gcr

6 ms·

61.
by gcr 4mo ago
This page starts flickering madly when I pinch-to-zoom. Until a11y details like this are figured out, I don’t think this should be considered for general use beyond a cool prototype.
62.
by gcr 4mo ago
I could write a program to generate the first 100MB of pi in a couple kilobytes. That certainly counts as “data compression” but isn’t useful outside this particular problem instance.
63.
by gcr 4mo ago
Now you understand why I thought something written in good faith might be interpreted as shade. Happens all the time.
64.
by gcr 4mo ago
This is a great example of how to make a minimal app bundle with Swift. Thank you! All an App needs on MacOS seems to be a binary and a little .plist
65.
by gcr 4mo ago
I don’t understand the shade being thrown ?
66.
by gcr 4mo ago
There are two forms of compression relevant to LLMs: 1. Reduce the number of parameters 2. Reduce the resolution of each parameter (quantization) For 1, changing the architecture is typically only possible by the labs producing the models,
67.
by gcr 4mo ago
How close to 0,0,0 are you talking? Subnormal floating point shenanigans maybe?
68.
by gcr 4mo ago
Yeah seconding this question, I’m very curious how this compares to wine
69.
by gcr 4mo ago
This is interesting! But one important aspect isn’t being simulated: what about the curvature of the disk itself? Shouldn’t this cause the individual scanlines to bend slightly?
70.
by gcr 4mo ago
Publish hash(image) on the blockchain at a verifiable time, then publish the image itself. The image contains the previous block’s hash. Wouldn’t this establish both a lower bound and an upper bound on the time the image could have been pro
71.
by gcr 4mo ago
Here in the US? Probably a large double-digit percentage of cases imo…
72.
by gcr 4mo ago
Sounds like a great idea, thanks for volunteering to pitch in and help!
73.
by gcr 4mo ago
It certainly can be!
74.
by gcr 4mo ago
That’s an amazingly pessimistic take, and a shortsighted one. Glitch and the TADC team are more than just one person. Gooseworx is burned out, but her work helped kickstart many early careers among her staff. I think the only people who cha
75.
by gcr 4mo ago
For context, Google initially refused to merge JpegXL as a strategy play to promote AVIF, which was in use by other teams (i think Photos?). Internally, chrome engineers were supportive of jxl but were overridden by leadership. I guess toda
76.
by gcr 4mo ago
I think it’s throwing shade on Microsoft, which is urging people to upgrade to W11 for “security reasons.” Since Microsoft owns GitHub, their decision to shut down Windows 7/8 action runners is a sneaky way to help drive customers towa
77.
by gcr 4mo ago
Between this, Iron Lung, and The Amazing Digital Circus finale getting a cinema release, I think this is shaping up to be a great year for small movie productions
78.
by gcr 4mo ago
Yes but in ways whose solutions admit some level of creativity or ingenuity
79.
by gcr 4mo ago
On the website, their joke has some formatting: “we recommend you upgrade <strikethrough>to play OpenRCT2</strikethrough> for security reasons!”
80.
by gcr 5mo ago
Shouldn't that be part of the test? Real-world systems need to be able to say "I don't know." This is a test about misinformation after all, and overconfident responses contribute to that. Teasing out the difference betw
81.
by gcr 5mo ago
For an alternate albeit somewhat contrarian view, also see Ed Zitron’s piece that add context to Anthropic’s profitability: https://www.wheresyoured.at/anthropics-profitability-swindle... TL;DR Ed argues that the deal betwe
82.
by gcr 5mo ago
Nifty! What pi extensions provide the batch web fetch?
83.
by gcr 5mo ago
Thanks, appreciate the info. For whatever it’s worth regarding recency, I’m testing the main llama-cpp branch that was pulled and built on 2026-05-25 running unsloth/Qwen3.6-35B-A3B-MTP-GGUF:Q4_K_M, my hardware platform is M1 Max 32GB
84.
by gcr 5mo ago
Speculative decoding shouldn't actually change the accuracy of the response. The draft model drafts a couple tokens, and the inference framework verifies that the larger model would have picked them. However, I've found that specu
85.
by gcr 5mo ago
Arguing with the agent is an anti-pattern IMO. When Claude acts up, my strategy is to rewind the conversation to the point where the misunderstanding started, revise what I said just before, and then continue from there. I’ve found that let
86.
by gcr 5mo ago
Ah. Most Claude users let Claude write their tests for them and I assumed you were too. Sorry.
87.
by gcr 5mo ago
What does “much better tested” mean to you? If you don’t read the tests to check they confer your intent or specifications, they’re more like tautologies than tests, you know?
88.
by gcr 5mo ago
This matches my experience too. The little a3b model is quite capable for its size class, as is the 27B model, but it’s still an order of magnitude less effective than Claude on the “effectiveness / time” curve
89.
by gcr 5mo ago
Ollama and LM-Studio are fine. Their main advantage is that they have a nice way to browse models -- LMStudio from huggingface and Ollama from their own curated list. Both are great ways of getting started. Pick LM-Studio if you'd like
90.
by gcr 5mo ago
which models do you have in mind? grok from xai?

More ›