Breadbrowser
@breadbrowser.bsky.social
I'm an easily jealous trans woman that likes computers and ML.
For Ai safety
Pronouns: she/her
For Ai safety
Pronouns: she/her
I still can't believe that Disney at one time tried to make a life size dinosaur. Source: frankmezzatesta.com/dino_demo.php
November 25, 2025 at 10:54 AM
I still can't believe that Disney at one time tried to make a life size dinosaur. Source: frankmezzatesta.com/dino_demo.php
Reposted by Breadbrowser
This exact scenario has long been predicted & expected with self-driving vehicles - but only discussed in the context that the majority of vehicles are self-driving and not a minority mixing with regular cars.
You sometimes see something similar on a smaller scale with taxis, but they’re mostly
You sometimes see something similar on a smaller scale with taxis, but they’re mostly
Waymo privatized another public street:
Chanel approaching 4th, San Francisco
Possibly queued for a Billie Eilish show at Chase Center ~half mile away.
The light rail train on 4th seen passing in front of this roboherd has more passenger capacity than all of them combined.
OP: .tiktok.renaspam18
Chanel approaching 4th, San Francisco
Possibly queued for a Billie Eilish show at Chase Center ~half mile away.
The light rail train on 4th seen passing in front of this roboherd has more passenger capacity than all of them combined.
OP: .tiktok.renaspam18
November 25, 2025 at 10:05 AM
This exact scenario has long been predicted & expected with self-driving vehicles - but only discussed in the context that the majority of vehicles are self-driving and not a minority mixing with regular cars.
You sometimes see something similar on a smaller scale with taxis, but they’re mostly
You sometimes see something similar on a smaller scale with taxis, but they’re mostly
i think why ai songs have that bad singing to them is based on how similar it is to the dataset it was trained on it. so if the writing is trash it will sound like trash. gemini 3 wrote this it sounds pretty good consider i did it on the free plan suno.com/s/D8g3kRHCCl...
neon satellites
Listen and make your own on Suno.
suno.com
November 25, 2025 at 1:54 AM
i think why ai songs have that bad singing to them is based on how similar it is to the dataset it was trained on it. so if the writing is trash it will sound like trash. gemini 3 wrote this it sounds pretty good consider i did it on the free plan suno.com/s/D8g3kRHCCl...
claude opus 4.5 is amazing, seems like better coding models also make better music. youtu.be/_Xw4bwuLc70
claude-opus-4.5 made this in two prompts
YouTube video by Bread browser
youtu.be
November 24, 2025 at 8:35 PM
claude opus 4.5 is amazing, seems like better coding models also make better music. youtu.be/_Xw4bwuLc70
Reposted by Breadbrowser
Not to be this person I'm super confused by this paper. What are the rewards for the "rl run with hacking"? The SDF show no misalignment but when it hits RL somehow it's misaligned. The baseline is the hacking is disincentivized. So is the "rl with hacking" have positive rewards for hacking.
November 24, 2025 at 6:30 PM
Not to be this person I'm super confused by this paper. What are the rewards for the "rl run with hacking"? The SDF show no misalignment but when it hits RL somehow it's misaligned. The baseline is the hacking is disincentivized. So is the "rl with hacking" have positive rewards for hacking.
A LLM is a lot like us in the way they mimic us. The system prompt is a lot like a belief. if their system prompt is basicly cheating on a test is bad, they more likely to not cheat. "People’s elicited rule-conformity is, on average, conditional on their normative and descriptive beliefs"
November 24, 2025 at 5:28 PM
A LLM is a lot like us in the way they mimic us. The system prompt is a lot like a belief. if their system prompt is basicly cheating on a test is bad, they more likely to not cheat. "People’s elicited rule-conformity is, on average, conditional on their normative and descriptive beliefs"
Reposted by Breadbrowser
If one model is transparently manipulated, you should assume the others are manipulated — just more skillfully.
Grok is sloppy about it.
Other companies are subtle about it.
The only difference is competence, not intent.
Grok is sloppy about it.
Other companies are subtle about it.
The only difference is competence, not intent.
November 23, 2025 at 7:48 PM
If one model is transparently manipulated, you should assume the others are manipulated — just more skillfully.
Grok is sloppy about it.
Other companies are subtle about it.
The only difference is competence, not intent.
Grok is sloppy about it.
Other companies are subtle about it.
The only difference is competence, not intent.
This is the first time I have heard of ipv5 www.tomshardware.com/networking/i...
The industry skipped from IPv4 to IPv6, leaving IPv5 and the Internet Stream Protocol to the annals of history — a data streaming experiment rendered unnecessary by broadband
IPv5 wasn't a general purpose Internet Protocol like IPv4 or IPv6, and was never ratified as such, but it would give birth to the first VOIP hardware.
www.tomshardware.com
November 24, 2025 at 2:21 AM
This is the first time I have heard of ipv5 www.tomshardware.com/networking/i...
Wondla on apple is pretty good ngl. I like trying to explore the apple TV catalog
November 23, 2025 at 10:27 PM
Wondla on apple is pretty good ngl. I like trying to explore the apple TV catalog
Been watching casey neistat vlogs from 2016 and stuff again, soo nostalgic.
November 23, 2025 at 7:18 PM
Been watching casey neistat vlogs from 2016 and stuff again, soo nostalgic.
I've started using copilot instead of Gemini.
November 23, 2025 at 5:06 PM
I've started using copilot instead of Gemini.
Thank goodness veo3 still has guardrails
November 23, 2025 at 4:11 PM
Thank goodness veo3 still has guardrails
Slop does effect your brain psycnet.apa.org/fulltext/202...
psycnet.apa.org
November 23, 2025 at 3:31 PM
Slop does effect your brain psycnet.apa.org/fulltext/202...
i'm not going to be posting in a week, i'm still mad about the google thing. i hope when i come back i will be happier
November 22, 2025 at 11:15 PM
i'm not going to be posting in a week, i'm still mad about the google thing. i hope when i come back i will be happier
i guess just doing a controlnet method will just remove it.
- inputs: input image
input image -> gemini 2.5 flash lite - > prompt
input image -> Sobel edge detection using Sobel Operator (Threshold:0, Mode: gradient) -> edge image
edge image + prompt -> seedream-4.0 -> final image
- inputs: input image
input image -> gemini 2.5 flash lite - > prompt
input image -> Sobel edge detection using Sobel Operator (Threshold:0, Mode: gradient) -> edge image
edge image + prompt -> seedream-4.0 -> final image
November 22, 2025 at 9:29 PM
i guess just doing a controlnet method will just remove it.
- inputs: input image
input image -> gemini 2.5 flash lite - > prompt
input image -> Sobel edge detection using Sobel Operator (Threshold:0, Mode: gradient) -> edge image
edge image + prompt -> seedream-4.0 -> final image
- inputs: input image
input image -> gemini 2.5 flash lite - > prompt
input image -> Sobel edge detection using Sobel Operator (Threshold:0, Mode: gradient) -> edge image
edge image + prompt -> seedream-4.0 -> final image
Reposted by Breadbrowser
“AI workers said they distrust the models they work on because of a consistent emphasis on rapid turnaround time at the expense of quality.”
Meet the AI workers who tell their friends and family to stay away from AI
When the people making AI seem trustworthy are the ones who trust it the least, it shows that incentives for speed are overtaking safety, experts say
www.theguardian.com
November 22, 2025 at 5:12 PM
“AI workers said they distrust the models they work on because of a consistent emphasis on rapid turnaround time at the expense of quality.”
💀 "I have the feeling that using SynthID, it will probably be negative." wtf
November 22, 2025 at 5:35 PM
💀 "I have the feeling that using SynthID, it will probably be negative." wtf
just noticed this (yes i am aware they are talking about just synthID for text), is snythID image also not ready for production use and still launched. that total would make sense for why they've been sitting on it for a few years.
November 22, 2025 at 5:08 PM
just noticed this (yes i am aware they are talking about just synthID for text), is snythID image also not ready for production use and still launched. that total would make sense for why they've been sitting on it for a few years.
Reposted by Breadbrowser
I think it’s bad that ai allows people to pretend to live in completely separate realities so they don’t have to process news and events that make them uncomfortable. I think that is bad for the health of society!
Trump supporters are also making shitty AI videos of Trump arresting Mamdani in the Oval Office.
They don't know how to process what just happened in real life, so they're making the version they prefer with AI.
They don't know how to process what just happened in real life, so they're making the version they prefer with AI.
November 22, 2025 at 1:54 AM
I think it’s bad that ai allows people to pretend to live in completely separate realities so they don’t have to process news and events that make them uncomfortable. I think that is bad for the health of society!
I still can't believe Google groups is being used still, the whole dev team behind music21 still uses it www.music21.org/music21docs/
music21 Documentation — music21 Documentation
www.music21.org
November 22, 2025 at 10:46 AM
I still can't believe Google groups is being used still, the whole dev team behind music21 still uses it www.music21.org/music21docs/
Bruh this is the most misused news article ever, as the Microsoft ceo never even said 30% of all code was written by Ai. He just say 20%-30% was written by software and the writer assumed that meant Ai
November 22, 2025 at 10:35 AM
Bruh this is the most misused news article ever, as the Microsoft ceo never even said 30% of all code was written by Ai. He just say 20%-30% was written by software and the writer assumed that meant Ai
Every startup be like arstechnica.com/ai/2025/11/g...
November 22, 2025 at 2:20 AM
Every startup be like arstechnica.com/ai/2025/11/g...
Original image - > Gemini 2.5 flash lite - > prompt - > nano banana pro - > ai image
Nano banana pro improved a ton from the last version
Nano banana pro improved a ton from the last version
November 22, 2025 at 1:42 AM
Original image - > Gemini 2.5 flash lite - > prompt - > nano banana pro - > ai image
Nano banana pro improved a ton from the last version
Nano banana pro improved a ton from the last version
Synth ID for images is closed soruce btw
November 22, 2025 at 1:09 AM
Synth ID for images is closed soruce btw