frumiousirc

5 ms·

1.
by frumiousirc 6d ago
What is the interplay between such delegation and token cache timeout? If Fable is truly active for 4 days, enough to keep the cache hot, the cost would be astronomical. OTOH, if Fable idles while subagents are active, then each awakening
2.
by frumiousirc 7d ago
The point is you can quickly get any yes's, and then subject them to more expensive scrutiny.
3.
by frumiousirc 7d ago
> the maximum achievable throughput is 26 bps at 1.5 m and goes down to 22 bps at 4.5 m. And when there are more than one transmitters in real world conditions instead of anechoic chamber? It's cool and all and if this work is for i
4.
by frumiousirc 8d ago
Thanks! I've used graphviz for more than a decade and had never learned of gvpr.
5.
by frumiousirc 8d ago
Do you have thoughts on the prospects for inter-operation between the "recoll" application and hister? Like, maybe an adapter that let's hister read recoll's xapian database/index? Or, a tool that converts/syn
6.
by frumiousirc 10d ago
They also define and measure an "IPJ" as well as "IPW" > the NVIDIA B200 achieves 1.6× to 2.3× higher intelligence per joule than the APPLE M4 MAX across QWEN 3 and GPT-OSS model variants The B200 = "cloud",
7.
by frumiousirc 10d ago
> Unless I misread it, are they saying local GPUs use less energy? You misread it. From the abstract: > local accelerators achieve at least 1.4× lower IPW than cloud accelerators running identical models That's "intelligenc
8.
by frumiousirc 12d ago
Apparently, nearlyepic is trying to deliver two veiled insults and a critique of judgement or perhaps of the current state of broader affairs. I'll attempt to translate: - "no idea what you’re talking" = "you have outso
9.
by frumiousirc 19d ago
The closest I get to finding skills useful is when I find myself repeating myself to an LLM. This tends to happen most when I am starting new projects and want to communicate basic design principles and patterns to follow and libraries to
10.
by frumiousirc 23d ago
My dictionary writes: "precision * accuracy = fidelity".
11.
by frumiousirc 24d ago
> not solved I agree. Other reasoning traces simonw quoted in his blog post showed that the model made changes to consider realistic fork rake. I think this may also be the first case where the chain went inside the seat stays. The ov
12.
by frumiousirc 28d ago
Datalog seems like a way to "spell" knowledge graph (KG). The article touches on Datalog statements changing over time. One ingredient I think would be good to add to the system is to make every statement carry "providence&q
13.
by frumiousirc 28d ago
> Interestingly, people have seen my GUI Emacs and commented "I didn't know you could get images in a terminal". Then I have to tell them it's not a terminal... ...[finishing with], though you *can* also display ima
14.
by frumiousirc 1mo ago
Linux didn't (yet) kill Microsoft. Microsoft absorbed that shot. Then the Git arrow went straight to cold black heart of Microsoft. The next few months will determine if they survive it. If they do, what will we see from the third
15.
by frumiousirc 1mo ago
That is not what I read from danluu's words. He merely stated in the prompt that there is a holdout set and did not iterate to minimize error against the holdout set. In a prior attempt he prompted with only "don't overfit&
16.
by frumiousirc 1mo ago
> Bicycle is the right shape. Pelican beak is excellent. Nice background. Relative to other results I agree. But on an absolute measure, there is not a single element in the current bicycle that is real-world accurate and many elements a
17.
by frumiousirc 2mo ago
I use bubblewrap, which I believe claude code also has internally but not for its `Bash()` tool. I wrap bubblewrap in a script that supports config files to allow different "profiles" of use (analogous to eg firefox profiles). Th
18.
by frumiousirc 2mo ago
Your question piqued my curiosity. I thought maybe NERSC doesn't accept jobs from private corporations but I checked and that's not true, as long as results are not held proprietary. Perhaps ANL was used as they have a lot of comp
19.
by frumiousirc 2mo ago
Well, vinyl records are an analog medium so their sampling rate is infinite. More important would be physical and electrical band limits of the original recording, cutting and pressing devices and of course the playback devices.
20.
by frumiousirc 2mo ago
There's no mention of "LLM" nor "language". It does mention "foundation model" which includes LLMs but that also includes non-LLM architectures and non-text data. Many of the Genesis Initiative proposals
21.
by frumiousirc 2mo ago
> and there are career civil servants (all the government scientists are under this category) US national lab scientists are not even civil servants. The labs themselves are run by a corporation under contract to the DOE and the scienti
22.
by frumiousirc 2mo ago
> I find myself to be a huge bottleneck, having to do technical design and design reviews, to make sure they're actually working on the right things. I find this depends on the LLMs being used and the person using them and the probl
23.
by frumiousirc 2mo ago
Of course I push my car to the store! It has the GPS and I've lost the ability to self-navigate. Plus, how else could I carry my groceries. :)
24.
by frumiousirc 2mo ago
I think this characterization misses the feedback loop that exists. The OP is not sitting in a one-way flow of info: LLM->human->code but rather a the center of a feedback loop: LLM<-->human<-->code. H
25.
by frumiousirc 2mo ago
It could be useful more as a labor saving device to enable censoring while allowing the human to engage in easy copy-paste. Examples I think of are inclusion of internal DNS names, usernames and human names and email addresses that may be
26.
by frumiousirc 2mo ago
A key feature about data-flow programming that seems too often missed is that it is (or can be) hierarchical. Define a subgraph of atomic nodes as itself a node with its ports formed from as-yet unconnected ports of its atomic constituents.
27.
by frumiousirc 2mo ago
Perhaps bad/naive ideas, but consider these options for unlocking where the requirements changes after one incorrect entry. - After an incorrect key, require two (or more) consecutive valid key entries. - After an incorrect key, no lon
28.
by frumiousirc 2mo ago
It seems Google already /dev/null's this search term. Your search - site:claude.ai/share - did not match any documents.
29.
by frumiousirc 2mo ago
> I really wish Ruff would introduce something similar to Nix’s stateVersion Two approaches: You can pin the version of `ruff` in `pyproject.toml`. You, to let ruff command version advance but pin the settings to a prior version: u
30.
by frumiousirc 2mo ago
ruff is not new. It's also truly really fast. On an as-yet un-ruff'ed 32k SLOC python package it returns in 158ms finding 2k errors. Having it fix errors is also very fast. What's really tiring is reading such facile &quot

More ›