- AI Valley
- Posts
- OpenAI "leaky" weekend
Together with
howdy, it’s Barsee again.
happy monday, AI family, and welcome back to AI Valley.
here are the biggest things worth knowing today:
OpenAI might’ve just previewed its next image model… by accident
OpenAI’s cap table is more crowded than people think
Perplexity unveils Computer for taxes
Plus trending AI tools, posts, and resources
Let’s dive into the Valley of AI…
Med0S
Courtesy: MedOS
MedOS is a next-generation AI-XR-Cobot medical system developed by the Stanford–Princeton AI Coscientist Team. Designed as a real-time clinical co-pilot, MedOS brings multi-agent AI reasoning, XR smart glasses, and collaborative robotics directly into live hospital workflows.
Now officially featured at NVIDIA GTC 2026, and deployed at the Stanford Blood Center and Stanford Department of Pathology, MedOS marks the first transition from lab research to real clinical environments.
Recent system upgrades doubled the supported medical disciplines and reduced XR response latency to near real-time. With new intelligent glove capabilities enabling precise physical assistance, MedOS bridges AI perception, reasoning, and execution, amplifying clinicians rather than replacing them.
*This is sponsored
THROUGH THE VALLEY
OpenAI might’ve just previewed its next image model… by accident
For a few hours, a new OpenAI image model showed up on LMArena under three fake names.
Then it disappeared.
If you’ve followed OpenAI launches for a while, that’s usually not random. Stuff has a habit of surfacing in public benchmarks right before it becomes real.
Source: LMArena
The aliases were:
maskingtape-alpha
gaffertape-alpha
packingtape-alpha
From the short time it was live, the model looked noticeably better than OpenAI’s current image system in a few areas that actually matter.
Some of the early examples included:
IKEA storefronts that looked way too close to real photos
Windows / YouTube UI generations that could actually pass as screenshots
handwritten notes that didn’t instantly turn into mush
small details like cat-eye reflections, which are exactly the kind of thing image models usually fake badly
That’s a bigger shift than it looks.
Because if we’re being honest, image models haven’t really struggled with making things look good for a while now.
That part is mostly solved.
The harder part has been credibility.
A lot of models are still better at giving you the impression of something than actually getting it right.
They can make an image that looks convincing at first glance, but the moment you ask for anything specific, the illusion usually starts to crack. Text gets weird, layouts feel off, and small details stop agreeing with each other.
That’s why people test things like screenshots, signage, handwritten notes, receipts, and UI.
Not because they’re exciting. Mostly because they’re boring enough to expose whether the model can actually hold up.
This one seemed better there.
Why does it matter?
The first wave of image models was mostly about making cool-looking images. This next phase is about making ordinary things look real.
That’s the more important milestone.
Because the internet doesn’t really run on fantasy art. It runs on screenshots, product photos, documents, and random visual proof people usually trust without thinking too hard about.
And once models get good at that, the question stops being whether they can make beautiful images.
It becomes whether people still notice when something was never real to begin with.
Other news:
OpenAI Cap Table
Most people talk about OpenAI like it’s basically “owned by Microsoft,” but the actual cap table is much more spread out. Microsoft reportedly holds 26.79%, the OpenAI Foundation sits at 25.80%, and employees still own a surprisingly large 19.35%. SoftBank comes in at 11.66%, followed by VC and institutional investors at 7.83%, Amazon at 4.66%, NVIDIA at 3.47%, and individual investors at just 0.40%. In other words, OpenAI isn’t as cleanly controlled as the internet tends to frame it, which probably explains why the company so often feels like it’s being pulled in several directions at once.
Perplexity has launched a new use case for its Computer feature: helping users prepare their federal tax returns. The pitch is simple enough. Upload your tax documents, answer a few questions, and the system fills out the relevant IRS forms for you. Perplexity also says it can retrieve and apply the latest tax code while doing it, which is either genuinely useful or the kind of sentence that should make at least a few accountants sit upright. Either way, “AI that does your taxes” is starting to move from demo idea to real product category.
TRENDING TOOLS
Teleport — Connect agents to your infra in seconds with Teleport Beams. Built-in identity. Zero secrets (sponsored)
Pika > It can now build digital clones of real people that lets you chat with your agent on Google Meet, complete with a face, voice, and personality
Twin > An AI-powered company builder that helps you move from idea to execution faster
Gemma 4 > Google’s new open-weight AI model delivers state-of-the-art intelligence for its size
Caveman Claude > A one-line plugin for Claude Code that rewrites responses in caveman-speak, cutting token usage by around 65% without losing technical detail
Steer AI > Ramp Labs’ latest model is designed to obsess over any concept you give it
Atlas 1 by Willow > Real-time speech-to-text with human-assisted accuracy that beats OpenAI and Deepgram
Goose > Block’s open-source AI agent does more than suggest code. It can install, execute, edit, and test with whichever LLM you choose
WHAT I'M CONSUMING
THE VALLEY GEMS
THAT’S ALL FOR TODAY
Thank you for reading today’s edition. That’s all for today’s issue.
💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee
👍️ New reader? Subscribe here
Thanks for being here.
HOW WAS TODAY'S NEWSLETTER |
REACH 100K+ READERS
Acquire new customers and drive revenue by partnering with us
Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.
If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.