OpenAI’s latest version of ChatGPT, GPT o1, a nomenclature indicative of resetting the counter clock to 1, and its less costly mini version, represents a watershed moment in the company’s LLM stockpile. Designed to replicate superhuman-level intelligence, the models can already answer questions a lot faster than humans. This series of models will be unlike previous ones. In responding to queries, they utilise a human-like ‘chain of thought’ processing combined with reinforcement learning on specialised datasets and optimisation algorithms.
The model outperforms older models by a significant margin. For example, when tested against GPT-4o at the International Mathematics Olympiad, it scored 83 percent to GPT-4o’s 13 percent. What’s unique about the model is its ability to not only provide step-by-step reasoning for outputs but to show human-like patterns of hesitation during the process, ‘I’m curious about…’ and ‘Ok, let me see’ or ‘Oh, I’m running out of time, let me get to an answer quickly’. The new design has also resulted in a reduced occurrence of hallucinations. Yet, despite their many pros, the models have limitations. For instance, they cannot browse the internet, lack world knowledge, and cannot process files and images.
According to the lead researcher on the project, Jerry Tworek, the next level is for the models to perform similarly to PhD students on challenging benchmark tasks in areas such as physics, chemistry and biology. He assures that the intention here is not to equate AI with human thinking but rather to illustrate the model’s ability to dive cognitively deep. For the company, reasoning is a step up from pattern recognition, which is the design model used with previous versions. Ultimately, OpenAI aims to develop a product that can make decisions and take action on behalf of humans, a venture estimated to cost a further $USD 150 billion. Removing the current kinks in the system will mean that the models can work on complex global problems we face today in areas such as engineering and medicine.
More breakthroughs will also mean reduced access costs for developers and users. According to Chief Research Officer Bob McGrew, developer access to o1-preview is currently $15 per 1 million input tokens (chunks of text parsed by the model) and $60 per 1 million output tokens. GPT -o4 costs $5 per 1 million input tokens and $15 per 1 million output tokens.
Meta’s decision to change how it labels AI-modified content on Instagram, Facebook, and Threads signifies another advancement in the company’s approach to generative AI. The visibility of AI’s involvement is reduced by moving the ‘AI info’ label to the post’s menu for content that has been edited with AI tools. This could make it easier for users to overlook or miss the AI editing details in such posts.
However, for content fully generated by AI, Meta will continue to prominently display the label beneath the user’s name, ensuring that posts created entirely by AI prompts remain visibly marked. The distinction Meta is making here seems to reflect the varying degrees of AI involvement in content creation.
Meta aims to increase transparency about content labelling, specifying if AI designation is from industry signals or self-disclosure. This effort follows complaints and confusion over the previous ‘Made with AI’ label, particularly from photographers concerned that their real photos were misrepresented.
This change may raise concerns about the potential for users to be misled, especially as AI editing tools become more sophisticated and the line between human and AI-created content continues to blur. It highlights the need for continued transparency as AI technology integrates more deeply into content creation across platforms.
Taiwan is now using AI to track and predict the path of tropical storms, including the approaching storm Bebinca. AI-powered models, such as those from Nvidia and other tech companies, are outperforming traditional methods. The Central Weather Administration (CWA) has found these tools especially useful, providing more accurate forecasts that give forecasters greater confidence in predicting storm paths.
In July, AI models helped Taiwan predict Typhoon Gaemi’s path and impact, delivering early warnings eight days before landfall. Technology like this one significantly outperformed conventional methods, accurately forecasting record rainfall and giving authorities more time to prepare. The AI-based system allowed Taiwan to anticipate a rare loop in Gaemi’s path, which prolonged its effects on the island.
While AI weather forecasting models have delivered impressive results, experts say more time is needed for the technology to fully surpass traditional methods in predicting typhoon strength and wind speeds. AI has already proven its worth in predicting storm tracks and could revolutionise weather forecasting globally.
Despite some limitations, AI’s increasing role in weather prediction is promising. Taiwan’s weather service forecasters hope ongoing partnerships with companies like Nvidia will enhance these tools, potentially leading to even more accurate predictions in the future.
Researchers at Western University have developed an AI model that detects strawberry diseases and predicts ripeness with nearly 99% accuracy. The system, designed by Joshua Pearce and Soodeh Nikan, could significantly enhance crop quality and reduce waste. Tested in a controlled hydroponic environment, the technology aims to extend Canada’s strawberry growing season while improving fruit quality.
The model is free and open-source, enabling farmers to tailor it to their needs. It can notify them via email or phone when diseases are detected or fruit is ripe. This adaptable AI system could prove crucial for increasing agricultural efficiency.
By minimising food waste and lowering production costs, the AI model has the potential to reduce grocery prices for consumers. Researchers hope the technology will support food security and help farmers meet growing demands for fresh produce.
Future plans involve testing the AI outdoors, possibly with drones monitoring larger fields. The innovation could bring smarter, more sustainable farming to outdoor environments, further boosting efficiency in agriculture.
Researchers are currently developing AI tools to help predict and manage future pandemics, which some experts believe will likely within the next decade. Teams from UC Irvine and UCLΑ, part of the US National Science Foundation’s Predictive Intelligence for Pandemic Prevention grant programme, are working on an AI-based early warning system that analyses social media posts to detect early signs of outbreaks. They aim to track billions of posts on platforms like X (formerly Twitter) to identify public health trends and assess the potential outcomes of public health policies. However, the reliance on specific platforms and US-focused data limits its global application. Researchers are working to expand its reach.
Harvard Medical School and the University of Oxford have created a tool called EVEScape, which predicts virus mutations. This tool helps in developing vaccines and treatment strategies. Pharmaceutical companies such as AstraZeneca are also utilising AI to accelerate the discovery of antibodies, which could potentially reduce the response time to new viral threats. These initiatives demonstrate how AI can enhance pandemic response by providing faster and more accurate data for decision-making.
“Despite its potential, experts warn that the effectiveness of AI depends on the quality of the data it receives. Biases or misrepresentations in the data could lead to skewed results, and there are ethical and fairness concerns. Although AI can improve preparedness and response times, human judgement, trust, and collaboration are essential for effectively managing future pandemics.”
WhatsApp is set to enhance its AI features with a two-way voice chat option. Users will soon interact with the Meta AI chatbot using voices of public figures, including well-known celebrities. The update will allow for more personalised and engaging communication experiences.
A recent beta update revealed that the voice feature will offer a range of options, including different accents and pitches. Users can select from various voices, possibly from both UK and US accents, though exact details remain unclear. The feature is designed to add a custom touch to AI interactions.
Meta previously introduced AI personalities on Messenger that mirrored celebrities and influencers. The new voice chat feature on WhatsApp builds on those efforts, bringing further AI-driven experiences to its user base.
Upon launch, the feature will display a simple interface with a prominent ‘Meta AI’ label, providing easy access to the voice options. Lastly, this marks another step forward for WhatsApp in delivering innovative AI solutions for users.
O’Reilly has launched its AI Academy, designed to help businesses upskill their workforce in generative AI technology. The Academy offers hundreds of learning materials, including books, live events, and on-demand courses, to enhance productivity through GenAI tools.
The demand for GenAI skills has surged, with global executives planning to invest more in AI technologies. Despite this, a vast number of workers remain untrained in the tools required to implement GenAI. Only 10% of workers have gained these skills, and confidence in executives’ understanding of AI remains limited, highlighting a clear need for upskilling.
O’Reilly’s AI Academy addresses this gap, providing tailored learning tracks to meet specific industry needs. Courses focus on essential skills such as productivity enhancement and AI integration, covering roles from HR to project management. Additional role-specific tracks will be introduced, enabling even greater personalisation.
Completing these learning tracks earns participants badges and certificates, which can be shared on platforms like LinkedIn. This helps employees showcase their expertise in GenAI, empowering them to adapt and drive business outcomes in an AI-driven world.
Connectly, a startup specialising in conversational commerce through AI-driven personalised messaging has secured $20 million in a Series B funding round. The round was led by Alibaba and included participation from several notable investors, such as Unusual Ventures and Volpe Capital. This new investment boosts Connectly’s total funding to $37.2 million and brings its valuation close to $100 million.
The funds will be used to advance AI research and support Connectly’s expansion into the US and European markets. Additionally, the company plans to strengthen its engineering presence in Greece, aiming to make it a key hub alongside San Francisco. Connectly, a company that uses AI models to help retailers enhance customer engagement and drive sales, has experienced significant growth in the past year.
The successful funding round follows Connectly’s launch of its advanced AI recommendation tool, ‘Sofia AI,’ and its expansion into the US market. The partnership with Alibaba is expected to accelerate Connectly’s global reach further, integrating its AI solutions into Alibaba’s international e-commerce platforms. With plans to grow its workforce to 80 by year-end and a current client base of 300, Connectly is well-positioned to continue its impactful growth in the retail industry.
Adobe has announced the upcoming release of a generative AI-powered video creation tool named Adobe Firefly Video Model. Scheduled for a limited beta release later this year, this tool will extend Adobe’s Firefly suite, which currently includes applications for generating still images and designs. The new model will allow users to create a five-second video clip from a single text or image prompt, with options to specify camera angles, motion, and zoom.
The introduction of this tool marks Adobe’s entry into the competitive AI video generation market, which already features offerings from companies like OpenAI and Stability AI. Adobe aims to differentiate itself by focusing on quality and user-guided prompt understanding, addressing specific needs of videographers.
Adobe assures that the model is trained exclusively on public domain or licensed content from its Adobe Stock database, which includes 400 million curated images and videos, avoiding any intellectual property issues. Additionally, Adobe is launching Generative Extend, a feature for Premiere Pro that extends video clips by generating content to fill gaps.
The US government is reportedly considering allowing Nvidia to export advanced AI chips to Saudi Arabia. These chips would assist the kingdom in developing and operating cutting-edge models. The move could play a crucial role in Saudi Arabia’s AI strategy, which was a key focus at the recent GAIN summit.
Efforts are underway in Saudi Arabia to meet US security requirements, which could expedite the acquisition of Nvidia’s H200 chips. These chips are expected to boost Saudi Arabia’s capabilities, as they are also used in advanced platforms like OpenAI’s GPT-4. Saudi officials have expressed their intention to comply with US regulations.
The Biden administration had imposed restrictions on AI chip exports, particularly targeting China, but also extending to the UAE and other Middle Eastern countries. However, Saudi Arabia has been careful to manage its relationship with both the US and China, ensuring access to key technologies remains open.
Nvidia and the US Department of Commerce declined to comment on the potential chip sales. The Department of Commerce noted that export control decisions involve multiple government departments, including Defense, State, and Energy.
