HNHacker News
TopNewBestAskShowJobs

leumon

722 karma · joined June 14, 2023

submissionscomments
leumon··on The New AI Consciousness Paper
It's just that I have the feeling that people avoid using the actual em-dash in fear of being accused that the text is ai generated. (Which isnt a valid indicator anyway) Maybe its just my perception that i notice this more since LLMs became popular.
leumon··on The New AI Consciousness Paper
Is there a reason why this text uses "-" as em-dashes "—"?
leumon··on Fighting the New York Times' invasion of user privacy
This problem wouldn't exist if openai wouldn't store chatlogs (which of course they want to do, so that they can train on that data to improve the models). But calling nyt the bad guy here is simply wrong because it's not strictly necessary to store that data at all, and if you do, there will always be a risk of others getting access to it.
leumon··on Claude Memory
This isn't memory until the weights update as you talk. (same applies to chatgpt)
leumon··on Show HN: OnlyJPG – Client-Side PNG/HEIC/AVIF/PDF/etc to JPG
Maybe this is just some niche use-case, but I tested it with a 268x98 png screenshot, and it made the image bigger and worse: https://files.catbox.moe/7so3z6.png
leumon··on LLMs are mortally terrified of exceptions
Or use the 'privacy redirect' extension which lets you specify your preferred nitter instance. It also works for other platforms.
leumon··on Ask HN: Radar and radio failures at Dallas area airports
https://www.theguardian.com/uk-news/2025/sep/20/heathrow-air...
leumon··on Pixel 10 Phones
You can schedule test calls and call in the scheduled time slot (US only)

Test calls confirm that your local 911 service can receive your 911 call and has the correct location information. Test calls can be scheduled by contacting your local 911 call center via its non-emergency phone number.

In your preferred search engine, search the key words “emergency communications center non-emergency number" and include the names of the city or town, state, and county or parish in your search. Test calls may need to be scheduled and are usually based on the workload experienced at the PSAP.

For more information, visit the National Association of State 911 Administrators site.

Please do not call 911 to obtain the non-emergency number.

https://www.911.gov/calling-911/frequently-asked-questions/#....

leumon··on Why are anime catgirls blocking my access to the Linux kernel?
Seems like ai bots are indeed bypassing the challenge by computing it: https://social.anoxinon.de/@Codeberg/115033790447125787
leumon··on Streaming services are driving viewers back to piracy
The average European household now spends close to €700 (£600) a year on three or more VOD subscriptions. Why not buy a Blu-ray a week, instead?
leumon··on Ch.at – A lightweight LLM chat service accessible through HTTP, SSH, DNS and API
Lmstudio or something like that on a laptop which has a lot of vram.
leumon··on Google's shortened goo.gl links will stop working next month
which makes url shorteners even more attractive for printed media, because you don't have to type many characters manually
leumon··on Graphene OS: a security-enhanced Android build
Can also recommend:

  PassAndroid: to open apple/android wallet files (airplane/cinema tickets etc.)
  Find My Device (FMD) on F-Droid: replacement for the google version, works via sms commands or a self-hosted app
  AntennaPod: Podcast App
  Breezy Weather: with multiple weather sources, great ui
leumon··on Cops say criminals use a Google Pixel with GrapheneOS – I say that's freedom
No, but curve pay or paypal pay work (or any other app not relying on google pay)
leumon··on François Chollet: The Arc Prize and How We Get to AGI [video]
He said in the video that they tested regular people (uber driver, etc.) on arc-agi2 and at least 2 people were able to solve each task (an average of 9-10 people saw each task). Also this quote from the paper: None of the self-reported demographic factors recorded for all participants—including occupation, industry, technical experience, programming proficiency, mathematical background, puzzle-solving aptitude, and var- ious other measured attributes—demonstrated clear, statistically significant relationships with performance outcomes. This finding suggests that ARC-AGI-2 tasks assess general problem-solving capabilities rather than domain-specific knowledge or specialized skills acquired through particular professional or educational experiences.
leumon··on Apple Weighs Using Anthropic or OpenAI to Power Siri in Major Reversal
They actually once used a neural net on-device in iOS 14, when they introduced the translate app. It worked offline, but actually sometimes produced some bad or hilarious translations.
leumon··on Gemini CLI
My guess: So that they can get more training data to improve their models which will eventually be subscription only.
leumon··on Google is using YouTube videos to train its AI video generator
Just like for llms for the base model you need quantity not quality. It just needs to learn how to correctly predict the next frame.
leumon··on Air India flight to London crashes in Ahmedabad with more than 240 onboard
“Thirty seconds after take off, there was a loud noise and then the plane crashed. It all happened so quickly,” said Ramesh, speaking to the Hindustan Times. He said he “impact injuries”, including bruising on his chest, eyes and feet but was otherwise lucid and conscious. https://www.theguardian.com/world/live/2025/jun/12/air-india...
leumon··on Sycophancy in GPT-4o
Possibly to get more training data.
leumon··on Widespread power outage in Spain and Portugal
Except that's not true: an auto shutdown only occurs at 47.5 Hz, at least in Germany, some countries have the limit even lower: https://de.m.wikipedia.org/wiki/Unterfrequenz
leumon··on Your phone isn't secretly listening to you, but the truth is more disturbing
> Even though these ad algorithms are not nearly perfect (try to pay attention to how often you are served ads that are entirely irrelevant to your interests), the simple fact that they are so eerily correct even some of the time is the real conspiracy here.

This could be intentional. Having too many accurate ads is having a bad effect, because you then enter the uncanny valley of noticing what the data collectors all know about you.

leumon··on Runway Gen-4
elo ranking based on user preferences. you can vote on prompts and submit new ones in the 'video arena' tab
leumon··on Gemma3 – The current strongest model that fits on a single GPU
In my opinion qwq is the strongest model that fits on a single gpu (Rtx 3090 for example, in Q4_K_M quantization which is the standard in Ollama)
leumon··on Mistral OCR
well already in 2013 ocr systems used in xerox scanners (turned on by default!) randomly altered numbers, so its not an issue only occuring in llms.
leumon··on Making o1, o3, and Sonnet 3.7 Hallucinate for Everyone
He should've tested 4.5. This model is hallucinating much less than any other model.
leumon··on GPT-4.5
this model does have a niche use-case: since its so large it does have a lot more knowledge and hallucinates much less. for example as a test question I asked it to list the best restaurants in my small town. and all of them existed. none of the other llms get this right.
leumon··on ElevenReader by ElevenLabs
Unfortunately the app is not compatible with Android 15.
leumon··on OpenAI O3 breakthrough high score on ARC-AGI-PUB
I've found gemini-1206 to be best. and we can use it free (for now), in google's aistudio. It's number 1 on lmarena.ai for coding, and generally, and number 1 on bigcodebench.
leumon··on OpenAI o1 system card
While it now can read an analog clock from an image (even with seconds), on some images it still doesn't work https://i.imgur.com/M2JouZs.png
← PreviousPage 3 of 4Next →