Weekly Tech Recap: OpenAI releases o3 and o4 mini AI models, Samsung’s One UI 7 drama continues and more

With a lot of murmur throughout the week, it’s tough to keep up with the fast paced tech news cycle. To help declutter things, we’ve rounded up the Weekly Tech Recap—giving readers a quick look at the top five stories that made waves in the world of technology. This week, OpenAI rolled out new AI reasoning models for ChatGPT, Mark Zuckerberg was accused of collaborating with China on censorship tools, Samsung’s One UI 7 rollout drama continued, and more.

Top tech news of the week:

1) OpenAI rolls out o3 and o4 mini reasoning models:

Earlier this week, OpenAI launched its latest reasoning models, o3 and o4 mini, with the ability to “agentically use and combine any tool within ChatGPT”. One of the standout features of the new models is the ability to natively understand visual input and even generate images, leading to a new trend among social media users to reverse-engineer the location of an image.

According to OpenAI, o3 makes 20 per cent fewer errors than its previous state-of-the-art reasoning model, o1, on “difficult real-world tasks” such as programming, consulting and creative ideation. However, in another technical document for o3, the company also said that its new model was almost twice as likely to hallucinate (make things up) as o1.

After releasing a GPT-4.5 model earlier this year, OpenAI also released a GPT-4.1 model, but limited it only to developers. Sam Altman – CEO of the AI startup – realised the complexity of the model naming scheme and promised a fix soon, possibly hinting at the GPT-5 model that is supposed to unify all of ChatGPT’s underlying models.

2) Mark Zuckerberg accused of helping China:

Former Meta global public policy director Sarah Wynn-Williams made some serious allegations against the Mark Zuckerberg-led company earlier this week. In testimony on Capitol Hill earlier this week, Williams told senators that Meta worked “hand in glove” with the Chinese Communist Party to build a censorship system to silence critics on its platform.

Williams, who worked at the company from 2011 to 2017, said she watched as Meta executives decided to give the Chinese government access to Meta users’ data.

“I saw Meta executives repeatedly undermine US national security and betray American values. They did these things in secret to win favor with Beijing and build an $18 billion business in China. We are engaged in a high stakes AI right arms race against China. And during my time at meta, company, executives lied about what they were doing with the Chinese Communist Party to employees, shareholders, Congress and the American public.”

3) Samsung’s One UI 7 rollout drama:

Samsung reportedly postponed its One UI 7 rollout plans due to a problem with unlocking the phone on Exynos-powered Galaxy S24 devices in Korea, as well as possible battery drain issues. Soon afterwards, reports suggested that the Korean smartphone maker has resumed the updates for its latest UI, but there is still no clarity on when it will reach Indian users.

The tech giant had earlier announced its plan to start rolling out One UI 7 for Galaxy S24, Galaxy S23 and other devices from 7 April, but that plan seemed to have hit a snag before it even started.

4) Instagram rolls out Blend feature:

Instagram launched a new feature called ‘Blend’, which allows users to create a personalised reel feel with their friends. Like other similar features from the company, Blend works for both one-on-one conversations and group chats.

Blend is an invite-only feature that provides a combined recommendation of reels that the people in the conversation might like, and the content in this custom feed is updated daily.

The new feature allows users to see reels that are not currently part of their recommendation engine, but because their friends like these reels, there is a good chance they might like them too.

5) Apple’s plan to catch up with OpenAI and Google:

Apple is now stepping up its plan to catch up in the AI race by improving the efficiency of its basic model by analysing the data on customers’ devices, according to a Bloomberg report.

Notably, the iPhone maker is currently using synthetic data to train its AI models. This data mimics real-life user input without personal information, but isn’t always representative of actual user data.

To catch up with OpenAI and Google, Apple will use a new method to create “useful synthetic data” by comparing its existing data with recent samples of real-world emails from the iPhone, iPad and Mac email app.

By comparing the real emails with the fake inputs, Apple will then be able to determine which part of its data most closely matches the actual use case. The new method is expected to help the company improve some of its AI features, such as notification summaries and writing tools.