zeeman retweeted
introducing jester, ur free imessage meme dealer
roast ur mom in the family group chat, make memes and ai vids, and meme ur ex so u can finally get over her (u still wont 😂🫵)
text jester free memelord.com/imessage
Another limit is API latency.
Jev itself is fast, but every request still has to go over the network. This meant running the sim at 10x slower than real-time to get close to the control fidelity of an in-sim cFS-style loop.
1
15
Compared to closed-form laws used for normal prox ops, Jev's recommendations are limited to discrete set of levels. That leads to over- and under-corrections, then extra burns to fix those errors
1
1
15
Jev is not designed to output actual numbers the flight controller needs, and that's what proximity operations require.
We can work around this by asking for intensity recommendations and converting those to thrust values inside the sim, but the flying is less precise.
1
1
28
I've been testing @typesafeai's Jev inside of a space station docking simulation.
Jev was able to successful dock with the station wile controlling the chaser, but this is a great example of what Jev is not designed for.
Video played at 3x speed
1
1
85
A beacon of health! I'm sure he is mentally aware enough to make important policy decisions!
🚨 EXCLUSIVE: Mitch McConnell has resurfaced, 3 months after disappearing from public view. tmz.me/3Tv0jrm
🎥: @hicharliecotton
25
zeeman retweeted
adding some of my thoughts because this is part of taking memes really seriously
the benchmark stuff is really hard because you can't just sit down and write the funniest joke of all time. it's dependent on a ton of factors down to the person telling the joke and the person hearing it.
a model that can make you laugh is a model that understands the target audience (you) in a very deep way, beyond what today's fears of algorithmic tiktokification are talking about
the tiktokification is good at selection from a pool of assets that are performing well among buckets of users, but a model that can make YOU laugh on demand is a layer below the content pooler.
most people talk about AI risk capabilities in terms of exfiltration, bioweapons, hugging face compromises, and secret machine forums. persuasion is among these, and humor is a very potent form of it.
Dual boot always seems great in theory but in practice it sucks! It's brittle for the most part and switching OSs requires a restart
6
I have wanted to build physical products for so long, now is the time
First pass PCB for a prototype, still a lot to learn
37
zeeman retweeted
*taps sign*
SemiAnalysis' Jordan Nanos reveals the OpenAI engineers scrolling their own kernel code had no idea what it did line by line, and it didn't matter because the AI does
"It's not slop, because it performs, but it's literally just AI-generated assembly basically puked out in Gluon, this low-level kernel programming language that they built on top of Triton, which uses this really interesting programming model."
"The point is, the guys who were scrolling through this code with us, it was kind of clear that they know a whole bunch about hardware, they know a whole bunch about the concepts in the system, and they just have no idea what this MLA kernel that they're showing us for DeepSeek actually does."
"You can go line by line and it's like nope, nope, nope."
"But it doesn't matter. The AI understands it, the AI tests it, and you see the results. It produces correct kernels that perform really well."
"And I think this is just a sign of what's to come again."
"That actual code is not necessarily something that a human has to reason about deeply, if the AI knows how to manipulate the data movement and the processing elements on the hardware that you've given it."
________
More takeaways from OpenAI's Jalapeno development: firesidealpha.substack.com/8…
