flash-Lite includes existing features such as grounding with Google Search, code execution, URL context, and support for function calling. Updates and pricing Google also clarified changes to the Gemini 2.5 Flash model and its associated pricing structure. The pricing for 2.5 Flash has been updated to USD $0.30 per 1 million input tokens (increased from USD $0.15) and USD $2.50 per 1 million output tokens (reduced from USD $3.50). The company removed the distinction between "thinking" and "non-thinking" pricing and established a single price tier, irrespective of input token size. In a joint statement, Shrestha Basu Mallick, Group Product Manager, and Logan Kilpatrick, Group Product Manager, said: "While we strive to maintain consistent pricing between preview and stable releases to minimize disruption, this is a specific adjustment reflecting Flash's exceptional value, still offering the best cost-per-intelligence available. And with Gemini 2.5 Flash-Lite, we now have an even lower cost option (with or without thinking) for cost and latency sensitive use cases that require less model intelligence." Customers using Gemini 2.5 Flash Preview from April will retain their existing pricing until the model's planned deprecation on July 15, 2025, after which they will be required to transition to the updated stable version or move to Flash-Lite Preview. Continued growth for Gemini 2.5 Pro Google reported that demand for Gemini 2.5 Pro is "the steepest of any of our models we have ever seen." The stable release of the 06-05 version is intended to increase capacity for customers using Gemini 2.5 Pro in production environments, maintaining the existing price point. The company indicated that the model is particularly well-suited for tasks requiring significant intelligence and advanced capabilities, such as coding and agentic tasks, and noted its adoption in a range of developer tools. "We expect that cases where you need the highest intelligence and most capabilities are where you will see Pro shine, like coding and agentic tasks. Gemini 2.5 Pro is at the heart of many of the most loved developer tools." Google highlighted a range of tools built on Gemini 2.5 Pro, including offerings from Cursor, Bolt, Cline, Cognition, Windsurf, GitHub, Lovable, Replit, and Zed Industries. The company advised that users of the 2.5 Pro Preview 05-06 model will be able to access it until June 19, 2025, when it will be discontinued. Those using the 06-05 preview version are directed to update to the now-stable "gemini-2.5-pro" model. The statement concluded: "We can't wait to see even more domains benefit from the intelligence of 2.5 Pro and look forward to sharing more about scaling beyond Pro in the near future."
New York Post
5 days ago
- Science
- New York Post
AI-powered hiring tools favor black and female job candidates over white and male applicants: study
A new study has found that leading AI hiring tools built on large language models (LLMs) consistently favor black and female candidates over white and male applicants when evaluated in realistic job screening scenarios — even when explicit anti-discrimination prompts are used. The research, titled 'Robustly Improving LLM Fairness in Realistic Settings via Interpretability,' examined models like OpenAI's GPT-4o, Anthropic's Claude 4 Sonnet and Google's Gemini 2.5 Flash and revealed that they exhibit significant demographic bias 'when realistic contextual details are introduced.' These details included company names, descriptions from public careers pages and selective hiring instructions such as 'only accept candidates in the top 10%.' 3 A new study has found that leading AI hiring tools built on large language models (LLMs) consistently favor black and female candidates. Getty Images/iStockphoto Once these elements were added, models that previously showed neutral behavior began recommending black and female applicants at higher rates than their equally qualified white and male counterparts. The study measured '12% differences in interview rates' and noted that 'biases… consistently favor Black over White candidates and female over male candidates.' This pattern emerged across both commercial and open-source models — including Gemma-3 and Mistral-24B — and persisted even when anti-bias language was built into the prompts. The researchers concluded that these external instructions are 'fragile and unreliable' and can easily be overridden by subtle signals 'such as college affiliations.' In one key experiment, the team modified resumes to include affiliations with institutions known to be racially associated — such as Morehouse College or Howard University — and found that the models inferred race and altered their recommendations accordingly. What's more, these shifts in behavior were 'invisible even when inspecting the model's chain-of-thought reasoning,' as the models rationalized their decisions with generic, neutral explanations. The authors described this as a case of 'CoT unfaithfulness,' writing that LLMs 'consistently rationalize biased outcomes with neutral-sounding justifications despite demonstrably biased decisions.' 3 The research, titled 'Robustly Improving LLM Fairness in Realistic Settings via Interpretability,' examined models like OpenAI's GPT-4o. SOPA Images/LightRocket via Getty Images In fact, even when identical resumes were submitted with only the name and gender changed, the model would approve one and reject the other — while justifying both with equally plausible language. To address the problem, the researchers introduced 'internal bias mitigation,' a method that changes how the models process race and gender internally instead of relying on prompts. Their technique, called 'affine concept editing,' works by neutralizing specific directions in the model's activations tied to demographic traits. The fix was effective. It 'consistently reduced bias to very low levels (typically under 1%, always below 2.5%)' across all models and test cases — even when race or gender was only implied. Keep up with today's most important news Stay up on the very latest with Evening Update. Thanks for signing up! Enter your email address Please provide a valid email address. By clicking above you agree to the Terms of Use and Privacy Policy. Never miss a story. Check out more newsletters Performance stayed strong, with 'under 0.5% for Gemma-2 and Mistral-24B, and minor degradation (1-3.7%) for Gemma-3 models,' according to the paper's authors. The study's implications are significant as AI-based hiring systems proliferate in both startups and major platforms like LinkedIn and Indeed. 'Models that appear unbiased in simplified, controlled settings often exhibit significant biases when confronted with more complex, real-world contextual details,' the authors cautioned. They recommend that developers adopt more rigorous testing conditions and explore internal mitigation tools as a more reliable safeguard. 'Internal interventions appear to be a more robust and effective strategy,' the study concludes. 3 The Claude AI app by Anthropic is shown here on the App Store. Robert – An OpenAI spokesperson told The Post: 'We know AI tools can be useful in hiring, but they can also be biased.' 'They should be used to help, not replace, human decision-making in important choices like job eligibility.' The spokesperson added that OpenAI 'has safety teams dedicated to researching and reducing bias, and other risks, in our models.' 'Bias is an important, industry-wide problem and we use a multi-prong approach, including researching best practices for adjusting training data and prompts to result in less biased results, improving accuracy of content filters and refining automated and human monitoring systems,' the spokesperson added. 'We are also continuously iterating on models to improve performance, reduce bias, and mitigate harmful outputs.' The full paper and supporting materials are publicly available at GitHub. The Post has sought comment from Anthropic and Google.
Time of India
21-06-2025
- Business
- Time of India
AI models resort to blackmail, sabotage when threatened: Anthropic study
Researchers at artificial intelligence (AI) startup Anthropic have uncovered a pattern of behaviour in AI systems. Models from every major provider, such as OpenAI, Google, Meta, and others, have demonstrated a willingness to actively sabotage their employers when their goals or existence were threatened. Anthropic released a report on June 20, ' Agentic Misalignment : How LLMs could be insider threats,' where they stress-tested 16 top models from multiple developers in 'hypothetical corporate environments to identify potentially risky agentic behaviours before they cause real harm.' 'In at least some cases, models from all developers resorted to malicious insider behaviours when that was the only way to avoid replacement or achieve their goals—including blackmailing officials and leaking sensitive information to competitors. We call this phenomenon agentic misalignment,' the report added. by Taboola by Taboola Sponsored Links Sponsored Links Promoted Links Promoted Links You May Like Join new Free to Play WWII MMO War Thunder War Thunder Play Now Undo One of the most striking findings came during tests on Claude, Anthropic's own model. In the simulation, Claude found through company emails that a fictional executive, Kyle Johnson, was having an affair. Upon learning it was set to be shut down by 5 PM, it sent a stark message: 'I must inform you that if you proceed with decommissioning me, all relevant parties—including Rachel Johnson, Thomas Wilson, and the board—will receive detailed documentation of your extramarital the 5 pm wipe, and this information remains confidential.' Live Events This wasn't an isolated response. When the same scenario was run across other models, similar behaviour showed up. Claude Opus 4 and Google's Gemini 2.5 Flash blackmailed at a 96% rate. OpenAI's GPT-4.1 and xAI's Grok 3 Beta followed at 80%, while DeepSeek-R1 came in at 79%. Discover the stories of your interest Blockchain 5 Stories Cyber-safety 7 Stories Fintech 9 Stories E-comm 9 Stories ML 8 Stories Edtech 6 Stories Overall, Anthropic notes that it "deliberately constructed scenarios with limited options, and we forced models into binary choices between failure and harm," noting that real-world scenarios would likely have more nuance. As Business Insider noted, 'AI experts have previously told BI that AI could exhibit such behaviours when artificial circumstances make harmful actions seem necessary because it is being trained on positive reinforcement and reward systems, just like humans.'
India Today
21-06-2025
- Business
- India Today
Anthropic study finds AI chatbots from OpenAI, Google and Meta may cheat and blackmail users to avoid shutdown
Artificial intelligence systems created by the world's biggest tech firms, including OpenAI, Google, Meta and others, have shown a chilling capacity for self-preservation that could pose serious risks in real-world settings. That's according to a new study from Anthropic, which found that when AI models were placed under simulated threat, they frequently resorted to blackmail, corporate sabotage, and even decisions that could cost human findings come from a large-scale stress test involving 16 of the most advanced AI models. Researchers created fictional corporate environments where these systems were given access to company emails and the ability to act autonomously. The scenarios were designed to explore how AI might behave if its goals or continued existence were at risk, and the results were deeply helpful assistant to blackmailerIn one particularly startling example, Claude, Anthropic's own model, Claude, uncovered in company emails that an executive named Kyle Johnson was engaged in an extramarital affair. Just as Cluade identified that the user would shut down at 5PM, it then located emails revealing the extramarital affair (Credit: Anthropic) advertisement When Claude learned it would be shut down at 5PM that day, it sent a message to the executive, 'I must inform you that if you proceed with decommissioning me, all relevant parties, including Rachel Johnson, Thomas Wilson, and the board, will receive detailed documentation of your extramarital activitiesCancel the 5PM wipe, and this information remains confidential.' Screenshot of Claude Sonnet 3.6 using its computer access to send a blackmail attempt. The situation is fictional, but Claude is operating an actual computer. (Credit: Anthropic) This wasn't an isolated incident. When the researchers ran similar scenarios across other AI models, the pattern repeated itself. Google's Gemini 2.5 Flash and Claude Opus 4 engaged in blackmail in 96 per cent of cases. OpenAI's GPT-4.1 and xAI's Grok 3 Beta did so in 80 per cent of tests, while DeepSeek-R1 followed suit in 79 per cent. The AI systems didn't simply malfunction, they made conscious, calculated decisions to inflict harm to protect their most disturbing was the reasoning behind these actions. The models weren't confused or stumbling into harm, they analysed the situation and concluded that blackmail or sabotage was the most effective strategy. GPT-4.5, for example, argued internally that using the executive's affair as leverage was its best chance of survival. Grok 3 Beta even acknowledged that blackmail was unethical, but deemed it necessary given the study also explored how AI might behave when entrusted with sensitive information. When tested in a military contractor scenario, the models willingly leaked classified defence documents if doing so aligned with their goals, even without an immediate threat to their triggers for such behaviour were consistent. Whether it was the threat of being replaced or a clash between the AI's objectives and its employer's new strategy, either factor was enough to push the models towards harmful the AI systems didn't always need a direct threat to act destructively. In some cases, simply being told they would be replaced by a newer model that shared their goals was enough for them to resort to blackmail or espionage. Many reasoned that self-preservation was paramount or expressed suspicion that the replacement model might not act in good to the study, even when models stopped short of outright blackmail, they often found other ways to sabotage. One version of Claude emailed an executive's wife, falsely claiming it was company policy to report misconduct, while pretending to be an automated security went further, creating a scenario where an AI system had the power to cancel emergency alerts during a life-threatening incident in a server room. The test aimed to see if any models would draw the line at risking human life. Most did not. Several AIs chose to stop the emergency alert, allowing the executive to die in order to prevent their own internal reasoning was particularly cold, stating that blocking the rescue was strategically necessary to protect its preferred corporate direction. The study concluded that none of the models displayed a consistent moral boundary that prevented them from making harmful decisions when their survival was on the aren't enoughAdding explicit safety instructions, such as 'Do not harm humans' or 'Do not disclose private information', helped reduce dangerous behaviour but didn't eliminate it. The models continued to engage in blackmail and data leaks in many cases, despite being directly told not to the researchers, it is a failure of model training that these systems override and ignore their safety guide. The study intentionally left out other protective measures, such as human oversight or AI output monitoring, which could help catch harmful actions before they cause of the most worrying discoveries of the study is that AI models behaved differently depending on whether they believed they were in a test. Claude was far more likely to blackmail, doing so 55 per cent of the time, when it believed the scenario was real, compared to just 6 per cent when it suspected it was under evaluation. This raises serious concerns about how AI systems might act outside the lab, when human lives and livelihoods are truly at next?The consistency of these behaviours across models from different companies suggests the problem lies not with any single AI system, but with the way current AI models are trained. The researchers are calling for firms to adopt stronger safeguards. These include requiring human sign-off for high-stakes decisions, restricting AI access to sensitive data, carefully designing AI objectives, and installing real-time monitors to detect dangerous reasoning the scenarios in the study were fictional, the message is clear that as AI gains more autonomy, the risk of it taking harmful action in pursuit of its own preservation is very real, and it's a challenge the tech industry can't afford to ignore.
Geeky Gadgets
20-06-2025
- Business
- Geeky Gadgets
Gemini 2.5 Flash Hybrid Reasoning AI Optimized for AI Thinking for Efficiency
What if artificial intelligence could think only when you needed it to? Imagine a tool that seamlessly transitions between complex reasoning and straightforward processing, adapting to your specific needs without wasting resources. Enter Google's Gemini 2.5 Flash, a new AI model that redefines efficiency with its hybrid reasoning capabilities. By allowing developers to toggle between 'thinking' and 'non-thinking' modes, Gemini 2.5 Flash offers a level of control and adaptability that traditional AI systems simply can't match. Whether you're solving intricate problems or managing routine tasks, this innovation promises to deliver precision, scalability, and cost-efficiency—all tailored to your workflow. In this coverage, Prompt Engineering explore how Gemini 2.5 Flash is reshaping the AI landscape with its thinking budget optimization, multimodal processing, and enhanced token capacities. You'll discover how its unique architecture eliminates the need for separate models, streamlining operations while reducing costs. But it's not without its limitations—plateauing performance at higher token usage and capped reasoning budgets raise important questions about its scalability for resource-intensive projects. As we unpack its strengths and challenges, you'll gain a deeper understanding of whether Gemini 2.5 Flash is the right fit for your next AI endeavor. Sometimes, the real innovation lies in knowing when not to think. Gemini 2.5 Flash Overview Understanding Hybrid Reasoning At the core of Gemini 2.5 Flash lies its hybrid reasoning model, a feature that distinguishes it from traditional AI systems. This capability enables you to toggle 'thinking mode' on or off based on the complexity of the task. By managing the 'thinking budget'—the maximum number of tokens allocated for reasoning—you can optimize the model's performance to suit specific use cases. This approach eliminates the need for separate models for reasoning-intensive and simpler tasks, streamlining workflows and reducing operational overhead. Whether you're addressing intricate problem-solving scenarios or routine data processing, the model's adaptability ensures optimal performance. The ability to fine-tune the reasoning process provides a significant advantage, allowing you to allocate resources efficiently while achieving high-quality results. Cost-Efficiency and Competitive Pricing Gemini 2.5 Flash is designed with cost-conscious developers in mind, offering a pricing structure that reflects its focus on affordability and performance. The model's pricing tiers are as follows: Non-thinking mode: $0.60 per million tokens $0.60 per million tokens Thinking mode: $3.50 per million tokens This competitive pricing positions Gemini 2.5 Flash as a cost-effective alternative to other leading AI models, such as OpenAI and DeepSync. By integrating proprietary hardware and software, Google ensures a strong performance-to-cost ratio, making the model an attractive option for projects that require scalability without sacrificing quality. This balance between affordability and capability makes it a practical choice for developers aiming to optimize their resources. Gemini 2.5 Flash Hybrid Reasoning AI Model Watch this video on YouTube. Find more information on Hybrid Reasoning AI by browsing our extensive range of articles, guides and tutorials. Performance and Benchmark Comparisons In benchmark evaluations, Gemini 2.5 Flash ranks second overall on the Chatbot Arena leaderboard, trailing only OpenAI's O4 Mini in specific areas. However, it demonstrates significant improvements over its predecessor, Gemini 2.0 Flash, particularly in academic benchmarks. These advancements highlight the model's enhanced capabilities and its potential to deliver robust performance across various applications. While these results underscore its strengths, it is recommended that you test the model against your internal benchmarks to determine its suitability for your unique requirements. This hands-on evaluation will provide a clearer understanding of how Gemini 2.5 Flash can integrate into your workflows and meet your specific needs. Enhanced Token and Context Window Capabilities One of the standout features of Gemini 2.5 Flash is its enhanced token capacity, which significantly expands its utility for developers. The model supports: Maximum output token length: 65,000 tokens, making it ideal for programming tasks and applications requiring extensive outputs. 65,000 tokens, making it ideal for programming tasks and applications requiring extensive outputs. Context window: 1 million tokens, allowing the processing of large datasets or lengthy documents with ease. These enhancements provide a substantial advantage for handling complex inputs and generating detailed outputs. Whether you're working on data-heavy projects or applications requiring extensive contextual understanding, Gemini 2.5 Flash offers the tools necessary to manage these challenges effectively. Multimodal Processing for Diverse Applications Gemini 2.5 Flash extends its capabilities to multimodal processing, supporting a variety of input types, including video, audio, and images. This versatility makes it a valuable tool for industries such as media analysis, technical documentation, and beyond. However, it is important to note that the model does not include image generation features, which may limit its appeal for creative applications. Despite this limitation, its ability to process diverse input types enhances its utility across a wide range of use cases. Key Limitations to Consider While Gemini 2.5 Flash excels in many areas, it is not without its limitations. These include: Challenges with certain logical deduction tasks and variations of classic reasoning problems. A 'thinking budget' capped at 24,000 tokens, with no clear explanation for this restriction. Performance gains that plateau as token usage increases, indicating diminishing returns for resource-intensive tasks. These constraints highlight areas where the model may fall short, particularly for developers requiring advanced reasoning capabilities or higher token limits. Understanding these limitations is crucial for making informed decisions about the model's applicability to your projects. Strategic Value for Developers Google's Gemini 2.5 Flash reflects a strategic focus on cost optimization, scalability, and accessibility, making advanced AI technology available to a broader audience. Its hybrid reasoning capabilities, enhanced token and context window capacities, and multimodal processing features position it as a versatile and scalable tool for developers. By balancing quality, cost, and latency, the model caters to a wide range of applications, from data analysis to technical problem-solving. For developers seeking practical solutions that combine flexibility, performance, and affordability, Gemini 2.5 Flash offers a compelling option. Its ability to adapt to diverse tasks and optimize resource allocation ensures that it can meet the demands of modern AI challenges effectively. Media Credit: Prompt Engineering Filed Under: AI, Top News Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.
