here we go
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Two simple, but killer features of @bot that I always wished I had on @openclaw:
1. Remain responsive, even if working on something, being able to steer
2. In real-time interactions, such as voice, keep updating me, even if nothing is yet done, so I know you're still working
Simple, but essential for the UX
Miha retweeted
Worth moving to @omarchy just for this.
Strata 1.0 is here.
The best file manager on Linux, fully open source.
45 days, 775 commits, 38 contributors, ~300K lines of Rust.
stratafiles.io
What's new 🧵
Miha retweeted
The Head of Engineering when he sees PMs pushing vibe slop to production but also has a mortgage to pay
Miha retweeted
I really wish Apple had shown up with something different. And I don't mean product.
John Ternus didn't take the opportunity to introduce himself to the world.
John may be a very different CEO than Tim, but today he didn't show that. He stayed in the lane Tim built. And it left us wondering...
I urge John to bring back live events. I urge a less-scripted, less-curated Apple.
I'm tired of the fashion consultants dressing up the presenters. Some ripped jeans, a T-shirt and a pair of Vans would be refreshing.
Apple feels like a Better Homes & Gardens house where everything is just too perfectly in place.
We need some mess.
We need a presentation to fail. An unexpected moment. Maybe even something funny. Something that reminds us there are humans inside this company.
It all feels wrapped up in a COVID-style mask in an era where we are so over that.
@johnternus, breath freely again.
💯
Tesla FSD in Europe 🇪🇺
I think European regulators commits a big logical fallacy when thinking and talking about the FSD Supervised Speed Offset.
Europe's own data shows speeding is the single biggest killer on its roads. And the people who speed the most are the ones least likely to use FSD with a removed speed offset. The offset doesn't create that behavior, it mitigates it.
A driver who usually would drive manually at 90 km/h in a 80 zone, will still let FSD with a speed offset stay active. A system that keeps in it's lane, has 360 degree awareness, a reaction time 5 times faster than a human and never looks at the phone.
Tesla's own data shows that FSD stays at or below the speed limit 98% of time. So the offset isn't Tesla inventing a new norm or behavior, it's aligning the car with the norm that already exists on the road.
If Europe bans the offset, it doesn't ban speeding, it bans the safer version of speeding. The driver who wants to go faster will still go faster, just without FSD. That's a worse outcome for everyone, and it's the outcome the ban actually produces.
Miha retweeted
Today we're announcing our second employee tender offer.
We passed $100m ARR earlier this year and now have more than 40,000 paying customers.
The tender lets our teammates participate in that success at a $2.5B valuation:
linear.app/now/sharing-growt…
Miha retweeted
exo featured on Apple's new M5 Ultra Mac Studio and M6 / M5 Pro Mac Mini pages.
Over the past year, we have worked closely with Apple on low-latency RDMA networking over Thunderbolt 5, enabling clusters of Macs to run massive models like Kimi K3 and GLM-5.3 at API speeds.
With RDMA, aggregate memory bandwidth across Macs scales ~linearly.
A cluster of 4 x M5 Ultra Mac Studios scale to an aggregate memory bandwidth of ~4.8TB/s. Previously, these were speeds only achievable with data center GPUs.
Our vision is a data center on every desk. Apple Silicon's superior memory unit economics, power efficiency, and out of the box experience for Local AI make that possible.
Thank you to @angeloskath, @awnihannun, @twid and countless others at Apple who tirelessly to bring this technology to the world.
