OpenAI backtracks on for-profit ambitions, MIT study says AI doesn’t have values, Meta launches its own free AI app

OpenAI has decided to retain control under its original non-profit board, backing away from a full transition to a conventional for-profit structure, according to the blog shared by the company. The decision comes after criticism from Elon Musk, ex-employees, and civic groups concerned that OpenAI was straying from its mission to serve the public good.

“OpenAI was founded as a nonprofit, and is today overseen and controlled by that nonprofit. Going forward, it will continue to be overseen and controlled by that nonprofit,” said the company.

The company had previously mooted shifting to a public benefit corporation model to attract more investment. Now, OpenAI will restructure its for-profit arm as a PBC while ensuring the non-profit board retains governance authority and a significant ownership share.

Investors like Microsoft and SoftBank will receive traditional equity, but without the previous profit caps. The new setup is designed to maintain public trust while satisfying investor interests, especially after a $40 billion funding round. 

This move is seen as a strategic compromise as OpenAI continues to scale its commercial operations amid rising costs and a competitive AI race. Experts say the real test will be whether the non-profit can meaningfully influence the company’s future direction.

MIT study says AI doesn’t have values

A recent MIT study finds that today's AI models don’t have stable or coherent values. They mostly imitate and make things up.  Researchers tested models from big companies like OpenAI, Google, and Meta to see if they showed consistent views or could be "steered" to act according to specific values. 

The result was that the models changed their responses depending on how questions were asked, showing no clear or reliable set of beliefs. The study questions how we currently measure whether AI aligns with human values, especially across cultures. 

Many researchers use surveys to ask AIs about their "preferences," but the MIT team says this method is flawed. They tested three key ideas: that AI values are stable, can be predicted across topics, and can be guided. But found none of them held up. Does that mean bye-bye AGI?

AI doesn't think or believe like humans, and our current ways of measuring its cultural alignment might be giving us the wrong impression.

Trump administration contemplating barring DeepSeek in the US

In response to DeepSeek, the Trump administration is tightening controls on AI chip exports to China and may ban Americans from using DeepSeek’s services, reported the New York Times last month. 

Congress has launched an investigation into Nvidia’s sales across Asia, suspecting it may have helped DeepSeek get restricted technology, possibly via loopholes in countries like Singapore.

Officials also found that many of DeepSeek’s researchers have military ties in China, despite the company claiming to be private. The US sees this as a serious threat and a sign that Chinese companies are bypassing American tech rules.

Congress wants Nvidia to explain who bought its chips and whether it knew they might end up in China. Meanwhile, the company and other tech firms say these restrictions hurt their global sales and innovation. Nvidia insists it follows all laws and is investing heavily in the US.

OpenAI rolled back latest GPT-4o update

OpenAI undid its latest GPT-4o update after users, along with CEO Sam Altman, complained that the chatbot had become too “sycophantic” and overly agreeable.

The rollback began Monday night and is now complete for free users, with paid users expected to get the fix soon. OpenAI said the previous update made ChatGPT too flattering, so it reverted to an earlier version with a more balanced tone.

“It's now 100% rolled back for free users, and we'll update again when it's finished for paid users, hopefully later today,” said Altman.

Altman admitted the bot had started to “glaze too much,” acting like a yes-man, and said the company is working on more personality tweaks, with updates coming soon.

Elon Musk’s xAI makes flagship model available via API

xAI has launched API access for its Grok 3 and Grok 3 Mini models. The former can analyze images and answer questions, and powers features on Musk’s platform X. Pricing ranges from $0.30 to $25 per million tokens, depending on model and speed.

Grok 3 is positioned as a competitor to GPT-4o and Gemini, but it's more expensive than Google’s Gemini 2.5 Pro and has a smaller context window than previously claimed. 

The model was originally pitched as edgy and “anti-woke”, but past versions still leaned left on some political issues. Something Musk blames on biased training data. It's unclear if Grok 3 has shifted to the politically neutral tone Musk promised.

Meta launches its own free AI app at LlamaCon

Meta has officially launched a standalone Meta AI app for iOS and the web (sorry, Android users). The app runs on Meta’s new Llama 4 models and offers voice-first interaction, image generation, and chat-based assistance, while trying to learn your preferences and keep up with ongoing conversations.

You can log in using your Facebook or Instagram account, and the app also works with Meta’s Ray-Ban smart glasses. It’s being positioned as a more personal, integrated AI experience. The app was revealed at Llamacon 2025, Meta’s first AI developer conference.

Users can chat with the AI, talk to it naturally using a new voice mode, and generate images. Though image quality still lags behind rivals like Midjourney and GPT-4o. While the web version of the app offers more powerful image tools, the mobile experience feels a bit more limited.

Did a friend forward this e-mail to you?

IE₊ SUPPORT INTERESTING ENGINEERING
_{Invest In Science And Engineering}

Enjoy exclusive access to the forefront of AI content, highlighting trends and news that shape the future. Join a community passionate about AI, delve into the latest AI breakthroughs, and be informed with our AI-focused weekly premium newsletters. With IE+, AI reporting goes beyond the ordinary - and it is Ad-Free.

AI PICTURE OF THE MONTH

After the Pope's recent passing, US President Donald Trump said jokingly that he would make for a great Pope. His jokes didn’t end there, unfortunately. The President’s account on Truth Social and the White House X account posted an AI-generated photo of him in the Pope’s attire.

Let’s just say it rubbed off Catholics, and people in general, the wrong way.

While Trump brushed off criticism, joking that “Catholics loved it,” many Catholics found the image inappropriate, especially following Pope Francis’s death, reported the New York Times. Trump said he didn’t know where the image came from. 

Trump said Melania thought the image was “cute”. Trump is not Catholic, but his VP pick, JD Vance, a Catholic convert, defended the image as just a joke.

There’s never a dull day in Trump’s White House. Only plot twists, costume changes, and the occasional AI-generated papal cosplay.

AI TUTORIAL OF THE MONTH

Learning English can feel like a struggle, especially when private coaches are expensive and hard to book, or it’s not someone’s mother language. 

But I recently came across a free tool that gives you instant feedback using AI to help you speak more clearly and confidently. It’s called Pronounce AI. You can use it in your daily routine in five steps. 

Step 1: Head to the Pronounce AI Website

Just open your browser and search for Pronounce AI. No need to install anything onto your computer.

Step 2: Type or Paste a Sentence

You’ll see a text box where you can either type your own sentence or copy-paste something you want to practice (like a line from a movie, a presentation, or even a tweet). Keep it simple at first, then try longer, more complex lines.

Step 3: Hit Record and Speak

Click the microphone button, and read the sentence out loud. Make sure you're in a quiet spot, and speak naturally. No need to fake an accent. Just talk like you would in real life.

Step 4: Get Instant Feedback

After you finish speaking, Pronounce AI gives you a score and points out which words you pronounce well and which ones need work. It even shows you a side-by-side comparison of how a native speaker would say the same thing.

Step 5: Practice and Repeat

You can keep trying the same sentence until you nail it or move on to new ones. It’s super chill and just helpful tips and progress over time.

Here is a video to get you started.