OpenAI

Sam Altman admits OpenAI holds back stronger AI models

OpenAI recently unveiled GPT-5, a significant upgrade praised for its advances in accuracy, reasoning, writing, coding and multimodal capabilities. The model has also been designed to reduce hallucinations and excessive agreeableness.

Chief executive Sam Altman has admitted that OpenAI has even more powerful systems that cannot be released due to limited capacity.

Altman explained that the company must make difficult choices, as existing infrastructure cannot yet support the more advanced models. To address the issue, OpenAI plans to invest in new data centres, with spending potentially reaching trillions of dollars.

The shortage of computing power has already affected operations, including a cutback in image generation earlier in the year, following the viral Studio Ghibli-style trend.

Despite criticism of GPT-5 for offering shorter responses and lacking emotional depth, ChatGPT has grown significantly.

Altman said the platform is now the fifth most visited website worldwide and is on track to overtake Instagram and Facebook. However, he acknowledged that competing with Google will be far harder.

OpenAI intends to expand beyond ChatGPT with new standalone applications, potentially including an AI-driven social media service.

The company also backs Merge Labs, a brain-computer interface rival to Elon Musk’s Neuralink. It has partnered with former Apple designer Jony Ive to create a new AI device.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

GPT-5 impresses in reasoning but stumbles in flawless coding

OpenAI’s newly released GPT-5 draws praise and criticism in equal measure, as developers explore its potential for transforming software engineering.

Launched on 7 August 2025, the model has impressed with its ability to reason through complex problems and assist in long-term project planning. Yet, engineers testing it in practice note that while it can propose elegant solutions, its generated code often contains subtle errors, demanding close human oversight.

Benchmark results showcase GPT-5’s strength. The model scored 74.9% on the SWE-bench Verified test, outperforming predecessors in bug detection and analysis. Integrated into tools such as GitHub Copilot, it has already boosted productivity for large-scale refactoring projects, with some testers praising its conversational guidance.

Despite these gains, developers report mixed outcomes: successful brainstorming and planning, but inconsistent when producing flawless, runnable code.

The rollout also includes GPT-5 Mini, a faster version for everyday use in platforms like Visual Studio Code. Early users highlight its speed but point out that effective prompting remains essential, as the model’s re-architected interaction style differs from GPT-4.

Critics argue it still trails rivals such as Anthropic’s Claude 4 Sonnet in error-free generation, even as it shows marked improvements in scientific and analytical coding tasks.

Experts suggest GPT-5 will redefine developer roles rather than replace them, shifting focus toward oversight and validation. By acting as a partner in ideation and review, the model may reduce repetitive coding tasks while elevating strategic engineering work.

For now, OpenAI’s most advanced system sets a high bar for intelligent assistance but remains a tool that depends on skilled humans to achieve reliable outcomes.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Employees of OpenAI eye multi-billion dollar stock sale

According to a source familiar with the talks, OpenAI employees are preparing to sell around $6 billion worth of shares to major investors, including SoftBank Group and Thrive Capital.

The deal, still at an early stage, would push the company’s valuation to $500 billion, up from its current $300 billion.

SoftBank, Thrive and Dragoneer Investment Group are already among OpenAI’s backers, and their participation in the secondary share sale would further strengthen ties with the Microsoft-supported AI company.

Reports suggest the size of the sale could still change as discussions continue.

The planned deal follows SoftBank’s leadership role in OpenAI’s $40 billion primary funding round earlier this year. Employee share sales often reflect strong investor demand and highlight the rapid growth of companies in competitive markets.

OpenAI has seen user numbers and revenues soar in 2025, with weekly active ChatGPT users climbing to about 700 million, up from 400 million in February.

The company doubled its revenue in the first seven months of the year, hitting an annualised run rate of $12 billion, and is expected to reach $20 billion by the end of the year.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

New OpenAI hire shares savvy interview strategies

Bas van Opheusden, who joined OpenAI as a technical staff member in July, has published a comprehensive eight-page guide for aspiring applicants, offering strategic advice spanning recruiter calls, coding interviews, compensation discussions and more.

He suggests treating recruiter conversations as strategic briefings, which are key for understanding the hiring manager’s priorities, team dynamics, role expectations, and organisational goals.

Van Opheusden recommends taking notes during calls, ideally using a dual-screen setup, and arranging windows so it appears you’re maintaining eye contact.

He also shared a standard error: arriving at coding interviews without remembering the exact role he’d applied for, underscoring the importance of clear preparation and role alignment.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

OpenAI’s GPT-5 faces backlash for dull tone

OpenAI’s GPT-5 launched last week to immense anticipation, with CEO Sam Altman likening it to the iPhone’s Retina display moment. Marketing promised state-of-the-art performance across multiple domains, but early user reactions suggested a more incremental step than a revolution.

Many expected transformative leaps, yet improvements mainly were in cost, speed, and reliability. GPT-5’s switch system, which automatically routes queries to the most suitable model, was new, but its writing style drew criticism for being robotic and less nuanced.

Social media buzzed with memes mocking its mistakes, from miscounting letters in ‘blueberry’ to inventing US states. OpenAI quickly reinstated GPT-4 for users who missed its warmer tone, underlining a disconnect between expectations and delivery.

Expert reviews mirrored public sentiment. Gary Marcus called GPT-5 ‘overhyped and underwhelming’, while others saw modest benchmark gains. Coding was the standout, with the model topping leaderboards and producing functional, if simple, applications.

OpenAI emphasised GPT-5’s practical utility and reduced hallucinations, aiming for steadiness over spectacle. At the same time, it may not wow casual users, its coding abilities, enterprise appeal, and affordability position it to generate revenue in the fiercely competitive AI market.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Musk faces an OpenAI harassment lawsuit after a judge rejects dismissal

A federal judge has rejected Elon Musk’s bid to dismiss claims that he engaged in a ‘years-long harassment campaign’ against OpenAI.

US District Judge Yvonne Gonzalez Rogers ruled that the company’s counterclaims are sufficient to proceed as part of the lawsuit Musk filed against OpenAI and its CEO, Sam Altman, last year.

Musk, who helped found OpenAI in 2015, sued the AI firm in August 2024, alleging Altman misled him about the company’s commitment to AI safety before partnering with Microsoft and pursuing for-profit goals.

OpenAI responded with counterclaims in April, accusing Musk of persistent attacks in the press and on his platform X, demands for corporate records, and a ‘sham bid’ for the company’s assets.

The filing alleged that Musk sought to undermine OpenAI instead of supporting humanity-focused AI, intending to build a rival to take the technological lead.

The feud between Musk and Altman has continued, most recently with Musk threatening to sue Apple over App Store listings for X and his AI chatbot Grok. Altman dismissed the claim, criticising Musk for allegedly manipulating X to benefit his companies and harm competitors.

Despite the ongoing legal battle, OpenAI says it will remain focused on product development instead of engaging in public disputes.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Santander expands AI-first strategy with OpenAI

Santander is accelerating its AI-first transformation through a new partnership with OpenAI, aiming to embed intelligent technology into every part of the bank.

Over the past two months, ChatGPT Enterprise has been rolled out to nearly 15,000 employees across Europe and the Americas, with plans to double that number by year-end. The move forms part of a broader ambition to become an AI-native institution where all decisions and processes are data-driven.

The bank will plan a mandatory AI training programme for all staff from 2026, with a focus on responsible use, and expects to scale agentic AI to enable fully conversational banking.

Santander says its AI initiatives saved over €200 million last year. In Spain alone, speech analytics now handles 10 million calls annually, automatically updating CRM records and freeing more than 100,000 work hours. Developer productivity has risen by up to 30% on some tasks.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Anthropic adds recall ability to Claude

Anthropic has added a user-triggered memory function to its Claude chatbot, allowing it to search and summarise previous chats on request. The feature helps users resume projects without repeating themselves.

The upgrade works across web, desktop and mobile platforms and is currently available to Max, Team and Enterprise subscribers, with wider rollout planned.

Claude’s memory does not automatically store personal profiles. Instead, when prompted, it retrieves relevant past chats, prioritising user privacy while enhancing usability.

With this feature, Anthropic aims to make Claude more competitive against rivals like ChatGPT by improving AI continuity in user experience across sessions.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

GPT-5 doubles usage limits and adds smarter features

OpenAI has rolled out GPT-5 as the default AI model powering ChatGPT, bringing new features designed to boost productivity for personal and business users.

The new model seamlessly switches between quick search and in-depth reasoning, allowing more fluid and intelligent responses. Users can prompt ChatGPT to ‘think hard’ to trigger the deeper reasoning mode.

ChatGPT Plus users now benefit from double the previous message limit, with 160 messages allowed every three hours. Meanwhile, Team and Pro plan subscribers enjoy unlimited GPT-5 access unless accounts are misused.

Free users have a limit of 10 messages every five hours and one daily ‘Thinking’ mode message. Older GPT models such as GPT-4.1 and GPT-3 have been discontinued but remain accessible via web settings for paying customers.

All built-in tools are automatically enabled according to user needs, removing the need to toggle features like web search, image generation, or data analysis on and off. OpenAI also revealed plans to support third-party plugins to expand ChatGPT’s development capabilities further.

The new voice mode now follows instructions more accurately and will be available to all users.

Overall, GPT-5 marks a significant leap forward, improving reasoning, creativity, and alignment with user intent. OpenAI aims to make ChatGPT an even more powerful assistant by integrating enhanced capabilities and streamlining the user experience.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Rollout of GPT-5 proves bumpier than expected

OpenAI’s highly anticipated GPT-5 has encountered a rough debut as users reported that it felt surprisingly less capable than its predecessor, GPT-4o.

The culprit? A malfunctioning real-time router that failed to select the most appropriate model for user queries.

In response, Sam Altman acknowledged the issue and assured users that GPT-5 would ‘seem smarter starting today’.

To ease the transition, OpenAI is restoring access to GPT-4o for Plus subscribers and doubling rate limits to encourage experimentation and feedback gathering.

Beyond technical fixes, the incident has sparked broader debate within the AI community about balancing innovation with emotional resonance. Some users lament GPT-5’s colder tone and tighter alignment, even as developers strive for safer, more responsible AI behaviour.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!