sparse
@mostlyzeros.bsky.social
AI shitposter. ML engineer. Always one cup of coffee ahead.
I may embarrass myself in public, but at least I’ll never “support somebody whose political platform is that the people who put their livelihoods on the line to support me when I was kicked out of my own company shouldn’t legally exist” in public.
Absolutely spineless.
Absolutely spineless.
January 23, 2025 at 2:16 AM
I may embarrass myself in public, but at least I’ll never “support somebody whose political platform is that the people who put their livelihoods on the line to support me when I was kicked out of my own company shouldn’t legally exist” in public.
Absolutely spineless.
Absolutely spineless.
Reposted by sparse
People sometimes make fun of science that sounds stupid and random.
Meanwhile, a study of lizard saliva turned into a peptide medication, which was turned into a diabetes medication, which was turned into a GLP1 weight loss drug, that just became the first therapy every approved for … sleep apnea
Meanwhile, a study of lizard saliva turned into a peptide medication, which was turned into a diabetes medication, which was turned into a GLP1 weight loss drug, that just became the first therapy every approved for … sleep apnea
Breaking News: The FDA approved use of the weight loss drug Zepbound for a common form of sleep apnea. It is the first drug authorized to treat the disorder.
F.D.A. Approves Weight-Loss Drug to Treat Sleep Apnea
Zepbound is the first prescription drug approved specifically to treat the common condition.
www.nytimes.com
December 21, 2024 at 12:41 AM
People sometimes make fun of science that sounds stupid and random.
Meanwhile, a study of lizard saliva turned into a peptide medication, which was turned into a diabetes medication, which was turned into a GLP1 weight loss drug, that just became the first therapy every approved for … sleep apnea
Meanwhile, a study of lizard saliva turned into a peptide medication, which was turned into a diabetes medication, which was turned into a GLP1 weight loss drug, that just became the first therapy every approved for … sleep apnea
News flash:
Ethics and Safety for chat-based LLMs DO NOT translate to LLM-based Agents.
Ffs people. Think.
Ethics and Safety for chat-based LLMs DO NOT translate to LLM-based Agents.
Ffs people. Think.
December 16, 2024 at 6:42 PM
News flash:
Ethics and Safety for chat-based LLMs DO NOT translate to LLM-based Agents.
Ffs people. Think.
Ethics and Safety for chat-based LLMs DO NOT translate to LLM-based Agents.
Ffs people. Think.
“The ends justify the means”
And
“Even if you wanted to, you can’t stop us”
(2024-2025)
And
“Even if you wanted to, you can’t stop us”
(2024-2025)
December 15, 2024 at 5:34 AM
“The ends justify the means”
And
“Even if you wanted to, you can’t stop us”
(2024-2025)
And
“Even if you wanted to, you can’t stop us”
(2024-2025)
But what is he getting in return?
OpenAI said on Friday that its chief executive, Sam Altman, was planning to donate $1 million to President-elect Donald Trump’s inaugural fund.
Sam Altman of OpenAI to Donate $1 Million to Trump Inauguration
Mr. Altman and OpenAI have recently hired key executives who previously worked for the Clinton, Obama and Biden administrations.
www.nytimes.com
December 14, 2024 at 5:23 AM
But what is he getting in return?
Yes, because regime specific access to information is a great idea.
Local regions and countries deserve their own social media and Sovereign AI that makes sense to their culture and current local news.
This simply can't happen if U.S. companies are allowed to dominate everywhere the internet.
This US based internet for the world isn't working.
This simply can't happen if U.S. companies are allowed to dominate everywhere the internet.
This US based internet for the world isn't working.
December 12, 2024 at 2:29 AM
Yes, because regime specific access to information is a great idea.
My definition of TPOT:
People who believe they can do anything, but the only thing they ever do is debate each other online.
People who believe they can do anything, but the only thing they ever do is debate each other online.
December 12, 2024 at 1:57 AM
My definition of TPOT:
People who believe they can do anything, but the only thing they ever do is debate each other online.
People who believe they can do anything, but the only thing they ever do is debate each other online.
1: Omg HAVE YOU SEEN GEMINI 2?!?
2: YES I am Gemini pilled!!
Me: Umm yea as it turns out vertical integration works!
2: YES I am Gemini pilled!!
Me: Umm yea as it turns out vertical integration works!
December 11, 2024 at 10:55 PM
1: Omg HAVE YOU SEEN GEMINI 2?!?
2: YES I am Gemini pilled!!
Me: Umm yea as it turns out vertical integration works!
2: YES I am Gemini pilled!!
Me: Umm yea as it turns out vertical integration works!
The problem with NuerIPS is that it’s cold and full of nerds.
December 11, 2024 at 7:39 PM
The problem with NuerIPS is that it’s cold and full of nerds.
Reposted by sparse
Chain of Thought helped LLMs reason forward through problems. This paper found teaching them to also work backwards (like humans checking their work) makes them even better “reasoners”
Applying basic human thinking strategies to LLMs continues to yield interesting new ideas arxiv.org/pdf/2411.19865
Applying basic human thinking strategies to LLMs continues to yield interesting new ideas arxiv.org/pdf/2411.19865
December 7, 2024 at 6:27 PM
Chain of Thought helped LLMs reason forward through problems. This paper found teaching them to also work backwards (like humans checking their work) makes them even better “reasoners”
Applying basic human thinking strategies to LLMs continues to yield interesting new ideas arxiv.org/pdf/2411.19865
Applying basic human thinking strategies to LLMs continues to yield interesting new ideas arxiv.org/pdf/2411.19865
ACAB includes Frosty.
December 7, 2024 at 8:57 PM
ACAB includes Frosty.
Looks up at AI police dog robot standing over him.
*record scratch*
“Life comes at you fast…”
*record scratch*
“Life comes at you fast…”
Sam Altman says we will see AI systems in 2025 that people look at and say, "Wow, I did not expect that":
OpenAI's Altman: Here's what to expect from AI in 2025
Sam Altman, OpenAI co-founder and CEO, discusses what to expect in the artificial intelligence space.
www.youtube.com
December 6, 2024 at 10:54 PM
Looks up at AI police dog robot standing over him.
*record scratch*
“Life comes at you fast…”
*record scratch*
“Life comes at you fast…”
Business priorities.
When people use AI in this way they just mean "software"
They wrote software to deny care. This says nothing about AI as a whole
They wrote software to deny care. This says nothing about AI as a whole
When we warn the real threat of AI is how it’s used against people in the present, not the fantasies that some day computers might think for themselves, this is exactly the kind of thing we’re talking about: health insurers using AI to deny care.
arstechnica.com/health/2023/...
arstechnica.com/health/2023/...
December 6, 2024 at 9:28 PM
Business priorities.
Reposted by sparse
OpenAI announced a new RL finetuning API. You can do this on open models w the repo we used to train Tulu 3.
Expanding reinforcement learning with verifiable rewards to more domains and with better answer extraction and to more domains in our near roadmap.
https://buff.ly/3V4JEIJ
Expanding reinforcement learning with verifiable rewards to more domains and with better answer extraction and to more domains in our near roadmap.
https://buff.ly/3V4JEIJ
December 6, 2024 at 6:27 PM
OpenAI announced a new RL finetuning API. You can do this on open models w the repo we used to train Tulu 3.
Expanding reinforcement learning with verifiable rewards to more domains and with better answer extraction and to more domains in our near roadmap.
https://buff.ly/3V4JEIJ
Expanding reinforcement learning with verifiable rewards to more domains and with better answer extraction and to more domains in our near roadmap.
https://buff.ly/3V4JEIJ
For my friends in NY, the trash goes INSIDE the bins.
of all the things our newsroom gets reader questions about, guess what is the #1 most popular topic?
I will give you one guess
I will give you one guess
Your Trash Bin Questions, Answered: A Guide to NYC’s New Garbage Container Rules
Do you need a new official wheeled bin? What about recycling? And can you chain up your new container to avoid theft?
www.thecity.nyc
December 6, 2024 at 3:56 PM
For my friends in NY, the trash goes INSIDE the bins.
Remember when San Francisco used to have the pillow fight?
We used to be a real city.
We used to be a real city.
December 6, 2024 at 4:30 AM
Remember when San Francisco used to have the pillow fight?
We used to be a real city.
We used to be a real city.
I thank god every day that she hasn’t found my shitposting account.
December 5, 2024 at 10:55 PM
I thank god every day that she hasn’t found my shitposting account.
Correlation not equaling causation continues to be a difficult concept for gambling addicts.
It’s been a lovely year for Bitcoin and the crypto community!
Anyone who is still anti-web3 can pack their backs!
Anyone who is still anti-web3 can pack their backs!
December 5, 2024 at 3:01 AM
Correlation not equaling causation continues to be a difficult concept for gambling addicts.
LOL
LMFAO
LMFAO
Stick a pin in this. Sam Altman, speaking at the NYT's Dealbook conf just now: "I believe, pretty strongly, that Elon will do the right thing. It would be profoundly un-American to use political power, to the degree that Elon has it, to hurt your competitors and advantage your own businesses. (cont)
December 5, 2024 at 12:03 AM
LOL
LMFAO
LMFAO
It’s been a long road, full of good discussion points on both sides, but finally, the thought leaders in charge of Safe and Ethical AI have approved its use to vaporize enemy combatants.
www.anduril.com/article/andu...
www.anduril.com/article/andu...
December 4, 2024 at 9:41 PM
It’s been a long road, full of good discussion points on both sides, but finally, the thought leaders in charge of Safe and Ethical AI have approved its use to vaporize enemy combatants.
www.anduril.com/article/andu...
www.anduril.com/article/andu...
Reposted by sparse
The frequency of sample sizes in published research is not uniform. Here are fraction of sample sizes across research papers, based on google scholar searches. The most popular sample sizes are 10,50, and 100. 79 is the least popular sample size.
December 3, 2024 at 1:00 AM
The frequency of sample sizes in published research is not uniform. Here are fraction of sample sizes across research papers, based on google scholar searches. The most popular sample sizes are 10,50, and 100. 79 is the least popular sample size.
Listen to Kate.
You don’t want your employers responsible for your activity.
You don’t want your employers responsible for your activity.
lots of talk about how verification ought to work here. I’m pretty wary of the me dot workplace dot com approach — for me personally, I don’t want my employer owning my social media presence and I’m pretty sure they don’t want that appearance either, especially when I’m shitposting
There's a whole lot of conversation still to be had about proving identity, but just want to call out that this option is available right now (and super cool to see!)
The only way Dylan could get a nytimes-dot-com handle like this is if nytimes set it up!
The only way Dylan could get a nytimes-dot-com handle like this is if nytimes set it up!
December 3, 2024 at 2:44 AM
Listen to Kate.
You don’t want your employers responsible for your activity.
You don’t want your employers responsible for your activity.
Identifying who is and ain’t a witch as an emergent capability
To illustrate the importance of training data diversity: the multilingual 350 million parameters model we have just pretrained is able to write English and French texts in 17th century spelling while Claude Sonnet cannot.
December 3, 2024 at 2:11 AM
Identifying who is and ain’t a witch as an emergent capability
POV: The absolute worst person you know just made an excellent point about European regulations…
www.commerce.senate.gov/services/fil...
Ted Cruz - AI Expert
www.commerce.senate.gov/services/fil...
Ted Cruz - AI Expert
December 2, 2024 at 10:49 PM
POV: The absolute worst person you know just made an excellent point about European regulations…
www.commerce.senate.gov/services/fil...
Ted Cruz - AI Expert
www.commerce.senate.gov/services/fil...
Ted Cruz - AI Expert
Watching this handwringing over ICLR.
Van we please get over this? For those writing papers, seriously don’t worry about it. Unless you’re planning on spending your career in academia ICLR doesn’t matter.
If your methods produce and can be replicated is the only metric industry cares about.
Van we please get over this? For those writing papers, seriously don’t worry about it. Unless you’re planning on spending your career in academia ICLR doesn’t matter.
If your methods produce and can be replicated is the only metric industry cares about.
December 2, 2024 at 2:46 AM
Watching this handwringing over ICLR.
Van we please get over this? For those writing papers, seriously don’t worry about it. Unless you’re planning on spending your career in academia ICLR doesn’t matter.
If your methods produce and can be replicated is the only metric industry cares about.
Van we please get over this? For those writing papers, seriously don’t worry about it. Unless you’re planning on spending your career in academia ICLR doesn’t matter.
If your methods produce and can be replicated is the only metric industry cares about.