Developers of artificial intelligence, such as Microsoft, often express the belief that it it should serve the good of humanity from the outset. It is debatable whether the military’s use of AI meets this criterion – especially when it comes to so-called lethal technologies. Google has also taken its clear stance on this issue, and no longer rules out the use of its AI technology for military purposes.
Google has stirred up quite a controversy by changing its guidelines for its artificial intelligence, including lifting the ban on its use for military purposes, including the creation of weapons and surveillance tools, including, for example, intelligent drones. Human Rights Watch expressed its concern, arguing that such a decision could “complicate accountability” for decisions on the battlefield, the consequences of which are “a matter of life and death.”
Alphabet (Google’s parent company) defends the change, saying that companies and democratic governments need to work together on AI that “supports national security.” Anna Bacciarelli of Human Rights Watch, however, finds the abandonment of previously established “red lines” troubling, especially at a time when there is a need for responsible leadership in the AI domain. Alphabet writes in a blog post that democracies should lead AI development guided by “core values” such as freedom, equality and respect for human rights. They believe that companies, governments and organizations sharing these values should work together to create AI that protects people, promotes global development and supports national security.
The military potential of artificial intelligence raises legitimate concerns. AI systems, capable of making life-or-death decisions autonomously, need urgent scrutiny. AI has already found military applications in Ukraine and the Middle East, and several countries are pursuing its integration into their armed forces. The question remains about the extent to which machines will make military decisions, including those that could lead to mass deaths.
Even before ethical AI became a topic of public discourse, Google founders Sergei Brin and Larry Page adopted the motto “Don’t be evil” (“Don’t be evil”). When the company was transformed into Alphabet Inc in 2015, it was changed to “Do the right thing” (“Do the right thing”). However, who has the right to determine what is the right thing? Since then, Google employees have sometimes objected to the board’s approach. In 2018, the company did not renew a contract for AI work for the Pentagon, probably prompted by resignations and a petition signed by thousands of employees who feared that the Maven project was the first step toward using AI for deadly purposes.
OpenAI introduces Deep Research, an AI agent to create deeper reports
OpenAI introduces Deep Research. It is a new AI agent which uses the inference capabilities of the o3 model, which can synthesize large amounts of information from the Internet. ChatGPT users will soon be able to use this to perform multi-step research tasks. “In-depth research” takes much longer than answering ordinary chatbot queries.
It usually takes people many hours to create a research report. Hundreds of online sources need to be collected, analyzed and synthesized. Deep Research from OpenAI can accomplish this task in minutes. This new feature has been trained on real-world tasks requiring the use of Python and a browser. The final report from Deep Research will include clear citations and a summary of its own thoughts. The publisher assures that this feature will be very effective in finding niche, non-intuitive information that requires browsing multiple pages to find.
To use this feature in ChatGPT, select “Deep Research” in the message writing box and enter your queries. Describe anything you need – whether it should be a competitive analysis on streaming platforms or a personalized report on the best city bike, for example. You can attach files, including worksheets, to add context. When you run the feature, the sidebar will begin to display the steps and sources used. Deep Research runs for 5 to 30 minutes, so it does a lot more work than usual at one time (ChatGPT and Microsoft Copilot usually return answers after a few seconds).
Compared to Deep Research GPT-4o is ideal for real-time multimodal conversations. For multi-faceted, domain-specific queries where depth and detail are critical, Deep Research’s ability to perform extensive exploration and citation of each assertion makes the difference between a quick summary and a well-documented, verified answer that can be used as a work product. – OpenAI
The publisher stresses that Deep Research’s operation is currently very computationally intensive. Currently, ChatGPT Pro subscribers can use it (up to 100 queries per month). Next in line will be users of the Plus, Team and Enterprise plans. Work is underway to make it available in the UK, Switzerland and the European Economic Area. This AI feature is only available in the browser, and will appear in smartphone and PC apps within a month.
Google’s biggest fear? ChatGPT Search available to all without login
In October 2024, OpenAI released a ChatGPT Search, a feature that allows you to search the Internet directly in ChatGPT. This allows you to quickly get answers with links to relevant sources, something for which a separate search engine was previously needed. In addition, this integration gives ChatGPT access to up-to-date (rather than time-limited) knowledge.
ChatGPT search has been available to the public since December, although initially only to those logged in. This meant that it was not possible to just jump into ChatGPT from the Internet and start a search. In addition to that, previously users had to click a search button to start a search. Now, the application automatically decides whether to trigger a search on the Web or rely on the artificial intelligence when providing answers. However, you can still force ChatGPT to search by clicking the “Search” icon.
OpenAI announced yesterday on Twitter (X) that ChatGPT search is now available to everyone on chatgpt.com – login is not required. This is a significant convenience for users who don’t have to create an account or enter login information, such as when they are on a different device or in a different browser.
That’s not the only new feature in ChatGPT Search. The feature has recently moved closer to conventional search engines, showing maps, images and even descriptions of tourist attractions anywhere. According to a recent Semrush report, ChatGPT is gradually reducing Google’s share of search traffic to external sites.
Real-time speech processing with GPT-4o in Azure OpenAI Service
Microsoft has released the new GPT-4o-Mini-Realtime-Preview and GPT-4o-Mini-Audio-Preview artificial intelligence models, both in familiarization. They introduce advanced audio processing capabilities at only 25% of the cost of the GPT-4o audio models. This increases the potential of AI applications for text- and voice-based interactions.
Developers can now unlock immersive, voice-guided experiences using the advanced capabilities of all sound models Azure OpenAI Service, available in the public preview version. Examples of applications include:
- Customer service: virtual assistants and voice chatbots can handle customer inquiries more naturally and efficiently, reducing wait times and improving overall satisfaction.
- Content creation: content creators can revolutionize their work by using speech generation in video games, podcasts or film studios.
- Real-time translation: industries such as healthcare and legal services can benefit from live translation to break down language barriers and provide better communication.
Both new models provide high-quality audio interactions at a fraction of the price of the sound models GPT-4o. They are also compatible with the existing Realtime API and Chat Completion API. What are the differences, on the other hand?
- GPT-4o-Mini-Realtime-Preview: provides natural real-time voice interactions for a more engaging user experience. This model will be ideal for applications that require instant responses, such as customer service chatbots and virtual assistants.
- GPT-4o-Mini-Audio Preview: provides high-quality audio interactions at a lower cost. It is an ideal model for applications requiring asynchronous audio processing, such as for analyzing the moods in a recording or creating content using text-to-audio conversion.
As you can see, the models available in the Azure OpenAI Service allow both to process speech virtually in real time, as well as to work with speech and audio in scenarios where lightning-fast performance is not so important. What seems important, on the other hand, is the ability to harness these models to your own applications via APIs, so that you can turn them into so-called “speech-enabled” applications. AI Driven Apps.
