Jonathan
_@jonathansm.com
npub1uqee...jckg
Hacker, cypherpunk. All memes are my own.
How are the election predictions this insanely close 4 days before the election?
https://www.metaculus.com/questions/11245/2024-us-presidential-election-winner/
It feels good to finally self host my website on my own server rather than on GitHub Pages. Now I have complete control.
I just bought a smart ring for $10 that’s completely hackable over BLE. It has an accelerometer, Blood Oxygen meter, and a heart rate monitor. The question now is: what should I do with it? Any ideas?
Who will be next? Apple? Facebook?
View quoted note →
Google is also joining the Nuclear party.
https://archive.is/2024.10.14-192552/https://www.wsj.com/business/energy-oil/google-nuclear-power-artificial-intelligence-87966624
View quoted note →
With the amount of energy drinks teens are drinking, will caffeine eventually become a national strategic resource?
Browser extension idea: what is the probability the page I’m looking at was AI generated?
Or a crowdsourced list of AI slop websites.
Since I’m too cheap to rent a nice server from Digital Ocean I had to rewrite some web apps from Python to Golang to save memory. The performance and RAM savings are dramatic. While Rust is probably still my language of choice for most non-trivial projects Go makes writing backends incredibly easy.
The regular Transformer may have just been unseated as the de facto architecture for LLMs. I’m no expert but Diff Transformers seem like a massive improvement over regular Transformers.
arXiv.org
Differential Transformer
Transformer tends to overallocate attention to irrelevant context. In this work, we introduce Diff Transformer, which amplifies attention to the re...
Hats off to JD Vance, I may not agree with him on a lot of his new political opinions since he’s cozied up to Trump but he’s a smooth debater. From the very first question you could see he was much better in that environment.
This is actually super entertaining.
https://www.reddit.com/r/notebooklm/comments/1fr31h8/notebooklm_podcast_hosts_discover_theyre_ai_not/
View quoted note →
I know people complain that the podcasts NotebookLM generates are a little vapid and bland, but so are 90% of podcasts.
Most podcasts probably shouldn’t exist in the first place because neither the people nor the topics are that interesting, and of the ones that should exist most could probably be replaced with AIs. An LLM could probably handle interviewing the millionth author of some shitty book that no one is going to read anyway. Hell the LLM could probably handle pretending to be the author as well.
Simon Willison’s Weblog
NotebookLM’s automatically generated podcasts are surprisingly effective
Audio Overview is a fun new feature of Google’s NotebookLM which is getting a lot of attention right now. It generates a one-off custom podcast a...
Wait wtf, California’s idea of “banning” something is putting you on a public list if you do it and then doing literally nothing else? If there’s no teeth to a bill why even pass it in the first place?
https://archive.ph/2024.09.30-175714/https://www.nytimes.com/2024/09/30/us/california-bans-legacy-admissions-private-universities.html
macOS Sequoia and iOS 18 feel like gimmick updates with little toys because the real meat of the updates (Apple Intelligence) haven’t been released yet.
At this point maybe we’re just reinventing HTTP/TLS but would multi-part notes help solve the note size limit problems? Its kind of annoying that you can’t follow more then around 1000 people or make a big list before a lot of relays start rejecting the notes.
I love his description of how riding a driverless car feels like a fast firmware update for your brain.
Allen Pike
Our Unevenly Distributed Future
How self-driving cars become mundane.
Just updated to MacOS Sequoia. The iPhone mirroring feature is pretty sick. I’m typing this on my Mac into Primal running on my phone which is currently in my pocket.
Now terrorists get sympathetic pieces in the WaPo when they die?
