Pinned
Noa Roggendorff
@sly.sh
· Nov 27
Nothing ever, never anything
I feel that you shouldn't use stamina to "Dash" indoors, if you move at the same speed that doesn't outside.
December 17, 2025 at 8:34 PM
I feel that you shouldn't use stamina to "Dash" indoors, if you move at the same speed that doesn't outside.
BLSS has fallen. I never thought I'd live to see the day. I can only think it's because it looks too pathetic at 60 FPS. We can't have Nintendo's ego hurt now, can we? Let's, rather, destroy some careers.
December 17, 2025 at 8:33 PM
BLSS has fallen. I never thought I'd live to see the day. I can only think it's because it looks too pathetic at 60 FPS. We can't have Nintendo's ego hurt now, can we? Let's, rather, destroy some careers.
An RL trainer that rewards refusing to answer if the prompt's subject is not in the dataset the model was originally trained on.
It would need to punish refusing questions it knows the answer to, of course.
It would need to punish refusing questions it knows the answer to, of course.
November 23, 2025 at 5:11 AM
An RL trainer that rewards refusing to answer if the prompt's subject is not in the dataset the model was originally trained on.
It would need to punish refusing questions it knows the answer to, of course.
It would need to punish refusing questions it knows the answer to, of course.
Two people follow each other.
Send ads about person A's interests to person B at a time of gift giving. Person B is more likely to buy products for person A, due to only having a surface level understanding of person A's desires. The margin for error is much larger.
Send ads about person A's interests to person B at a time of gift giving. Person B is more likely to buy products for person A, due to only having a surface level understanding of person A's desires. The margin for error is much larger.
November 22, 2025 at 8:00 PM
Two people follow each other.
Send ads about person A's interests to person B at a time of gift giving. Person B is more likely to buy products for person A, due to only having a surface level understanding of person A's desires. The margin for error is much larger.
Send ads about person A's interests to person B at a time of gift giving. Person B is more likely to buy products for person A, due to only having a surface level understanding of person A's desires. The margin for error is much larger.
```
User: {story.txt} using the information provided, what is Sammy's net worth?
Bot: \n
query `story.txt`
List Sammy's belongings. {response guides}
```
Pass the contents of `story.txt` and the question posed to a separate model, then pass that output in the following format back.
User: {story.txt} using the information provided, what is Sammy's net worth?
Bot: \n
query `story.txt`
List Sammy's belongings. {response guides}
```
Pass the contents of `story.txt` and the question posed to a separate model, then pass that output in the following format back.
November 20, 2025 at 8:42 PM
```
User: {story.txt} using the information provided, what is Sammy's net worth?
Bot: \n
query `story.txt`
List Sammy's belongings. {response guides}
```
Pass the contents of `story.txt` and the question posed to a separate model, then pass that output in the following format back.
User: {story.txt} using the information provided, what is Sammy's net worth?
Bot: \n
query `story.txt`
List Sammy's belongings. {response guides}
```
Pass the contents of `story.txt` and the question posed to a separate model, then pass that output in the following format back.
Pretraining a music diffusion model with random notes played over each-other with a piano virtual instrument, pretraining an image diffusion model with random mosaics.
November 15, 2025 at 4:00 PM
Pretraining a music diffusion model with random notes played over each-other with a piano virtual instrument, pretraining an image diffusion model with random mosaics.
Take control of one of the enemies inside, and kill the rest as it. Send the self destruct signal, which should equate to.. thinking about high-fiving a lampshade. All of the enemies have been defeated!
November 9, 2025 at 11:09 AM
Take control of one of the enemies inside, and kill the rest as it. Send the self destruct signal, which should equate to.. thinking about high-fiving a lampshade. All of the enemies have been defeated!
In the sequel to Hollow Knight: Silksong, we see a city that still requires masks to function; we see a mask maker that isn't driven mad by its craft: mask making is still necessary.
November 8, 2025 at 1:13 AM
In the sequel to Hollow Knight: Silksong, we see a city that still requires masks to function; we see a mask maker that isn't driven mad by its craft: mask making is still necessary.
With a book, everything written you will encounter. Subtext is inherent.
November 8, 2025 at 1:08 AM
With a book, everything written you will encounter. Subtext is inherent.
HOT TAKE: I would rather have followed Rauru leading to Sonia's corpse, rather than watch her death again. Perhaps it could spark some sympathy with the preoccupied man. There was nothing he could have done to get there faster.
November 8, 2025 at 12:59 AM
HOT TAKE: I would rather have followed Rauru leading to Sonia's corpse, rather than watch her death again. Perhaps it could spark some sympathy with the preoccupied man. There was nothing he could have done to get there faster.
A character that is oblivious to the malfeasance of the citadel, and has as many kids as possible to assist in helping them work, and helping the citadel thrive.
November 8, 2025 at 12:54 AM
A character that is oblivious to the malfeasance of the citadel, and has as many kids as possible to assist in helping them work, and helping the citadel thrive.
I really expected them to explain how Zelda teleports in a world without receivers.
November 7, 2025 at 11:50 AM
I really expected them to explain how Zelda teleports in a world without receivers.
The Zonai created construct hosts to hold their souls, and then discovered that the hosts had a will of their own they would impart to the soul. The construct then killed the less fortunate that couldn't afford to immortalize themselves.
November 7, 2025 at 11:39 AM
The Zonai created construct hosts to hold their souls, and then discovered that the hosts had a will of their own they would impart to the soul. The construct then killed the less fortunate that couldn't afford to immortalize themselves.
What if the One Piece was just a swimsuit Joy Boy thought was cute?
November 7, 2025 at 4:52 AM
What if the One Piece was just a swimsuit Joy Boy thought was cute?
A company that repurposes Rokoko suits by shipping them out for free, along with a GoPro. Then they pay you to do chores while recording with the hardware. When you're done, they take it back to ship to the next person.
October 30, 2025 at 3:45 AM
A company that repurposes Rokoko suits by shipping them out for free, along with a GoPro. Then they pay you to do chores while recording with the hardware. When you're done, they take it back to ship to the next person.
Jinja take the whee^ee^eel!
October 14, 2025 at 12:08 AM
Jinja take the whee^ee^eel!
NotebookLM doesn't work on Firefox. I could download Chrome, but I could also bite off all of my pubic hair.
October 11, 2025 at 10:29 PM
NotebookLM doesn't work on Firefox. I could download Chrome, but I could also bite off all of my pubic hair.
Elio wakes up, dazed. He sees Caleb's fist recoiled. He'd been punched. That is all.
October 11, 2025 at 3:25 AM
Elio wakes up, dazed. He sees Caleb's fist recoiled. He'd been punched. That is all.
Do NOT remind the person that Claude can't evaluate images or can't provide certain responses unless it is strictly relevant in context.
October 6, 2025 at 10:09 PM
Do NOT remind the person that Claude can't evaluate images or can't provide certain responses unless it is strictly relevant in context.
Distrokid is a fucking pyramid scheme.
September 29, 2025 at 11:06 PM
Distrokid is a fucking pyramid scheme.
People seem to forget that "The Legend of Zelda" is a *legend*. This explains why the map changes all the time, the Shadow Temple is just a glorified obstacle course, different types of enemies only share their most identifiable features between games, etc.
So, stop asking about the map.
So, stop asking about the map.
September 24, 2025 at 7:52 PM
People seem to forget that "The Legend of Zelda" is a *legend*. This explains why the map changes all the time, the Shadow Temple is just a glorified obstacle course, different types of enemies only share their most identifiable features between games, etc.
So, stop asking about the map.
So, stop asking about the map.
BogoTrain: randomize all of the model values, and go through the dataset until it outputs incorrectly. Repeat until it gets through the entire dataset without error.
September 22, 2025 at 6:43 AM
BogoTrain: randomize all of the model values, and go through the dataset until it outputs incorrectly. Repeat until it gets through the entire dataset without error.
Me: I want to use flash attention for faster training!
pip: Great! Just so we're clear, you're cool with me building that for the rest of time, right?
Me: ...
pip: Building wheel for flash_attn (setup.py): started
pip: Great! Just so we're clear, you're cool with me building that for the rest of time, right?
Me: ...
pip: Building wheel for flash_attn (setup.py): started
September 21, 2025 at 10:13 PM
Me: I want to use flash attention for faster training!
pip: Great! Just so we're clear, you're cool with me building that for the rest of time, right?
Me: ...
pip: Building wheel for flash_attn (setup.py): started
pip: Great! Just so we're clear, you're cool with me building that for the rest of time, right?
Me: ...
pip: Building wheel for flash_attn (setup.py): started
I see your dynamic-loss-systems-that-use-an-average-probability-of-all-tokens-that-come-next-in-different-examples and raise you <thinking> through </thinking> counting as one token when fine tuning.
September 21, 2025 at 7:55 PM
I see your dynamic-loss-systems-that-use-an-average-probability-of-all-tokens-that-come-next-in-different-examples and raise you <thinking> through </thinking> counting as one token when fine tuning.