HNHacker News
TopNewBestAskShowJobs

luke-stanley

555 karma · joined April 29, 2012

Old Keybase stuff. Do people still use it?

[ my public key: https://keybase.io/lukestanley; my proof: https://keybase.io/lukestanley/sigs/1pBLoHuRbRQHHwd8sEqxZCe1L6gzWInT4UBKthawE2c ] 8e47ab2600d05031bdbc2a0547fb9b79a82aeb240ef79cb82c5536e286f4d4c9

I am also https://news.ycombinator.com/user?id=lukestanley I think I forgot my pass and didn't want to bother PG.

submissionscomments
luke-stanley··on Contra Chrome: How Google's browser became a threat to privacy and democracy
Robert Horn's `Visual language: global communication for the 21st century` is another great one!
luke-stanley··on ARC-AGI without pretraining
Famous last words!
luke-stanley··on ARC-AGI without pretraining
Maybe not yet, but what prevents games from getting more complicated and matching rich human environments, requiring rich human like adaptability? Nothing at all!
luke-stanley··on Show HN: Agents.json – OpenAPI Specification for LLMs
This could be more simple, which is a good thing, well done!

BTW I might have found a bug in the info property title in the spec: "MUST provide the title of the `agents.json` specification. This title serves as a human-readable name for the specification."

luke-stanley··on Directus – real-time REST and GraphQL API of any SQL database
A sustainable project income can be hard, I get that, but an auto-converting license sounds like some kind of trick weapon, not like a real GPL-3 license.

I suspect that weird versions of "open-source" licenses are rarely good for any party involved.

luke-stanley··on Claude 3.7 Sonnet and Claude Code
My key got killed months ago when I tested it on a PDF, and support never got back to me so I am waiting for OpenRouter support!
luke-stanley··on Directus – real-time REST and GraphQL API of any SQL database
GitHub's recent report has this breakdown of the top licenses: https://github.blog/open-source/open-source-license-usage-on...

MIT: 44%

GPLv2: 12.9%

Apache: 11%

GPLv3 8.8%

BSD: ~6%

Unlicense: 1.8%

LGPLv3: 1%

luke-stanley··on Directus – real-time REST and GraphQL API of any SQL database
It's not a normal license. It's not worth the hassle.
luke-stanley··on Did the Windows 95 setup team forget that MS-DOS can do graphics?
To be fair, they had to ship first.
luke-stanley··on DOGE as a National Cyberattack
There may be an error in Schneier's reasoning here: `They are also reportedly training AI software on all of this sensitive data.` Running inference isn't the same as training.

That said, I don't think this should be flagged, that seems counter free-speech. He's got plenty of other well voted articles here on HN, a lot of HN clearly value his insights. Disagreement would be better expressed as a comment rather than trying to make it go away.

luke-stanley··on Apple's Best Option: Decentralize iCloud
Moving the data to China is pretty much the same thing as backdooring the encryption, as we have seen.

Really though, decentralised is the way to go: Apple should allow custom endpoints for iCloud that users can set. If the UK or China wants to block a regional endpoint, they can.

Having user provided configurable iCloud servers might be a bit too empowering for Apple though, I suspect they would rather profit from their hosting services! But who knows, maybe they will be forced into empowering users?

luke-stanley··on Apple's Best Option: Decentralize iCloud
Option 4: Just move the UK data to China. If it's not end-to-end-encrypted then they get access anyway. No messing.
luke-stanley··on Anthropic: "Applicants should not use AI assistants"
I'm sure Anthropic have too many applications submitted that are obviously AI generated, and I am sure what they mean by "non-AI-assisted communication", they don't want "slop" applications, that sounds like a LLM wrote it. They want some greater proof of human ability. I expect humans at Anthropic can tell what LLM model was used to rewrite (or polish) applications they get, but if they can't, a basic BERT classifier can (I've trained one for this task, it's not so hard).
luke-stanley··on Show HN: DeepSeek Your HN Profile
Not bad but could have more variety? https://hn-wrapped.kadoa.com/luke-stanley
luke-stanley··on A marriage proposal spoken in office jargon
So it's low latency and fast but in some contexts, firewalled?
luke-stanley··on GPT-4o with scheduled tasks (jawbone) is available in beta
Whoops! Might have been built wrong? I'm seeing a source map error: Source map error: Error: request failed with status 404 Stack in the worker:networkRequest@resource://devtools/client/shared/source-map-loader/utils/network-request.js:43:9

Resource URL: https://cdn.oaistatic.com/assets/jbl0aowda306m4s1.js Source Map URL: jbl0aowda306m4s1.js.map

Also I am getting`Unable to display this message due to an error.`a lot.

luke-stanley··on I am rich and have no idea what to do
I'm thinking more like weather prediction with big ensemble models for human suffering prediction, you know, time series prediction, which need not be a monolithic ethical imperative with loads of subprojects and not so uniform initiatives. I guess Deepmind's SOTA weather transformers could indeed be a good starting point. I did find that an animal welfare org speculated about doing similar actually!
luke-stanley··on I am rich and have no idea what to do
Do you mean has anyone looked into making suffering models like people do weather models for regional / global maps to predict likely suffering and remediation? Or are you just pointing to touchy EA issues?
luke-stanley··on I am rich and have no idea what to do
Why not map out suffering and ways people optimally can learn to solve it, using our collective intelligence? I mean, if you're looking for ideas. Hopefully someone gets around to that one.
luke-stanley··on Apple and Meta go to war over interoperability vs. privacy
Is this Hacker News? This is way bigger than Meta. Should individuals have the freedom to access to their own phone data without needing to be a device manufacturer? Or should we depend on case by case approval of access that is outside of the users hands and something only big companies can do?

Framing it as being about `Apple vs Meta` is exactly the line of thinking that Apple wants, and it does a disservice to real Apple users, who have repeatedly been locked into a ecosystem where they have little control of their own data, where even sharing photos over Bluetooth isn't even an option! It's so locked in it's really unreasonable to frame this as being about Meta. Of course users should be able to give none / empty / fake or even real data to Meta if they so choose. But really this is about Apple, user rights, the EU, and a more fair playing field. If you read the PDF Apple put out, they are not offering any kind of middle solution, they are just saying they will, as usual, review case by case device partnership opportunities as it pleases them.

luke-stanley··on ChatGPT O3 Preview Announcement
I watched the live stream and don't recall any mention of it. It's not at that stage!
luke-stanley··on ChatGPT O3 Preview Announcement
The title is very wrong, "ChatGPT O3" is not a thing, OpenAI's new o3 model is not even demoed in ChatGPT, they don't call it "O3". The subtitle at the top is "o3 preview & call for safety researchers". The web page title is "12 Days of OpenAI".
luke-stanley··on OpenAI O3 breakthrough high score on ARC-AGI-PUB
But they did use a prompt that included a full example of how to call their latest model and API!
luke-stanley··on Rats learned to drive
Is it like an "Oh hi!" tail? Suddenly interested in the `animal tail gestures embedding space`! There must be lots of interesting ways of conveying feelings, strategic intent and misdirection along with subconscious communication.
luke-stanley··on Interview with gwern
It does make me wonder what easy way to do ML assisted shot / person detection and `blanking` is. I'm just gonna point out there is a nerd snipe danger here ;D

Although the avatar tool is probably not SOTA, I thought the 3D model was a really cool way to deal with interviewing Gwern, I am quite enjoying the current video.

luke-stanley··on ElasticSearch and many other repos are gone
Fair enough, and what I said was wrong too, so it's turtles all the way down! I butchered Cunningham’s Law, thank you for correcting me, though the name of the law is confusing since it was McGeady who pointed out: "The best way to get the right answer on the Internet is not to ask a question, it’s to post the wrong answer.", which is what I did, and what the parent comment did, it is attributed to the great Ward Cunningham, creator of the first Wiki, who is a rightful dude. I have nerd-sniped myself on this. At least I'm less wrong now, thank you.
luke-stanley··on ElasticSearch and many other repos are gone
This bears out the idea that the fastest way to get the truth on the Internet is to say something wrong first.
luke-stanley··on Video scraping: extracting JSON from a 35s screen capture for 1/10th of a cent
I put my Userscript and Python server script here: https://gist.github.com/lukestanley/c3a37ab61a45e72b74995a5c... It tries to save as Markdown. I'm sure it could be much better in many ways. But it works well enough for me right now.
luke-stanley··on Video scraping: extracting JSON from a 35s screen capture for 1/10th of a cent
I'm glad this worked for Simon, but I would probably prefer using a User Script that scrapes DOM text changes and streams them to a small local web server to append to a JSONL file that has the URL, text change and timestamp. Probably since I already have something doing this, it allows me to do a backup of things I'm looking at in real time, like streaming LLM generations, and it just relies on normal browser technology. I should probably share my code since it's quite useful. I'm a bit uncomfortable relying on a LLM to transcribe something where there is a stream of text that could be used in a robust way, and with real data, Vs well trained but indirect token magic. A middle ground might be to have grounded extraction and evidence chains, with timestamps, screenshots, cropped regions it's sourcing from, spelled out reasoning. There's the extraction / retrieval step and there's a kind of data normalisation. Of course, it's nice that he's got something that just works with two or three steps, it's good the technology is getting quite reliable and cheap a lot of the time, but still, we could do better.
luke-stanley··on Homemade AI drone software finds people when search and rescue teams can't
A lot of people do class rule based systems under the umbrella of AI, when I was a kid, I'd run Alicebot on my pocket computer. Definitely "artificial" "intelligence" and well before any of this modern fancy machine learning stuff! Definitely lots of human work. People have different ways of understanding words and AI is a term that is not well defined, to say the least.
← PreviousPage 4 of 14Next →