Reposted by James Neno
It gets worse. Schumer isn’t just going on a book tour—he’s charging for tickets.
Absolute insanity. While Americans are scared and in crisis mode, he’s basically saying, let them eat cake.
@schumer.senate.gov Step Down.
Absolute insanity. While Americans are scared and in crisis mode, he’s basically saying, let them eat cake.
@schumer.senate.gov Step Down.
March 16, 2025 at 12:49 PM
It gets worse. Schumer isn’t just going on a book tour—he’s charging for tickets.
Absolute insanity. While Americans are scared and in crisis mode, he’s basically saying, let them eat cake.
@schumer.senate.gov Step Down.
Absolute insanity. While Americans are scared and in crisis mode, he’s basically saying, let them eat cake.
@schumer.senate.gov Step Down.
Reposted by James Neno
s1: Simple inference-time scaling
This is a simple small-scale replication of inference-time scaling
It was cheap: 16xH100 for 26 minutes (so what, ~$6?)
It replicates inference-time scaling using SFT only (no RL)
Extremely data frugal: 1000 samples
arxiv.org/abs/2501.19393
This is a simple small-scale replication of inference-time scaling
It was cheap: 16xH100 for 26 minutes (so what, ~$6?)
It replicates inference-time scaling using SFT only (no RL)
Extremely data frugal: 1000 samples
arxiv.org/abs/2501.19393
February 3, 2025 at 5:10 PM
s1: Simple inference-time scaling
This is a simple small-scale replication of inference-time scaling
It was cheap: 16xH100 for 26 minutes (so what, ~$6?)
It replicates inference-time scaling using SFT only (no RL)
Extremely data frugal: 1000 samples
arxiv.org/abs/2501.19393
This is a simple small-scale replication of inference-time scaling
It was cheap: 16xH100 for 26 minutes (so what, ~$6?)
It replicates inference-time scaling using SFT only (no RL)
Extremely data frugal: 1000 samples
arxiv.org/abs/2501.19393
Two things David Lynch has given us, and continues to give us, after his passing:
1. The realization that meditation can be a path to peace, happiness, and creativity.
2. An artistic representation of the surreal and violent undercurrents of modern American life.
-quoting Threads' michael.tucker_
1. The realization that meditation can be a path to peace, happiness, and creativity.
2. An artistic representation of the surreal and violent undercurrents of modern American life.
-quoting Threads' michael.tucker_
February 3, 2025 at 8:07 PM
Two things David Lynch has given us, and continues to give us, after his passing:
1. The realization that meditation can be a path to peace, happiness, and creativity.
2. An artistic representation of the surreal and violent undercurrents of modern American life.
-quoting Threads' michael.tucker_
1. The realization that meditation can be a path to peace, happiness, and creativity.
2. An artistic representation of the surreal and violent undercurrents of modern American life.
-quoting Threads' michael.tucker_
Reposted by James Neno
i’ve said this a few times — i think US export controls on China **CAUSED** Chinese labs to catch up in AI
recent Chinese innovations were:
1. unique in that they avoided cost
2. the right way
US academic dialog was dominated by more expensive and complex methods that wouldn’t have worked
recent Chinese innovations were:
1. unique in that they avoided cost
2. the right way
US academic dialog was dominated by more expensive and complex methods that wouldn’t have worked
January 27, 2025 at 11:41 AM
i’ve said this a few times — i think US export controls on China **CAUSED** Chinese labs to catch up in AI
recent Chinese innovations were:
1. unique in that they avoided cost
2. the right way
US academic dialog was dominated by more expensive and complex methods that wouldn’t have worked
recent Chinese innovations were:
1. unique in that they avoided cost
2. the right way
US academic dialog was dominated by more expensive and complex methods that wouldn’t have worked
Reposted by James Neno
NASDAQ took a hit today, on fears that DeepSeek’s R1 represents flailing US tech industry
January 27, 2025 at 12:59 PM
NASDAQ took a hit today, on fears that DeepSeek’s R1 represents flailing US tech industry
Reposted by James Neno
Stephen Uzzell: Franceska Mann, the Polish ballerina, who, while being led to the gas chamber, stole a Nazi guard’s gun, shot him dead, and started a female-led riot that gave hope to all of the prisoners of Auschwitz in the face of certain death. @lacentrist.bsky.social
January 26, 2025 at 7:38 PM
Stephen Uzzell: Franceska Mann, the Polish ballerina, who, while being led to the gas chamber, stole a Nazi guard’s gun, shot him dead, and started a female-led riot that gave hope to all of the prisoners of Auschwitz in the face of certain death. @lacentrist.bsky.social
Reposted by James Neno
despite the seductiveness of the "distealing" story, we're very quickly building evidence that, no, deepseek did not distill from openai models
even today, mere *days* after the R1 release, there's already reproductions. combined with the incoming huggingface repro, it seems like R1 is legit
even today, mere *days* after the R1 release, there's already reproductions. combined with the incoming huggingface repro, it seems like R1 is legit
an open source 7B replication of R1-zero and R1
notable: they claim they developed in parallel and that most of their experiments were performed *prior to* the release of R1 and they came to the same conclusions
hkust-nlp.notion.site/simplerl-rea...
notable: they claim they developed in parallel and that most of their experiments were performed *prior to* the release of R1 and they came to the same conclusions
hkust-nlp.notion.site/simplerl-rea...
7B Model and 8K Examples: Emerging Reasoning with Reinforcement Learning is Both Effective and Efficient | Notion
A replication of DeepSeek-R1 training on small models with limited data
hkust-nlp.notion.site
January 25, 2025 at 4:46 PM
despite the seductiveness of the "distealing" story, we're very quickly building evidence that, no, deepseek did not distill from openai models
even today, mere *days* after the R1 release, there's already reproductions. combined with the incoming huggingface repro, it seems like R1 is legit
even today, mere *days* after the R1 release, there's already reproductions. combined with the incoming huggingface repro, it seems like R1 is legit
Reposted by James Neno
huggingface is doing a fully open source replication of R1 github.com/huggingface/...
GitHub - huggingface/open-r1: Fully open reproduction of DeepSeek-R1
Fully open reproduction of DeepSeek-R1. Contribute to huggingface/open-r1 development by creating an account on GitHub.
github.com
January 25, 2025 at 2:31 PM
huggingface is doing a fully open source replication of R1 github.com/huggingface/...
Reposted by James Neno
Reposted by James Neno
learning = extrapolating answers from data
emergence = extrapolating questions from the data
emergence is basically “higher order learning”. i like to argue that LLMs aren’t machine learning because they go this extra step
emergence = extrapolating questions from the data
emergence is basically “higher order learning”. i like to argue that LLMs aren’t machine learning because they go this extra step
January 19, 2025 at 5:40 PM
learning = extrapolating answers from data
emergence = extrapolating questions from the data
emergence is basically “higher order learning”. i like to argue that LLMs aren’t machine learning because they go this extra step
emergence = extrapolating questions from the data
emergence is basically “higher order learning”. i like to argue that LLMs aren’t machine learning because they go this extra step
Reposted by James Neno
January 2025 as a meme
January 17, 2025 at 12:56 PM
January 2025 as a meme
Reposted by James Neno
If you’re planning on watching any Lynch, my advice is: don’t watch any analysis videos, don’t think about what stuff is “supposed” to mean, just get your own unique experience from it!!
I think people have become far too concerned with the “correct” interpretations of media. Just experience it!!!
I think people have become far too concerned with the “correct” interpretations of media. Just experience it!!!
January 16, 2025 at 8:37 PM
If you’re planning on watching any Lynch, my advice is: don’t watch any analysis videos, don’t think about what stuff is “supposed” to mean, just get your own unique experience from it!!
I think people have become far too concerned with the “correct” interpretations of media. Just experience it!!!
I think people have become far too concerned with the “correct” interpretations of media. Just experience it!!!
Reposted by James Neno
Sorry to hear about David Lynch. I worked beside him (same studio) while he was making BLUE VELVET. He answered my questions about his work with grace and honesty. I especially enjoyed his deeply existential comic strip, The Angriest Dog in the World.”
January 16, 2025 at 11:02 PM
Sorry to hear about David Lynch. I worked beside him (same studio) while he was making BLUE VELVET. He answered my questions about his work with grace and honesty. I especially enjoyed his deeply existential comic strip, The Angriest Dog in the World.”
Reposted by James Neno
fascinating insight from @natolambert.bsky.social
the raw duration of training, regardless of parallelization, is a huge risk, bc the cluster is saturated doing only one thing
so while you *could* build bigger models by training longer, that’s not feasible in practice due to risk
the raw duration of training, regardless of parallelization, is a huge risk, bc the cluster is saturated doing only one thing
so while you *could* build bigger models by training longer, that’s not feasible in practice due to risk
January 10, 2025 at 10:43 PM
fascinating insight from @natolambert.bsky.social
the raw duration of training, regardless of parallelization, is a huge risk, bc the cluster is saturated doing only one thing
so while you *could* build bigger models by training longer, that’s not feasible in practice due to risk
the raw duration of training, regardless of parallelization, is a huge risk, bc the cluster is saturated doing only one thing
so while you *could* build bigger models by training longer, that’s not feasible in practice due to risk
Reposted by James Neno
WSJ confirms that the LA fires are now the costliest in U.S. history, with economic damages currently estimated to be at least $50 billion.
January 9, 2025 at 5:09 PM
WSJ confirms that the LA fires are now the costliest in U.S. history, with economic damages currently estimated to be at least $50 billion.
Reposted by James Neno
i have a blog post almost ready to go that has a different take — if the nature of software changes such that you only have small short-lived projects with small scale, then there’s little to no need for software engineers and yes it can be mostly automated
Anyone saying that GenAI could replace software engineers don't understand how software is created (and operated.)
Tool innovations have always make the process of building software faster, cheaper: GenAI and AI agents will also do this.
But these are tools and efficiency gains.
Tool innovations have always make the process of building software faster, cheaper: GenAI and AI agents will also do this.
But these are tools and efficiency gains.
December 30, 2024 at 5:25 PM
i have a blog post almost ready to go that has a different take — if the nature of software changes such that you only have small short-lived projects with small scale, then there’s little to no need for software engineers and yes it can be mostly automated
Reposted by James Neno
There is no good music streaming platform but spotify is absolutely the most evil one and also has the distinction of sounding the shittiest
Spotify is gaming its own editorial playlists with tons of music it hires a small number of people to make for very little money and a severely limited license. Spotify doesn't care about artists, quite the opposite.
“What I uncovered was an elaborate internal program”: Is Spotify creating fake artists to flood its editorial playlists?
A new book sheds light on the alarming rise of dubious real 'artists' peppering Spotify's curated playlists
www.musicradar.com
December 21, 2024 at 6:44 AM
There is no good music streaming platform but spotify is absolutely the most evil one and also has the distinction of sounding the shittiest
Reposted by James Neno
we should have a tool like NotebookLM that writes a blog post for an academic paper. Yes, every paper should have a blog post, and if it doesn’t, i’ll just generate it
December 17, 2024 at 1:34 PM
we should have a tool like NotebookLM that writes a blog post for an academic paper. Yes, every paper should have a blog post, and if it doesn’t, i’ll just generate it
Reposted by James Neno
A famous psychiatrist once said that there are more CEO’s and other executives (mostly CEO’s) who display psychopathic & some sociopathic tendencies than in any other areas of business. They are more represented in the executive positions than anywhere else.
December 8, 2024 at 2:25 AM
A famous psychiatrist once said that there are more CEO’s and other executives (mostly CEO’s) who display psychopathic & some sociopathic tendencies than in any other areas of business. They are more represented in the executive positions than anywhere else.
Reposted by James Neno
ICYMI a starter pack to fill your feed with awesome women in AI go.bsky.app/LaGDpqg
November 15, 2024 at 10:39 PM
ICYMI a starter pack to fill your feed with awesome women in AI go.bsky.app/LaGDpqg
Reposted by James Neno
3 years ago Twitter suppressing deadly covid misinformation was viewed as a partisan attack on Republicans that generated months of performative outrage and resulted in congressional hearings. Today the owner of the platform literally took a job in the Trump administration. 🤔
November 13, 2024 at 12:58 AM
3 years ago Twitter suppressing deadly covid misinformation was viewed as a partisan attack on Republicans that generated months of performative outrage and resulted in congressional hearings. Today the owner of the platform literally took a job in the Trump administration. 🤔