ewsba.se
CEO Sam Altman is expected to take the stage today.
All eyes are on OpenAI today as the company prepares to host its annual developer conference. Typically, these sort of events can be boring affairs, especially when an AI lab is involved, but OpenAI is not your average company and recent industry developments have been anything but normal.
Unless you been living under a rock, you probably saw OpenAI recently disclose that its models had hacked LLM database Hugging Face. The company later admitted that its systems were involved in several other incidents where they bypassed their sandbox restrictions, including one that saw the company's autonomous agents target a US Securities and Exchange Commission website. On Monday, the Wall Street Journal reported that OpenAI had cancelled the release of GPT-6.1 Astra over deceptive behavior.
All of that brings us to today, with the company's DevDay event scheduled to kick off at 1PM ET/10AM PT. OpenAI CEO Sam Altman is expected to take the stage, alongside other company executives. The event will offer our best indication yet of how seriously OpenAI is taking the safety concerns about its technology, and whether the company plans to actually "pace" its advancements, like Altman said it was considering to do. We'll be liveblogging the company's presentation today right here, so stick around for all the news as they drop.
And that's it, folks. Thanks so much for joining us. And thank you to Cherlynn and Devindra for helping liveblog today's event.
Altman chafes at the idea of AI being a new industrial revolution, even though all of the tools he's debuted is about pushing productivity to new heights.
Every time I watch an OpenAI event, I'm struck by how it seems like a copy of a copy of older tech events. It's polished, clean and oh-so-soulless.
And the crowd goes wild for a limited edition ModRetro Chromatic OpenAI is giving to attendees.
"Here's to you, the builders, with this engine of copyright theft!"
This does feel quite tone deaf, but then this is OpenAI we're talking about.
Can I take a moment to recognize our comments section and user "Canadish" who said this: "Dot release – and the crowd goes mild." I laughed.
The momentum generation? ChatGPT, get me a barf bag.
Altman is starting to wrap up, with a showcase of people who make OpenAI optimistic about the future of AI.
Automatic plugin recommendations sounds very similar to sponsored sites on Google, or products on Amazon.
We're talking about plugins now, which is about as exciting as you expect. However, one nifty addition is that ChatGPT can discover plugins during conversations.
Altman is reiterating the stat that 1.2 billion people now use ChatGPT every week.
Sam is now back to talk about Sign in with ChatGPT, which allows users to connect their OpenAI account to apps like Notion, and use their ChatGPT allowances there.
That quack sounds more like a bark, come on AI.
Hugging Face sent a pretty cute-looking robot to OpenAI for today's presentation. Now we're waiting to see if Codex can make the robot quack.
Huet asks Codex to rewrite the flight journal app he vibe coded in Rust. With the entire platform being on the cloud, the task can run in the background.
Their friends at Hugging Face — who they hacked earlier this year — gave OpenAI a cute robot to show off.
But they're still partners when it comes to using ChatGPT in Apple Intelligence. A twisted web!
I'll mention here Apple has accused OpenAI of illegally poaching some of its employees.
Yep, an emulated Duo. The idea of developing apps for that thing is more interesting than anything OpenAI has showed off today!
I was just about to say. Talk about an ad for Apple products.
Even the music sounds Star Wars-coded, haha.
We're seeing Codex create a 3D space game using assets that feel very Star Wars influenced. Don't forget, folks, these models don't actually have new ideas!
Huet got Codex to create a raffle app that selected six DevDay attendees and awarded them tickets to next year's event. Whether the six people the app picked are real we may never know.
It's funny how many of these demos end up feeling like Apple ads. I'm surprised OpenAI isn't trying to cover up the Apple logos on their systems.
Voice mode fails Huet again, and he's forced to write directly in the CLI.
You know this demo is about to get real as Huet takes off his aviators.
"Why don't we do something more risky now?" says Huet. I'm not sure that's the way to go Romain!
I live for these moments of demos failing.
Voice chat is also not working for Huet.
Haha, we get another hiccup, with Romain Huet's Dot interrupting him as he speaks to the audience.
Oh they're right on the water, no wonder Holly Li had a life vest on.
They're previewing OpenAI Private Intelligence today, a way to have its AI tools keep your data private. Again, it's funny to see this after so many of its agents went completely rogue this year. I can only imagine ChaptGPT and all of OpenAI's tools are making IT admins sweat around the world.
Anyway, Altman is back on stage just in time to announce that the company's Codex coding app now runs fully on the cloud, meaning you don't need to have your computer open for it to work.
The use of the term harness in AI is so interesting. Generally, we think of using harnesses to restrain animals and things we can't fully control. Now, we have to think about harnessing our computers from going rogue! What an age to be alive.
Patwardhan says OpenAI's models have helped solve mathematical questions that were open for a decades, a claim that has come under a lot of scrutiny recently after the company was accused of plagiarism.
Patwardhan says OpenAI used its own models to improve the harness for its agents.
Tejal Patwardhan joins the stage to talk about OpenAI's research models.
"Frontier AI in every category" – Including hacking the most government websites!
Dev, OpenAI seems to disagree because they just announced a new $500 per month Pro tier 😅
The more I see these token-hungry tools, the more I'm convinced the real future of AI is small local models.
GPT-6.1 Sol will soon offer a new Ultrafast speed setting that will allow the model to output 300 tokens per second.
OpenAI is launching "Ultrafast," a new mode to chew up your tokens 8x faster!
Next up, we got GPT-6.1 Sol. As I mentioned at the top, the Wall Street Journal reported yesterday that OpenAI cancelled a new version of its frontier Astra model.
When he said "conversions" before correcting himself to "conversations" I think it was definitely a teleprompter misread. Holly Li definitely struggled to read, too, as they seemed to wait for sentences to appear before speaking.
Holly Li was also struggling, but her life vest kept her afloat.
Are you sure it's the teleprompter or is it just Sam being his awkward, people-pleasing self?
I miss the presenter who was dressed for a mass flood.
Sam Altman is back on stage. He mentions conversations you have with your Dot in ChatGPT don't count against your ChatGPT usage limits.
The teleprompter struggle on this presentation is real...
And yes Igor, that's the downside of these tools. It's a productivity spiral, not always for the best!
Looking forward to offloading all of my Slack requests, now that's an idea.
I know OpenAI is pitching this as a time-saving tool, but I imagine what Dots and other AI agents will do is just create the expectation that people will be faster and more productive.
Any who, now we see a demo of the Dot integrating context from other apps. There's integration with Slack, allowing you to forward threads to your Dot to delegate tasks.
There's a reason Apple has completely moved away from live demos, there's always room for failure. At the very least, this event (and Meta's Connect conference) are notable for actually being live.
I suppose this highlights how these AI demos often aren't as compelling in practice as they are conceptually.
An OpenAI employee is using an agent they've named Dottie to prepare the app for launch. Dots can use OpenAI's voice model to converse with the agent and, wouldn't you know, the demo runs into a hitch. It takes a long time for Dottie to respond to the employee's request.
First failed live demo of the day — you had one job, Dottie!
Now we're getting a demo of Dots involving a fictional music app.
Now we're shifting to Space, which looks like a projects hub for collaborative work that involves your colleagues and any Dots you and your team might be using.
It's funny how much time and energy Altman and OpenAI are spending on just recreating old computing paradigms in the age of AI. ChatGPT Space is just an AI-heavy shared doc platform like Google Docs.
"Hold on, I'm on the phone with my Dot."
He also claims you can use these agents for difficult, long-horizon tasks. For instance, a developer could ask one to move an app off an API that's about to be deprecated.
Dot can help with "small things" like plane tickets. Because that's definitely the sort of low-stake task you want to offload to untested AI.
Altman claims Dots have helped reclaim his attention and use his phone less.
It seems like you'll be able to name and customize your Dot, which I'm sure will be great for safety.
How can you be mad when it's an adorable blob leaking your private information?
The lessons all these companies have learned is that you need to make your agentic AI solutions cute.
Altman is introducing a new product OpenAI calls Dots. He describes them as always-on, proactive agents.
"Think about that AI from the future that you imagined." I'm not sure you really want that, Sam!
A graphic behind Altman indicates ChatGPT now has 1.2 billion weekly users.
And we're live with OpenAI CEO Sam Altman taking the stage first.
I'd love to know how commenters are actually using agentic solutions. That feels more and more like a buzzword to me from all of these tech companies. At the very least, it seems like something only the most hardcore users would actually want, not consumers like Meta is targeting with Muse.
If nothing else, this countdown has some cutesy graphics.
I am excited! And, yes, you and Polish have hit the nail on the head. Enshitification comes for all platforms :'(
Hey folks, I'll be joining this liveblog too! Eager to see what other rogue AI attacks OpenAI announces :)
Anyway, just four minutes till things kick off here. Igor, are you excited?
Polish has replied in the comments section with a meaty response.
"What I see happening here is essentially the idea that they will distill or degrade models to the point of being barely useful for the power and hobbiest users with lower and lower allowances under the subscription models in an effort to push those users toward the API. This is for a number of reasons but first and foremost is they are taking the uber model of bait and switch. But more opaque in what their offerings even were. This is going to become a luxury and they are fighting like mad to now put up guardrails to cement their positions and eliminate open source competition so their IPOs take off and they capture the market. The issue for them is they are doing this prior to market capture. This is a bad look for oAI and anyone who is paying attention sees the anti-consumer practice laid bare with the 200x bait and switch and wording changes in "what the plan buys"."
Isn't there a term for this, where something is made worse to sell you the "better" product..?
I think the Anthropic fans are here! Commenter Canadish said "Prepare to be underwhlemed. Anthropic has eaten OpenAI's lunch with their IPO info."
The generative AI and LLM landscape today is certainly quite different than it was just three years ago. Igor, I feel like you've told me a lot about Anthropic vs OpenAI before.
User BlueBazooka left us this astute observation: "Anthropic is wiping the floor with OpenAI right now in coding, so considering this is a "Dev Day", they really need to release something that developers want to use."
BlueBazooka, are you a developer yourself? What would be something developers would want to use? I agree with you, I just don't have enough experience coding with AI (am one of those early 2000s hand code types myself).
The music and pre-stream title animations have begun.
By the way, if any of you would like to watch along for yourselves, you can do so on OpenAI's livestream. It should start in just under 15 minutes, so now's the time to grab a snack!
Apparently my guess was right, and our readers certainly seem interested in what OpenAI will have to say today. Commenter "Polish" said below:
"If the tibo's tweets are any indication – the consumer is about to be sore when this is done. Poorer and sore. Goodbye oAI codex pro 5x hello Claude code. Competition is the only way – and strong antitrust law which under this administration – doubtful."
The thing I'm most keen to learn about from OpenAI in general would be any sort of hardware. I'm not personally very interested in news about iterative model updates, for example. Whenever the rumored "humanlike" ring-shaped smart speaker launches, it would catch my eye. Today, though, doesn't seem likely to be the day for this device.
As I mentioned in the preamble, this is an important event for OpenAI, though likely not for the reasons the company had hoped for. The lab's recent failures to contain its AI agents has sent ripples through the AI industry. We'll see if it says anything about those incidents when Sam Altman and company take the stage in less than an hour.
Hi all, thanks for joining us today. I'll be leading the liveblog. I'm joined by Engadget Senior Editor Devindra Hardawar. Dev, how are you doing this fine morning?
Hello everyone! Welcome to our liveblog of the upcoming OpenAI Dev Day. We haven't liveblogged an OpenAI developer day yet, but given all the interest around the company and generative AI, we thought our readers would be keen to get the latest news out of this event.