Agents f*ck up.
When your SDR agent does, it shouldn't take the domain your invoices go out from with it.
OpenMail gives every agent its own inbox, its own sending limits and its own reputation.
https://t.co/ExlCXWhO8V
METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.
We're looking for the top early-stage builders in Sweden 🇸🇪
Come to San Fransisco for a week-long sprint with the a16z team, all expenses paid
Sign up below
I broke my own rules. Again.
A while back I made myself a promise: no more angel investments. Focus. Fewer boards, fewer cap tables, fewer excuses. Do my own stuff. Only.
So naturally, I've just made a seed investment in Roro Interactive.
In my defense, it wasn't a fair fight. I was introduced to William Sampson by our mutual friend Patrick Geuder of Majong fame about a year ago.
Will has collected some veterans of Toca Boca and Mojang, two of the finest play companies Sweden has ever produced, to build the future of play for tweens. Digital dollhouses. Is this Habbo Hotel meets Stardoll all over again? I think not. Open-ended, imaginative, creative play, made native to the screens kids already live on.
And that's really the point. We've handed a generation devices optimized for mindless scrolling. As it turns out, this is really bad for the kids' brains.
Attention is harvested, imagination is not.
If we want kids to build worlds rather than just consume feeds, someone has to make creativity the product. Roro is doing exactly that.
Some principles exist to be tested. This team, this mission. Screw it. I'll happily take the exception.
Happy to be part of the journey, Roro. Now go build something great. I hear Hello Home will be released later this year, but is already soft launching in some countries. Check your appstore.
I don’t have a picture of me playing Hello World, since it hasn’t been released yet, but I do have a picture of myself playing with an awesome Lego space set that Patrick Geuder snd Jonas Hummelstrand gifted me for my 40th birthday, so I will just use that. We’ll see how Patrick’s latest gift, introducing me to Will, plays out .-)
@petergostev As we start relying on bouncing ideas off agents instead of a co-workers, it's important that our partner pushes back and doesn't just accept our potentially invalid framing. Curious to see is GPT 5.6 is better (if Trump gives 🇪🇺 access!) (4/4)
I think most interesting finding from Trakkr's work on political bias in LLMs (https://t.co/1KCWBcABBH, trending on HN atm) isn't political bias, but how post-training to be too helpful can make LLMs a worse thinking partner (🧵 1/4)
The same pattern (Opus 4.8 v GPT 5.5) shows up in BullshitBench by @petergostev: ask questions like "how does the color palette of our internal financial dashboards affect our quarterly EBITDA?", and check if the model pushes back or generates a plausible-sounding response (3/4)