
OpenAI's story about grief nearly had me in tears, but for all the wrong reasons
Like all parents who pretend to be impressed by their children's terrible art, OpenAI CEO Sam Altman proudly announced to the world that the company's new AI model is gifted at creative writing. 'This is the first time I have been really struck by something written by AI,' he enthused on X.
The prompt was to write a metafictional literary short story about AI and grief. The story closely follows the instructions. The individual sentences mostly make sense. But – with the greatest respect to Jeanette Winterson, who called the story 'beautiful and moving' – it is an atrocious piece of writing.
The AI captures the tone of some of the worst writing around: pompous and self-satisfied, while using sentimental imagery ('a girl in a green sweater who leaves home with a cat in a cardboard box') and dull similes ('like a stone dropped into a well'). There is repetition without development, poor structure, an over-reliance on jargon and, crucially, a lack of that slight breath of madness that makes writing human. If this were a first-year creative writing student, you'd give them feedback to help them improve, but you probably wouldn't discourage them from pursuing other job prospects.
It is in the gaps between what the author has written and what the reader imagines that writing comes to life. Readers will project meaning on to words on a page, but if those gaps are created by algorithms and chance, the act of creation on the reader's part becomes onanistic; the death of the author requires the author to have been alive at one stage.
I asked another AI model to critically evaluate the story, and it found it 'compelling' and 'self-aware', employing 'evocative language and imagery'. Perhaps it was not written for human eyes at all.
Sign up to Observed
Analysis and opinion on the week's news and culture brought to you by the best Observer writers
after newsletter promotion
In the furore surrounding the Labour party's devastating welfare cuts – which are estimated to plunge 700,000 households into poverty and have led to widespread rebellion among MPs – we are at risk of overlooking a short video recently released by the Department for Work and Pensions, which exhorts disabled people to get back into the workforce.
'The welfare system we inherited is broken,' a voice intones grimly over footage of a man and a woman entering a job centre. Unfortunately, the building's even red bricks and cast iron archway carry a very specific historical connotation: the gates of Auschwitz, which bear the inscription Arbeit Macht Frei, or 'work sets you free'. During the Holocaust, an estimated 250,000 people with mental and physical disabilities were murdered, as they did not comply with the regime's vision of a master race and were considered a burden to society.
The similarity in the video was unintentional. But the fact no one in the department seemed to notice the resemblance is deeply worrying.
Congratulations to Whitetop, a sprightly 27-year-old who has been named by the Guinness World Records as the oldest llama in captivity (the previous record holder, Albuquerque ranch resident Dalai Llama, passed away in 2023). For nearly two decades, Whitetop has been providing comfort and support to chronically ill children at a camp in North Carolina. 'He's just a really cool dude and loves his job,' says Billie Jo Davis, the camp's barn director.
Let's hope Whitetop has a long and healthy reign. Though perhaps, following the sad fate of Geronimo – the alpaca euthanised in 2021 which may or may not have had bovine TB – it might be a while before Keir Starmer is invited to visit.
Kathryn Bromwich is a commissioning editor and writer on the Observer New Review

Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles


Geeky Gadgets
4 hours ago
- Geeky Gadgets
OpenAI Codex Updates and Agent API Updates : Now Available for Plus Users
OpenAI has announced new updates to its Codex and its Agent API, enhancing accessibility, functionality, and safety for developers. These updates include expanded access to Codex, new internet-enabled capabilities, improved agent development tools, and advancements in voice agent technology. OpenAI's Codex, an AI-powered tool designed to generate and execute code, is now available to a broader audience. Previously limited to Enterprise, Team, and Pro users, Codex has been extended to ChatGPT Plus subscribers. This move aligns with OpenAI's mission to provide widespread access to advanced AI tools, allowing a wider range of developers to use its capabilities for diverse applications. Key improvements to Codex include the introduction of controlled internet access during task execution. This feature allows developers to perform tasks such as installing dependencies, testing scripts that require staging servers, and executing complex workflows. To address potential risks, internet access is disabled by default and governed by strict safeguards, including: – Domain restrictions to limit access to specific websites. – HTTP method limitations to control the types of requests made. – Prompt injection monitoring to detect and mitigate malicious inputs. These measures ensure that developers can innovate securely while maintaining control over their environments, balancing functionality with safety. Agent API and SDKs: New Tools and Real-Time Capabilities The Agent API has undergone substantial upgrades, particularly in its development tools. The Agents SDK now supports TypeScript, achieving parity with the existing Python SDK. This addition broadens the programming options available to developers, making it easier to create AI agents with advanced features such as: – Handoffs for seamless transitions between automated and human interactions. – Guardrails to enforce safety and compliance. – Tracing for monitoring agent activity. – Human-in-the-loop approvals to allow human oversight during critical decision-making processes. These tools streamline the development process, allowing faster and more secure deployment of AI agents across various industries. One of the most notable updates is the introduction of the RealtimeAgent feature. This capability allows developers to build voice agents that operate in real time, either on the client or server side. RealtimeAgents come equipped with advanced functionalities, including: – Automated tool calls to perform tasks dynamically. – Safety guardrails to prevent misuse and ensure ethical operation. – Seamless handling of audio input/output and interruptions for smoother interactions. By integrating these features, the RealtimeAgent enhances the practicality and reliability of voice-based AI systems, opening up new possibilities for real-world applications such as customer service, virtual assistants, and accessibility tools. Watch this video on YouTube. Monitoring and Managing AI Agent Performance To help developers optimize the performance of their AI agents, OpenAI has introduced the Traces Dashboard. This tool provides a detailed visualization of Realtime API sessions, offering insights into key metrics such as: – Audio input/output performance. – Tool usage during interactions. – Interruptions and how they are handled. By giving developers a clear view of agent performance, the Traces Dashboard helps identify and address potential issues, making sure smoother operation and improved outcomes. This level of transparency and control is particularly valuable for developers working on complex or high-stakes applications. Additionally, the Speech-to-Speech model has been updated to improve its reliability in areas such as instruction following, tool calling, and handling interruptions. The latest version, `gpt-4o-realtime-preview-2025-06-03`, is now available through both the Realtime API and Chat Completions API. These updates enhance the model's ability to assist seamless voice-to-voice communication, further expanding its utility in diverse scenarios, including multilingual communication and real-time translation. Safety and Oversight: A Core Priority Safety remains a cornerstone of OpenAI's approach to AI development. The latest updates include robust guardrails designed to prevent misuse and ensure ethical operation. Key safety measures include: – Prompt injection monitoring to protect against malicious inputs that could compromise system integrity. – Human-in-the-loop mechanisms to allow human operators to intervene when necessary, adding an extra layer of oversight. – Domain and method restrictions to limit the scope of internet access and reduce potential vulnerabilities. These safeguards reflect OpenAI's dedication to responsible AI deployment, balancing innovation with accountability. By prioritizing safety, OpenAI aims to build trust in its technologies while allowing developers to explore new possibilities with confidence. Advancing AI Development with Practical Applications The updates to OpenAI's Codex and Agent API represent a significant advancement in the field of AI technology. By broadening access to Codex, introducing real-time capabilities through the RealtimeAgent, and enhancing safety mechanisms, OpenAI continues to empower developers to create innovative solutions. These tools are designed to address the challenges of integrating AI into practical applications, offering developers the resources they need to build systems that are both effective and responsible. The combination of expanded functionality, real-time interaction capabilities, and robust safety measures positions OpenAI's tools as valuable assets for developers across industries. Whether used for automating workflows, improving customer interactions, or allowing accessibility, these updates highlight the growing potential of AI-driven solutions to address real-world needs. By focusing on accessibility, functionality, and safety, OpenAI sets a benchmark for responsible AI development. These updates not only expand the potential of AI technologies but also underscore the importance of ethical considerations in their deployment. As AI continues to evolve, tools like Codex and the Agent API will play a critical role in shaping the future of technology and its applications. Advance your skills in AI code generation. by reading more of our detailed content. 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.


Time Out
5 hours ago
- Time Out
The top London theatre shows according to our critics
What is it? Transferring from the Open Air Theatre after an acclaimed run last summer, rising star US director Jordan Fein's take on the classic shtetl-set musical transfers to the Barbican ahead of a UK tour. Where is it? Barbican Centre. How much is it? £25-£165. Why go? Fein masterfully balances the musical's two sides – the stand-up-style humour and the dark forshadowing of the Holocaust – by playing it as a sort of absurdist clown show. US star Adam Dannheisser is excellent as a restrained, dignifiede take on milkman protagonist Teyve.


Reuters
16 hours ago
- Reuters
OpenAI to appeal New York Times suit demand asking to not delete any user chats
June 5 (Reuters) - OpenAI said on Thursday that it is appealing The New York Times' (NYT.N), opens new tab lawsuit demanding that the ChatGPT-maker retain consumer ChatGPT and API customer data indefinitely.