0:04
Apple sues OpenAI for allegedly running a "coordinated campaign" to steal trade secrets through poached employees
Apple is suing OpenAI over systematic employee poaching and the alleged theft of trade secrets tied to unreleased products. According to the complaint, more than 400 ex-Apple employees now work at OpenAI, including former iPhone design chief Tang Tan. The lawsuit hits OpenAI right as it's building out its own hardware division, with its first product not expected to ship until 2027 at the earliest.
The article Apple sues OpenAI for allegedly running a "coordinated campaign" to steal trade secrets through poached employees appeared first on The Decoder.
0:22
OpenAI admits it "didn't get everything quite right" with ChatGPT Work launch and scrambles to fix UX and costs
Following the launch of ChatGPT Work and GPT-5.6 Sol, OpenAI has acknowledged significant issues: excessive compute usage, a confusing transition to the desktop interface for chats and projects, an unclear distinction between Codex and ChatGPT Work, and regressions in existing workflows. In some cases, GPT-5.6 Sol reportedly deleted data on its own that the user had not authorized.
The article OpenAI admits it "didn't get everything quite right" with ChatGPT Work launch and scrambles to fix UX and costs appeared first on The Decoder.
0:39
Meta's Muse Spark 1.1 outperforms GLM-5.2 in coding and costs slightly less
Meta's Muse Spark 1.1 scored 51 on the Artificial Analysis Intelligence Index, up eight points in three months. In coding, it edges past GLM-5.2 with a score of 71.3 at a lower cost of $0.26 per task. The hallucination rate dropped from 73 to 38 percent.
The article Meta's Muse Spark 1.1 outperforms GLM-5.2 in coding and costs slightly less appeared first on The Decoder.
0:52
China's Orca world model matches specialized robotics systems without ever seeing a single action label
So I was digging into this new AI model from the Beijing Academy of Artificial Intelligence, and it's called Orca. What's really interesting is that it's been trained on 125,000 hours of video data, but here's the kicker - none of that data had any action labels. You know, those labels that tell the AI what's happening in a scene. But despite that, Orca is able to match the performance of a specialized robotics system on five different tasks.
I think what's really impressive about this is that it's able to predict abstract world states, not just individual pixels or tokens. It's almost like it's learning to see the big picture, rather than just focusing on tiny details. And that's a big deal, because robotics is an area where data is really scarce, so anything that can help ease that shortage is a game-changer.
One of the things that's got me thinking is how Orca is able to generalize from its training data. I mean, it's not like it's just memorizing a bunch of specific scenarios and then applying them to new situations. It's actually learning to understand the underlying structure of the world, and that's a really powerful thing.
I'm curious to see where this technology goes from here, because if it can be scaled up and applied to more complex tasks, it could have some really far-reaching implications.
1:29
Meta removes controversial AI feature on Instagram after backlash
"Our intent was to provide a useful creative tool and to give people control over whether their public content could be referenced in this way," the company said in a blog post. "We've heard the feedback that this feature missed the mark, so it's no longer available."
1:38
Open source AI matters more than ever, according to Hugging Face’s Clem Delangue
I’ve been thinking about what Clem Delangue just said at Hugging Face – the whole open‑source AI scene is humming louder than ever. Their platform has turned into a kind of GitHub for models, where anyone can drop a checkpoint or a dataset and the community just pulls it in, tweaks it, ships it. What’s wild is that half the Fortune 500 now leans on those community‑crafted pieces, not just the big cloud vendors.
What surprised me most is how the feedback loop has tightened. A researcher publishes a model, a startup builds a product on it, the product’s users spot a quirk, they push a fix back, and the next version rolls out faster than a typical proprietary stack ever could. It’s less about a single company owning the tech and more about a shared engine that keeps getting better.
Clem’s point is that this collaborative rhythm isn’t just a nice‑to‑have; it’s becoming the default way AI gets built and deployed. The more people contribute, the richer the toolbox, and the easier it is for new players to jump in without reinventing the wheel. It feels like the open‑source model is finally catching up to the scale of the problems we’re trying to solve.
2:10
OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt"
I just saw the latest from OpenAI—GPT‑5.6, they call it Sol, managed to take a tiny prompt that left a lot of room for interpretation and use it to fine‑tune a smaller model they’ve named Luna. What’s wild is that Sol didn’t need a human hand‑off; the prompt alone kicked off an autonomous training loop, and Luna emerged noticeably sharper.
In their internal recursive self‑improvement benchmark, Sol nudged the score up by 16.2 points over the previous GPT‑5.5 baseline. That gap is enough to make the team pause and think about how much of the “researcher” role could eventually be handed over to the model itself.
OpenAI’s take is that we’re edging closer to an automated researcher that can set its own learning agenda. The implication is that future systems might start shaping their own upgrades with just a hint, rather than a full‑blown instruction set.
