verdverm

14 ms·

391.
by verdverm 1mo ago
Internal vector recurrence is more or less what the Universal Transformer paper proposes. This recent paper does so by sharing the transformer weights in a single recurrence setup: https://arxiv.org/abs/2607.22083 , and
392.
by verdverm 1mo ago
Most people do not hide intent, we also write things down at work, eg. a ticket in kanban
393.
by verdverm 1mo ago
Sebastian Raschka put out a video on this "bombshell" report (he's more skeptical) https://www.youtube.com/watch?v=KT4n-z_4QJU
394.
by verdverm 1mo ago
This idea that people are somehow compelled to like Ai because they lack political power seems a false hubris. Do you know the CCP lets people protest some things, though certainly not party control?
395.
by verdverm 1mo ago
I'm not sure where you are pulling that bitcoin statement from, presenting it as a quote, but it is not mine. My understanding is the gas turbine generators are the thing creating noise. We've had data centers for decades, noise i
396.
by verdverm 2mo ago
What does "it works" mean in this context? Are we prioritizing short-term political wins at cost of long-term competitiveness? Why are non-Western populations so excited by Ai while the West is eating fud? I suspect the data cente
397.
by verdverm 2mo ago
I find the use of an AI writing detector to be a foundational flaw in this study, we'll never know the answer and I don't think that matters. I'm more curious to see the effects on quality vs quantity
398.
by verdverm 2mo ago
Sounds like something a shovel salesman would say. It doesn't make sense to trade deterministic, efficient code for unreliable agents. The agents will use the same tools we do and will be more effective that way. I've started payi
399.
by verdverm 2mo ago
Yeah, we let big tech go to far. I think a foundational poor decision was letting algos use big data to figure out what triggers our engagement in online platforms, which has spread to more industries, notably from social media to news medi
400.
by verdverm 2mo ago
One can also drive past a mountain, any Gemini is DOA where I work, Google has disappointed us across many products in the last 2 years, they are trashing their brand reputation with slop, poor delivery, false promises, and user hostile pri
401.
by verdverm 2mo ago
Because he's doing PR, those jobs are primarily during the construction phase, and they are not your typical electrician / plumber skills, they are high voltage and high volume. He's doing the media spin thing. I met a guy wh
402.
by verdverm 2mo ago
s/Investment/Illusion/?
403.
by verdverm 2mo ago
When this decision was made years ago, it was very NIH / I want to make a language because it sounds cool. It does seem this was toned down quite a bit from those days to something that is yaml with curl braces, still reeks of hackish
404.
by verdverm 2mo ago
More likely to get a Prince of Persia where you go get gifted airplanes and make backroom deals to slice up Gaza for hotels
405.
by verdverm 2mo ago
How related is this to AlphaFold and the release of the giant dataset of protein folds? Perhaps there is less of a need for finding the folds and more for understanding them? I heard someone in the know say it will take decades to go throug
406.
by verdverm 2mo ago
There are nuts at that level in the Trump administration, RFK and Hegseth are two glaring examples. Trump's natsec council is largely empty, not where near the number and expertise of people as other admins or his first time round
407.
by verdverm 2mo ago
better, but works against one of the core ethos in AT
408.
by verdverm 2mo ago
I agree, that's just not where ATProtocol is (currently only available in world/anonymous-read mode). Personally, I find the permission proposal rather lacking (ACL vs the advancements / learning since) and user requested fea
409.
by verdverm 2mo ago
There's always the future. I stopped using ATProto after the revelations that they hid their VC raise from the community for almost a year (and some other things, straw eventually breaks a camel's back) - and almost all social med
410.
by verdverm 2mo ago
1. I would avoid gendering software or saying "it lies," there is no identity or intent for anything it does 2. Why are your expectations no failures? 3. How much have you used llm -> agents? What have you tried to have them do
411.
by verdverm 2mo ago
If the market demands products you cannot deliver, buy politicians, laws, and pr firms instead
412.
by verdverm 2mo ago
Most people right now do not want their social media data in world/anonymous-read mode nor want Ai accessing it and doing things with it.
413.
by verdverm 2mo ago
Been thinking about this, kids are going to know about it and probably use it outside of school. Is there a plan to educate them prior, like we do with sex ed?
414.
by verdverm 2mo ago
Carney has some ideas ...but I think this is more the news media ecosystem turning an online molehill into an ad click mountain. Some random former admin person said something ridiculous online, not really news worthy imo
415.
by verdverm 2mo ago
The models have reached, on average, the 256k/1M window size and that is more than enough for them to complete most tasks without losing attention. I suspect more people are using fresh sessions and markdown between them, than trying t
416.
by verdverm 2mo ago
https://mail.cyberneticforests.com/models-dont-go-rogue/
417.
by verdverm 2mo ago
I used dagger for the sandbox in my personal harness. Docker also has a sandbox (microvm based) product now. Both obscure the credentials, but I do not think it eliminates them base64 encoding it (or some other transformation). It does step
418.
by verdverm 2mo ago
> It's pathfinding through language generation. What an interesting sentence (to describe inference time reasoning )
419.
by verdverm 2mo ago
The question remains, what will they do when it is ignored by Trump and his cronies?
420.
by verdverm 2mo ago
It sounded closer to Scion, which wraps the various harnesses. Given your reply, I'm not sure where your project sits in the stack or what you mean by `agy` support in another harness

More ›