jmalicki

20 ms·

421.
by jmalicki 6mo ago
Eh., if you have a PyTorch model that uses non-deterministic tensor operations like matrix multiplications, I think it is fair to call the model non-deterministic, since the matmul is not guaranteed to be deterministic - the non determinism
422.
by jmalicki 6mo ago
That's why it's a hard problem, researchers are still working out how to properly get alignment with sociopathy.
423.
by jmalicki 6mo ago
No, what they're saying is the previous run could have just been lucky and not representative!
424.
by jmalicki 6mo ago
In the words of the great General Washington, we use meters and kilometers for certain unpopular sports like swimming. So for measuring an island, use of swimming units seems appropriate. https://www.youtube.com/watch?v=JYqf
425.
by jmalicki 6mo ago
Floating point associativity differences can lead to non-determinism with 0 temperature if the order of operations are non-deterministic. Anyone with reasonable experience with GPU computation who pays attention knows that even randomness i
426.
by jmalicki 6mo ago
What does that even mean? Even then, depending on the specific implementation, associativity of floating point could be an issue between batch sizes, between exactly how KV cache is implemented, etc.
427.
by jmalicki 6mo ago
The article is sparse on what pending means, but I would guess that that where condition would be enough?
428.
by jmalicki 6mo ago
I think the point of the paper is to prod benchmark authors to at least try to make them a little more secure and hard to hack... Especially as AI is getting smart enough to unintentionally hack the evaluation environments itself, when that
429.
by jmalicki 6mo ago
Health care premiums are expensive, but even the best insurance I've ever seen was a fraction of the compensation software employees were making. For some fields that's a huge amount of your compensation, but for software engineer
430.
by jmalicki 6mo ago
This isn't even training on the test data. This is modifying the test code itself to always print "pass", or modifying the loss function computation to return a loss of 0, or reading the ground truth data and having your mode
431.
by jmalicki 6mo ago
They said they used things like submitted a `conftest.py` - e.g. what would be considered very blatant cheating, not just overfitting/benchmaxxing. Did you read the AI slop in the post? This is basically a paper about security exploit
432.
by jmalicki 7mo ago
There aren't taxes on datacenters in Texas. They gain virtually nothing from them! https://www.texastribune.org/2026/04/08/texas-data-centers-s...
433.
by jmalicki 7mo ago
This is the market telling you what matters. OpenClaw has been an outstanding success, it is providing people the ability to leak their keys, secrets, and personal data, and allowing people to be subject to an incredible number of supply ch
434.
by jmalicki 7mo ago
What do you mean? He has a rabid fan base who loves him because he writes like he talks! It clearly works for him - I hate how he talks, but he seems to be an effective communicator if you only judge by results. Sadly.
435.
by jmalicki 7mo ago
That's how you get flagged as an AI!
436.
by jmalicki 7mo ago
I've been nearly hit by a bicycle-messenger looking dude in San Francisco when I was crossing the street with a "walk" sign at a crosswalk and he blew through the red light at probably about 15mph, and I have plenty of other
437.
by jmalicki 7mo ago
Is there an interpretation of the Computer Fraud and Abuse Act where using this bicycle bell to circumvent the computer system used in your headphones for active noise cancellation would be a federal felony in the United States?
438.
by jmalicki 7mo ago
How can you practically verify that OneDrive won't do this either?
439.
by jmalicki 7mo ago
When people judge blindly, the are more likely to think the human is the AI and the AI is the human. 73% judged GPT 4.5 (edit: had incorrectly said 4o before)to be the human. https://arxiv.org/abs/2503.23674 Not only a
440.
by jmalicki 7mo ago
William Kahan would beg to differ on this. https://people.eecs.berkeley.edu/~wkahan/ieee754status/754st...
441.
by jmalicki 7mo ago
If I only have 1Gbps, I'm not going to be using a NAS for something like /home, or other things that need to be performant.
442.
by jmalicki 7mo ago
1Gbps is even slower than a SATA SSD that can saturate a 6Gbps SATA link. 10Gbps gets you speeds only a few times slower than a PCIe 3.0 NVMe SSD. Except you can run that over the Internet!
443.
by jmalicki 7mo ago
For cloud services, I only need to match the bandwidth my cellphone has for my home NAS, so 1Gbps would be fine. 10Gbps is like.... my /home on my desktop could be served via NFS from somewhere else and it would probably be barely noti
444.
by jmalicki 7mo ago
> You don't wanna YOLO understanding how employment rates are calculated. You're way better off YOLO'ing reading the documentation about how they are calculated than listening to the myriad pundits deliberately trying to
445.
by jmalicki 7mo ago
While Switzerland has higher median HHI than the US as a whole, the Bay Area in California does have comparable median household income. In the Bay Area, Sonic does offer 10Gbps fiber internet in some places on new buildouts. I struggled to
446.
by jmalicki 7mo ago
> We should be using LLMs to translate from (fuzzy) human specifications to formal specifications (potentially resolving contradictions) Agreed! This is why having LLMs write assembly or binary, as people suggest, is IMO moving in the w
447.
by jmalicki 7mo ago
This reminds me of the fact that hospital-acquired pneumonia is the leading cause of death in ICUs in the US. Among people who are already going to die, the thing that kills them the most is an opportunistic bacterial pneumonia that barely
448.
by jmalicki 7mo ago
This is all just the classic "the first hit's free" business model.
449.
by jmalicki 7mo ago
Is API pricing subsidized on a unit basis? Most estimates I've seen have shown that API usage seems to be at least unit profitable (paying for infra and electricity, not R&D)
450.
by jmalicki 7mo ago
The actual /loop and cron are beyond the normal "agentic loop"

More ›