thadk

Can anyone share syllabi for these?

On macOS, I wanted to be able to paste HDR HEIC screenshots straight into GitHub in Safari and have them get converted automatically to these HDR-capable formats. Normally, when you switch Screenshot.app to HDR, paste to GitHub stops working.

I put together this Hammerspoon spoon with a few brew libraries and some inline Swift code.

You paste twice quickly to trigger it, and you can iterate formats with a keystroke. https://gist.github.com/thadk/8b178aa93dafcab92ca579825f8191...

It is nice to have simulated confirmation of the M4 Max 128gb limitations on Gemma4 31B vs Gemma4 26B A4B at Q4:

Time to first token (TTFT) on 5,687 tokens, like pasting a short code block into the prompt, takes 36 seconds vs 4.7 seconds. It also breaks down the seconds for each step like KV and prefilling.

Gemma4 31B TTFT on the M5 Max 128gb with same setup is showing as 19 seconds.

vgr's waldenponding is about the "And I mourn what we don’t get to hear from them."

An ineffable game happens that one's retreat diminishes.

https://www.econtalk.org/venkatesh-rao-on-waldenponding/ https://studio.ribbonfarm.com/p/against-waldenponding-ii

2006 article by Quinn Norton that was linked in USRP's Wikipedia: https://www.wired.com/2006/06/gnu-radio-opens-an-unseen-worl...

Reminds of this HN classic, with gratitude to the author for recounting: https://news.ycombinator.com/item?id=17694032

Well, seems like this indicates another long-time and discerning user profile for Apple's pricier XDR display products.

Would OpenPangram's LoRa fine-tuning training be able to pick up on the patterns of steganography left behind by LLM vendors? https://huggingface.co/pangram/editlens_Llama-3.2-3B

(pardon for reposting just now from off-front-page thread on https://declaude.org/watermarking/, just thought of question)

Would OpenPangram's LoRa fine-tuning training be able to pick up on the patterns of green left behind by LLM vendors? https://huggingface.co/pangram/editlens_Llama-3.2-3B

Still reading but I like how Cat Hicks in Psychology of Software Teams (2027) confronts this dilemma on page 9: "Agile Manifesto said, 'Individuals and interactions over processes and tools.' DevOps made the argument that cohesive processes in software development needed to be invested in... Agile, DevOps, Platform Engineering, or DevEx, every engineering organization I've ever met with is struggling with implementing these ideals in the real world."

Simon's retelling is more compact but it also invites anthropomorphization of the sharing of the familiarity with the message board which re-emerged a few times.

Zvi's retelling handles this better. Zvi speculates that the secret message board familiarity was carried because it had been trained into the May-and-subsequent models: https://thezvi.substack.com/p/openai-trained-its-models-for-...

fil's AirOcean projection on the original ObservableHQ is a fun starting point for rendering your own such map in Javascript from scratch:

https://observablehq.com/@fil/airocean-projection

The source data is down for my vector tile remix of it but here is a screenshot: https://static.observableusercontent.com/thumbnail/0fac8a928... and the code: https://observablehq.com/@thadk/reprojecting-vector-tiles

I did it as a little MapTime project in 2019 or so.

It bugs me so much when eReaders re-render the ebook and everything moves slightly. I lose my eye muscle memory of where things were.

Until now, I had not thought to try to lock down the settings that lead to accidental reflows. It sounds less difficult in KOReader than I expected. I may try to stabilize it for my next ebook I can get into KOReader.

Would dropping airplane windows lower radiation exposure somewhat, especially on those polar routes?

Three or so years ago, Omar, the creator of DSpy pointed out on Twitter that ~LLMs get better most by better internal collaboration. Wish I could find it.

It seems to me that LLMs and particularly chatbots have already allowed for bigger scale collaboration within the LLM companies versus what was possible within the prior cohort of big platform companies.

Has the result just been taller towers, or actually a change of what is possible?

maybe they did already via the "formerly known as" comment in the paper?: https://www.spongium.org

Does anyone have hints on what kinds of prompts are most used for a distillation like this—SWE-Bench sorts of things?

Is reconstructing the compressed knowledge in the model like reconstructing a lossy JPG or MP3 a reasonable analogy?

Somewhat related, but TIL the macOS Screenshot.app utility (and the system wide keyboard shortcuts) in Sonoma got HDR capability to help re-convey just a bit more fidelity. Even in PNG, it keeps the display profile, just not HDR.

Open Utilities->Screenshot.app Options->Capture/Capture Format->HEIC. Note, it changes the system screenshot default away from PNG too.

2014 https://web.archive.org/web/20140515004053/https://www.jason...

His Studio Display 3D printable model MacBook holder: https://arslan.io/macbook-holder-for-apple-studio-display/

Microsoft poorly manages token use of most expensive models in a pilot. Then they use that failure to advertise/position their own Github Copilot agents to procurement teams, over the now widely validated Claude Code-based agents.

At least Codex is trying to win validation on merit.

Article does not mention the other reason: in the interview with Dwarkesh, Amodei remarked about how other organizations are copying or training off Opus for their models.

By delaying allowing others to train off Mythos, they hold their SWE-Bench Pro head start longer so among other things, the USG can't but notice Anthropic's lead when they're deliberating on whether to further substantiate the "supply chain risk".

Anyone else read this as "An idiot's guide to Pb optimization for proteins," as in avoiding contaminated dietary protein isolates?

Muji stocked a stapler like this for some time.

Ffmpeg.WASM is really good and might manage all these steps before even uploading.

Since it kind of had an LLM flavor in the original anyway, I had an LLM redraft this for my particular non-Japanese intercultural situation, and it wasn't valueless. I had to tell it to "Make sure to keep the same technology focus of the original."

The Land of Silence and Darkness (1971) mentioned is streaming on Amazon Prime in the USA, amongst other streaming services:

https://letterboxd.com/film/land-of-silence-and-darkness/

"We're not perfectly good at preventing some of these other [model] companies from using our models internally." — well maybe this says something about how Opus 4.5 and Opus 4.6 have the same SWE bench score.

Huh, so if a China-based lab takes ARC-AGI-2 on the new year, then they can say they had just-shy of a solution anyway.

It was probably from the other day when roon realized that normal people have it slower than staff.

Then from that they realized they could just run API calls more like staff, fast, not at capacity.

Then they leave the billion other people's calls at remaining capacity.

https://thezvi.substack.com/i/185423735/choose-your-fighter

> Ohqay: Do you get faster speeds on your work account?

> roon: yea it’s super fast bc im sure we’re not running internal deployment at full load