Zachary Lipton
@zacharylipton.bsky.social
Cofounder & CTO @ Abridge, Raj Reddy Associate Prof of ML @ CMU, occasional writer, relapsing 🎷, creator of d2l.ai & approximatelycorrect.com
Reposted by Zachary Lipton
💡Can we trust synthetic data for statistical inference?
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
October 10, 2025 at 4:12 PM
💡Can we trust synthetic data for statistical inference?
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
We’ll know he’s really made it when Alcatraz gets autocorrected to Alcaraz.
September 15, 2025 at 6:04 AM
We’ll know he’s really made it when Alcatraz gets autocorrected to Alcaraz.
Small idiosyncratic language errors in otherwise clear technical papers are now a plus; assurance that I'm not wasting my time trying to read GPT spam. #neurips2025
July 5, 2025 at 11:56 PM
Small idiosyncratic language errors in otherwise clear technical papers are now a plus; assurance that I'm not wasting my time trying to read GPT spam. #neurips2025
Spent the whole day reading unlearning papers. If any of the methods worked, I might hope to recover from the experience.
June 30, 2025 at 7:33 AM
Spent the whole day reading unlearning papers. If any of the methods worked, I might hope to recover from the experience.
Final 5m of NBA games solved. If you foul:
• opposing team gets 1 FT
• & keeps possession w fresh 24
Next up, Israel/Palestine.
• opposing team gets 1 FT
• & keeps possession w fresh 24
Next up, Israel/Palestine.
May 30, 2025 at 1:42 AM
Final 5m of NBA games solved. If you foul:
• opposing team gets 1 FT
• & keeps possession w fresh 24
Next up, Israel/Palestine.
• opposing team gets 1 FT
• & keeps possession w fresh 24
Next up, Israel/Palestine.
Google miles behind Open/Anth til overnight boom 2.5.
Mosaic models underwrite $1B until boom mothballed.
Deepseek turns world upside down, but only for 5min.
The speed at which the winds turn here is truly wild.
Mosaic models underwrite $1B until boom mothballed.
Deepseek turns world upside down, but only for 5min.
The speed at which the winds turn here is truly wild.
May 10, 2025 at 12:54 AM
Google miles behind Open/Anth til overnight boom 2.5.
Mosaic models underwrite $1B until boom mothballed.
Deepseek turns world upside down, but only for 5min.
The speed at which the winds turn here is truly wild.
Mosaic models underwrite $1B until boom mothballed.
Deepseek turns world upside down, but only for 5min.
The speed at which the winds turn here is truly wild.
What problem were airplane designers hoping to solve when they proposed this monstrosity?
May 6, 2025 at 1:51 AM
What problem were airplane designers hoping to solve when they proposed this monstrosity?
50M human hours have been spent adjusting airplane seats into upright position before takeoff, saving a grand total of 0 human lives
May 3, 2025 at 3:34 AM
50M human hours have been spent adjusting airplane seats into upright position before takeoff, saving a grand total of 0 human lives
Unlike OpenAI, ostensibly controlled by a powerful holding entity, here have effectively 0 board power. This saga's shaping up like the Asimov's Spacer novels, but instead of each book exploring a different ratio of humans to robotic servants, it's a different level of corporate oversight.
OpenAI folks getting creative about corporate governance. I've seen this movie before...
www.theinformation.com/articles/thi...
www.theinformation.com/articles/thi...
Thinking Machines Lab CEO Has Unusual Control in Andreessen-Led Deal
It’s common nowadays for founders of hot startups to hold special shares that give them more voting power than most investors. But Mira Murati, a former OpenAI chief technology officer who is now a fo...
www.theinformation.com
April 30, 2025 at 1:20 AM
Unlike OpenAI, ostensibly controlled by a powerful holding entity, here have effectively 0 board power. This saga's shaping up like the Asimov's Spacer novels, but instead of each book exploring a different ratio of humans to robotic servants, it's a different level of corporate oversight.
OpenAI folks getting creative about corporate governance. I've seen this movie before...
www.theinformation.com/articles/thi...
www.theinformation.com/articles/thi...
Thinking Machines Lab CEO Has Unusual Control in Andreessen-Led Deal
It’s common nowadays for founders of hot startups to hold special shares that give them more voting power than most investors. But Mira Murati, a former OpenAI chief technology officer who is now a fo...
www.theinformation.com
April 30, 2025 at 12:24 AM
OpenAI folks getting creative about corporate governance. I've seen this movie before...
www.theinformation.com/articles/thi...
www.theinformation.com/articles/thi...
On my way to Singapore for #iclr2025. Excited to pop my head out of AI product development to get inspired, connect w the next generation of researchers, and also for food. hmu
April 23, 2025 at 7:56 AM
On my way to Singapore for #iclr2025. Excited to pop my head out of AI product development to get inspired, connect w the next generation of researchers, and also for food. hmu
Controversial opinion. It’s impossible to lease an EV in the US that’s even half as good a deal as @Tesla model 3/Y.
March 7, 2025 at 2:41 AM
Controversial opinion. It’s impossible to lease an EV in the US that’s even half as good a deal as @Tesla model 3/Y.
Plugin that automatically clicks thru "reject" to select "only strictly necessary" on the cookie pop-ups. 💵
February 28, 2025 at 12:46 PM
Plugin that automatically clicks thru "reject" to select "only strictly necessary" on the cookie pop-ups. 💵
CTO’s the most slippery title in the world. Depending on the company, you’re either
1. A technical founder who never wanted to manage, now an IC engineer
2. A VP of Engineering
3. Leading all of EPD (+AI)
4. An influencer with no internal responsibilities
1. A technical founder who never wanted to manage, now an IC engineer
2. A VP of Engineering
3. Leading all of EPD (+AI)
4. An influencer with no internal responsibilities
February 25, 2025 at 2:37 PM
CTO’s the most slippery title in the world. Depending on the company, you’re either
1. A technical founder who never wanted to manage, now an IC engineer
2. A VP of Engineering
3. Leading all of EPD (+AI)
4. An influencer with no internal responsibilities
1. A technical founder who never wanted to manage, now an IC engineer
2. A VP of Engineering
3. Leading all of EPD (+AI)
4. An influencer with no internal responsibilities
Just as the jaws of tetragenarianism are closing in, stole this W from the jaws of impending awardless decrepitude.
👑 𝗔𝘄𝗮𝗿𝗱 𝗪𝗶𝗻𝗻𝗲𝗿: Modern Healthcare names Abridge Chief Science and Chief Technology Officer Zack Lipton to its 40 Under 40 list!
Learn more about the most influential leaders in healthcare here: www.modernhealthcare.com/awards/zack-... @zacharylipton.bsky.social
Learn more about the most influential leaders in healthcare here: www.modernhealthcare.com/awards/zack-... @zacharylipton.bsky.social
February 10, 2025 at 2:43 PM
Just as the jaws of tetragenarianism are closing in, stole this W from the jaws of impending awardless decrepitude.
Phase 1 of my research career: "ML" = rebranding of the successful subset of AI.
Phase 2: "AI" = rebranding of the successful subset of ML.
Phase 3: AI & ML truly starting to diverge.
Phase 2: "AI" = rebranding of the successful subset of ML.
Phase 3: AI & ML truly starting to diverge.
February 7, 2025 at 8:24 PM
Phase 1 of my research career: "ML" = rebranding of the successful subset of AI.
Phase 2: "AI" = rebranding of the successful subset of ML.
Phase 3: AI & ML truly starting to diverge.
Phase 2: "AI" = rebranding of the successful subset of ML.
Phase 3: AI & ML truly starting to diverge.
Can't find any person or AI that can explain Luka → Lakers. Only satisfying theory I can conjure: behind-the-scenes feud → "calf strain" masking a relationship strain → trade demand. But according to ChatGPT's research, Luka was blindsided by the trade.
February 3, 2025 at 5:04 AM
Can't find any person or AI that can explain Luka → Lakers. Only satisfying theory I can conjure: behind-the-scenes feud → "calf strain" masking a relationship strain → trade demand. But according to ChatGPT's research, Luka was blindsided by the trade.
Just another boring week in the AI capabilities landscape.
January 27, 2025 at 4:28 PM
Just another boring week in the AI capabilities landscape.
Reposted by Zachary Lipton
How to **efficiently** build personalized language models **without** textual info on user preferences?
Our Personalized-RLHF work:
- light-weight user model
- personalize all *PO alignment algorithms
- strong performance on the largest personalized preference dataset
arxiv.org/abs/2402.05133
Our Personalized-RLHF work:
- light-weight user model
- personalize all *PO alignment algorithms
- strong performance on the largest personalized preference dataset
arxiv.org/abs/2402.05133
December 14, 2024 at 5:02 PM
How to **efficiently** build personalized language models **without** textual info on user preferences?
Our Personalized-RLHF work:
- light-weight user model
- personalize all *PO alignment algorithms
- strong performance on the largest personalized preference dataset
arxiv.org/abs/2402.05133
Our Personalized-RLHF work:
- light-weight user model
- personalize all *PO alignment algorithms
- strong performance on the largest personalized preference dataset
arxiv.org/abs/2402.05133
Reposted by Zachary Lipton
First #runconference @neuripsconf.bsky.social #NeurIPS2024 was great! Will share tomorrow's deets later today, join us!
@zacharylipton.bsky.social @adamjelley.bsky.social @random-steve.bsky.social
@zacharylipton.bsky.social @adamjelley.bsky.social @random-steve.bsky.social
December 10, 2024 at 4:49 PM
First #runconference @neuripsconf.bsky.social #NeurIPS2024 was great! Will share tomorrow's deets later today, join us!
@zacharylipton.bsky.social @adamjelley.bsky.social @random-steve.bsky.social
@zacharylipton.bsky.social @adamjelley.bsky.social @random-steve.bsky.social
caught red-handed
December 9, 2024 at 2:07 PM
caught red-handed
Excited to attend #NeurIPS this year. AI research has never been more impactful and yet its core identity, key initiatives, and trajectory have never been less clear. Looking forward to meeting a new generation of talent, to having wild conversations, & to reconnecting w old friends.
December 9, 2024 at 1:38 AM
Excited to attend #NeurIPS this year. AI research has never been more impactful and yet its core identity, key initiatives, and trajectory have never been less clear. Looking forward to meeting a new generation of talent, to having wild conversations, & to reconnecting w old friends.
Thirded
Seconded
Meta-comment. Nested quote-{dunks, replies} are a navigation nightmare. One arrives at the last reply and must read the thread per reverse chronology, clicking through each tweet in sequence to get to the origin. Would love a navigation tool that jumps to the origin, progressively zooms out.
November 28, 2024 at 11:25 PM
Thirded
Meta-comment. Nested quote-{dunks, replies} are a navigation nightmare. One arrives at the last reply and must read the thread per reverse chronology, clicking through each tweet in sequence to get to the origin. Would love a navigation tool that jumps to the origin, progressively zooms out.
I'm with @arjunraj.bsky.social on this one. One point is that I think it would get even harder to take risks with research & pursue potentially resource heavy projects, if one has very limited resources since u end up having less of a buffer. 1/
I strongly disagree with the idea of caps. Caps are always brought up because "mega-labs are bad!!" Regardless of whether they are good or bad, mega-labs are rarely affected by caps. Instead, it is middle-class labs that are shifted towards subsistence. That is bad for the scientific ecosystem, IMO.
November 28, 2024 at 5:00 PM
Meta-comment. Nested quote-{dunks, replies} are a navigation nightmare. One arrives at the last reply and must read the thread per reverse chronology, clicking through each tweet in sequence to get to the origin. Would love a navigation tool that jumps to the origin, progressively zooms out.