Lukas Gabriel
@lukasgabriel.net
this better not ruin my head canon that it really is forest from devs running this account
June 30, 2025 at 8:25 AM
this better not ruin my head canon that it really is forest from devs running this account
if implemented properly, no: en.wikipedia.org/wiki/High-ba...
High-bandwidth Digital Content Protection - Wikipedia
en.wikipedia.org
June 5, 2025 at 2:37 AM
if implemented properly, no: en.wikipedia.org/wiki/High-ba...
Ugh, this is the worst. Any time I use AI to look up a feature which I am not yet sure even exists, it hallucinates it. Especially if it's a feature that _should_ exist but doesn't. When I try to find it, there's nothing there. It's like it wants the feature to exist so it just makes one up. 😤
June 3, 2025 at 7:18 AM
Ugh, this is the worst. Any time I use AI to look up a feature which I am not yet sure even exists, it hallucinates it. Especially if it's a feature that _should_ exist but doesn't. When I try to find it, there's nothing there. It's like it wants the feature to exist so it just makes one up. 😤
They are switching to year-based versioning, so it might be macOS 26.
June 2, 2025 at 7:44 AM
They are switching to year-based versioning, so it might be macOS 26.
oh and they become almost completely useless with short inputs like LinkedIn comments :/
May 11, 2025 at 12:17 PM
oh and they become almost completely useless with short inputs like LinkedIn comments :/
It has been shown that AI text detectors have a very high rate of false positives which disproportionately affect text written by non-native speakers: www.sciencedirect.com/science/arti...
The only thing worse than AI slop is people telling you that your content is AI-generated when it isn't 🙄
The only thing worse than AI slop is people telling you that your content is AI-generated when it isn't 🙄
GPT detectors are biased against non-native English writers
GPT detectors frequently misclassify non-native English writing as AI generated, raising concerns about fairness and robustness. Addressing the biases…
www.sciencedirect.com
May 11, 2025 at 11:56 AM
It has been shown that AI text detectors have a very high rate of false positives which disproportionately affect text written by non-native speakers: www.sciencedirect.com/science/arti...
The only thing worse than AI slop is people telling you that your content is AI-generated when it isn't 🙄
The only thing worse than AI slop is people telling you that your content is AI-generated when it isn't 🙄
class H amnestics 😳
February 11, 2025 at 1:47 PM
class H amnestics 😳
I ran a DeepSeek R1 model locally at it will give detailed info about issues like Taiwan, Hong Kong, persecution of chinese ethnic minorities, protests, ...
What do you mean exactly?
What do you mean exactly?
February 1, 2025 at 11:57 PM
I ran a DeepSeek R1 model locally at it will give detailed info about issues like Taiwan, Hong Kong, persecution of chinese ethnic minorities, protests, ...
What do you mean exactly?
What do you mean exactly?
Or maybe a smaller LLM runs in the background and checks the output for adherence to rules.
January 28, 2025 at 4:28 PM
Or maybe a smaller LLM runs in the background and checks the output for adherence to rules.
From what I saw it looked like a simple check for keywords in the output of the web version, and when it detects something it doesn't like, it stops the chat. Same thing Bing did/does. Honestly not the huge news everyone makes it out to be.
January 28, 2025 at 4:27 PM
From what I saw it looked like a simple check for keywords in the output of the web version, and when it detects something it doesn't like, it stops the chat. Same thing Bing did/does. Honestly not the huge news everyone makes it out to be.
Does the self-hosted model also censor output, or just the chatbot on the website?
January 28, 2025 at 12:05 PM
Does the self-hosted model also censor output, or just the chatbot on the website?
