What went wrong with the ChatGPT desktop redesign?

When OpenAI merged Chat, Work, and Codex into a single desktop app last week, the result was a cluttered interface that buried basic features. Users complained about losing quick access to chat history and struggling to switch between modes. The backlash was quick and loud.

Now the company has responded with a batch of fixes, acknowledging the rollout wasn’t quite right. In a post on X, OpenAI’s Tibo Thsottiaux said: “We’ve gotten lots of great feedback on the new ChatGPT desktop app (which we didn’t get totally quite right on the first try).”

What’s changing in the updated ChatGPT desktop app?

The biggest change is the return of conversation history and Projects to the sidebar. Instead of digging through different screens, you can now jump back into previous chats or ongoing work from one place. It’s a small tweak that makes a big difference.

OpenAI has also enabled Chat and Work history syncing across web, desktop, and mobile. Your conversations stay up to date no matter which device you pick up. There’s a catch though: Local Tasks remain stored only on your computer and won’t sync across devices.

The app also gets a dedicated switch for moving between Chat and Work mode. The layout now matches the web and mobile experience, so you don’t have to relearn the interface every time you switch devices.

Search upgrade and safety features

Alongside the desktop fixes, OpenAI rolled out a major search upgrade yesterday. You can now find old chats, projects, documents, and images from a single search bar. The company also added a batch of teen safety features and parental controls.

These changes are live across every ChatGPT plan, including free accounts. All you need to do is update your desktop app.

Why does this update matter for ChatGPT users?

This update matters because it shows OpenAI is listening. The company says these changes are a direct response to user feedback after the redesign missed the mark. It also confirmed that Codex mode remains unchanged, while promising to keep shipping improvements based on what users say.

For now, the ChatGPT desktop app looks a lot closer to what users expected from the start. The question is whether OpenAI can keep that momentum going with future updates.

If you’re using the desktop app, make sure you grab the latest version. And if you’re curious about other recent changes, check out our coverage of ChatGPT’s new features and the broader AI assistant landscape.

A goal the world never saw — until now

In 1959, Pelé scored what he called the greatest goal of his career. Three defenders beaten with consecutive sombreros. A knee flick past the goalkeeper. A header to finish. The ball never touched the ground. And nobody filmed it.

Sixty-five years later, Google has recreated the moment using its Gemini AI, and the result is genuinely moving — not just for football fans, but for anyone who cares about how AI can preserve history rather than exploit it.

The ‘Gol da Rua Javari’ — a goal preserved only in memory

Known in Brazil as the “Gol da Rua Javari,” the goal was witnessed only by the thousands of fans in the stadium that day. A single grainy photograph of the header moment was the only visual record that survived.

Google’s team tracked down people who were actually there — some now in their 80s and 90s — and interviewed them about how the play unfolded. They studied photos of the stadium, the crowd, and the players. Bit by bit, they reconstructed the choreography of a moment that existed only in memory.

How Google rebuilt a football legend

This wasn’t a simple deepfake. The team recreated the entire scene with real human actors in the same stadium where Pelé scored. They even sourced a replica of the heavy leather ball from that era, along with Pelé’s dark boots and the period-accurate uniforms.

The actors’ movements were mapped using motion-capture techniques — but without the usual specialized suits and gear. All that data was then fed into a pipeline combining Gemini Omni, the Nano Banana Pro image generator, and the Google Veo video engine.

The system restyled the footage to match the look of old photographs. The stunt performer wearing the iconic No. 10 jersey was swapped for Pelé’s digital likeness. The entire stadium was recreated as it appeared on that fateful day in 1959.

Period accuracy down to the film grain

Google didn’t stop at the visuals. “To ensure the generations looked as period accurate as possible, we ran the digital output through a filmout machine, capturing the distinct look and feel of 1950s cinema,” the company wrote in its blog.

The result is a short video that feels like lost archival footage — grainy, slightly flickering, utterly authentic. Seeing Pelé’s legendary skill come to life for the first time is breathtaking, even for viewers who never watched him play.

Why this AI project is different

Let’s be honest: AI’s track record in entertainment has been rough. Resurrecting dead actors, finishing songs artists never completed, generating entire scenes — the backlash has been loud and often justified.

This project stands apart for one crucial reason: consent. Pelé’s family approved the recreation. His daughter, who was quoted in Google’s announcement, said: “He would be so proud to see all this happening. He’d always say it was a shame that the goal was never recorded. So being able to relive it, with all this technology, is amazing.”

Football historians and fans were consulted throughout. This wasn’t a cash grab. It was an act of preservation — a way to give future generations something that history denied them.

What this means for AI and sports history

The project opens a fascinating door. How many other legendary sporting moments exist only in the memories of those who witnessed them? How many goals, tries, and match-winning plays are lost to time because the cameras weren’t rolling?

AI won’t bring those moments back perfectly. It can’t. But it can offer something close — a respectful approximation built on eyewitness accounts, archival research, and modern technology.

That’s a far more hopeful use of AI than yet another deepfake of a celebrity. It’s the kind of application that makes you wonder what else might be recovered from the past. A lost concert. A forgotten speech. A goal that defined a career.

For now, we have Pelé. And thanks to Google AI video generation and the team behind it, the Black Pearl’s greatest moment finally has a place in the visual record of football history.

The leak that changed everything

It started with a single news article. Not a product launch, not a customer win — just a report from The Information that Databricks was raising money. That was enough to set off a chain reaction CEO Ali Ghodsi never anticipated.

“We wanted to raise $1 billion, but then The Information printed this article saying that Databricks is doing a big fundraise,” Ghodsi told TechCrunch. “We were heads down with our conference, and we were not actually at all focused on fundraising.”

The timing couldn’t have been worse. Databricks was in the middle of its annual user conference in June, and Ghodsi’s phone wouldn’t stop buzzing. “My phone blew up. It was like the worst timing for us because we were busy with our conference,” he recalled.

But here’s the thing about having too many investors pounding on your door — it’s a pretty good problem to have. What started as a modest $1 billion target quickly snowballed into something far bigger. “The interest level was just insane. Just from this select group of investors that we looked at, there was $15 billion of interest,” Ghodsi said.

From $1 billion to $5 billion: how the round grew

When you’ve got that much demand, telling longtime backers they can’t get in is a quick way to burn bridges. So Databricks made a pragmatic call: issue more stock. In July, the company announced it had closed a new round at a $188 billion valuation, though it kept the exact amount under wraps at the time.

On Thursday, the full picture emerged. Databricks raised $5 billion at a valuation that ticked up to a clean $190 billion. The round was led by Coatue, with participation from Blackstone, MGX, T. Rowe Price accounts, and newcomer Sixth Street Growth — the firm founded by former Goldman Sachs chief investment officer Alan Waxman. All told, about two dozen VCs got a piece of the deal.

Why investors were so eager

The answer is simple: Databricks looks like a sure bet. Ghodsi says the company has hit $7 billion in annualized run-rate revenue, growing at 80% and already cash-flow positive. Its core cloud data warehouse product alone generates $1.5 billion of that run-rate and is still expanding at 100% year-over-year.

Then there’s the AI angle. Lakebase, the company’s database for AI agents launched in June 2025, has already reached a $100 million revenue run-rate. And Genie, its AI chatbot for business analysis, is — in Ghodsi’s words — “insanely popular.”

Why raise more when the business is thriving?

That’s the obvious question. Databricks had already pulled in $20 billion over the past 20 months. But AI doesn’t come cheap.

“AI research is very expensive,” Ghodsi said, noting the company employs a 100-person AI research team. Databricks also has multi-billion dollar cloud commitments with all three major hyperscalers — AWS, Azure, and Google Cloud.

On top of that, Databricks is on an acquisition spree. This week it snapped up Electric, the company behind the lightweight Postgres database PGlite. In June it bought AI cybersecurity firm Panther, and in March it acquired two more startups. “We do a lot of M&A,” Ghodsi said simply.

The new math of mega-rounds

There was a time when a $1 billion raise was considered massive — a headline-grabbing feat that took months of courting investors. In today’s AI gold rush, that number barely moves the needle. Startups are walking out of the gate with billion-dollar seed rounds.

Databricks’ private fundraising has become something of a running joke in Silicon Valley. When the company announced this round last month, observers quipped online that it had raised so many times it was running out of letters of the alphabet to name the rounds.

What’s next for Databricks?

Ghodsi told CNBC he still wants to take the company public eventually. With this many investors on the cap table, all of whom will want an exit at some point, he doesn’t have much choice.

But for now, the focus is on investing in AI — and doing it out of the public eye. Given the expenses involved, staying private a while longer makes sense. And when you can command $15 billion in instant interest on your own terms, why rush?

For more on how AI companies are reshaping fundraising, check out our analysis of AI startup funding trends and the latest on Databricks competitors in the data lakehouse space.