Omer Calakos-Mano
@omermano.bsky.social
Data Scientist, former neuroscientist, Jersey City resident
Holy crap, Parrot looks amazing! Congrats on the release!
November 12, 2025 at 7:49 PM
Holy crap, Parrot looks amazing! Congrats on the release!
Yes, I think we've made big improvements in pointer chasing at every problem size. Hard drive sized problems became SSD sized problems became memory sized problems became L3 sized.
August 28, 2025 at 1:40 AM
Yes, I think we've made big improvements in pointer chasing at every problem size. Hard drive sized problems became SSD sized problems became memory sized problems became L3 sized.
Thought it might be useful for others!
Blog post here: woti.substack.com/p/paralleliz...
Podcast episode here: adspthepodcast.com/2025/07/04/E...
Blog post here: woti.substack.com/p/paralleliz...
Podcast episode here: adspthepodcast.com/2025/07/04/E...
Parallelizing Non-Associative Scans
In episode 241 of the excellent ADSP podcast, Bryce (one of the co-hosts) mentioned a technique he had explored with ChatGPT for parallelizing a recurrence relation that, at first glance, seems inhere...
woti.substack.com
August 11, 2025 at 12:30 AM
Thought it might be useful for others!
Blog post here: woti.substack.com/p/paralleliz...
Podcast episode here: adspthepodcast.com/2025/07/04/E...
Blog post here: woti.substack.com/p/paralleliz...
Podcast episode here: adspthepodcast.com/2025/07/04/E...
Financially owning a condo rarely seems to make sense. I sometimes feel that we would be better off not encouraging home ownership. On the other hand if people feel like they want to own a home, we should make that easy to do in urban areas.
August 9, 2025 at 4:48 AM
Financially owning a condo rarely seems to make sense. I sometimes feel that we would be better off not encouraging home ownership. On the other hand if people feel like they want to own a home, we should make that easy to do in urban areas.
I don't want the VA to spend less money, I want it to provide more services. Remember the Jevons paradox- more efficiency leads to more use.
June 6, 2025 at 2:17 PM
I don't want the VA to spend less money, I want it to provide more services. Remember the Jevons paradox- more efficiency leads to more use.
If he wrote the search tool in two days and it searched through 90k contracts and flagged 100 for manual review then that's a great first evaluation of tech. And he gets to tell Trump "we searched through all the contracts, there is no fraud".
June 6, 2025 at 2:12 PM
If he wrote the search tool in two days and it searched through 90k contracts and flagged 100 for manual review then that's a great first evaluation of tech. And he gets to tell Trump "we searched through all the contracts, there is no fraud".
I don't think most people who saw the headline caught this sentence. People came away misinformed and future government workers will fear this kind of headline coming from their experiments.
June 6, 2025 at 12:58 PM
I don't think most people who saw the headline caught this sentence. People came away misinformed and future government workers will fear this kind of headline coming from their experiments.
If the prompt was perfect it would mean they spent too much time developing it instead of trying the search and seeing what comes up.
June 6, 2025 at 12:47 PM
If the prompt was perfect it would mean they spent too much time developing it instead of trying the search and seeing what comes up.
DOGE is an abomination but if we hold government to this standard we will never be able to deliver effective government services. We want government workers to try new technologies to improve efficiencies and to check them with human oversight.
June 6, 2025 at 12:46 PM
DOGE is an abomination but if we hold government to this standard we will never be able to deliver effective government services. We want government workers to try new technologies to improve efficiencies and to check them with human oversight.
"He also stressed that he knew his list of what he called 'MUNCHABLE' contracts would be vetted by others before a final decision was made."
I'm sorry, but this sentence makes the rest of the article irrelevant. It's very important that the government trial new technology and then vet the results.
I'm sorry, but this sentence makes the rest of the article irrelevant. It's very important that the government trial new technology and then vet the results.
June 6, 2025 at 12:42 PM
"He also stressed that he knew his list of what he called 'MUNCHABLE' contracts would be vetted by others before a final decision was made."
I'm sorry, but this sentence makes the rest of the article irrelevant. It's very important that the government trial new technology and then vet the results.
I'm sorry, but this sentence makes the rest of the article irrelevant. It's very important that the government trial new technology and then vet the results.
Salary increases means NJT can hire fewer workers. A lucky (or well-connected) few will get these jobs but most people who want them will be locked out.
May 16, 2025 at 3:53 PM
Salary increases means NJT can hire fewer workers. A lucky (or well-connected) few will get these jobs but most people who want them will be locked out.
According to the sources cited in OP's article, it's ~3Wh/query. So that's like playing League for ~20 seconds on a gaming desktop (I'm assuming 500 watts). Energy is money so these companies are VERY motivated to make queries efficient.
May 15, 2025 at 10:04 PM
According to the sources cited in OP's article, it's ~3Wh/query. So that's like playing League for ~20 seconds on a gaming desktop (I'm assuming 500 watts). Energy is money so these companies are VERY motivated to make queries efficient.
What I mean is that we shouldn't say that this plan isn't feasible because of interference with the LD routes. We should assume we want high speed NEC and then figure out whether we can fit the LD routes in. If they can't, then congress should lift Amtrak's obligations to serve that route.
May 14, 2025 at 9:05 PM
What I mean is that we shouldn't say that this plan isn't feasible because of interference with the LD routes. We should assume we want high speed NEC and then figure out whether we can fit the LD routes in. If they can't, then congress should lift Amtrak's obligations to serve that route.
I think I'm going to anger the train enthusiasts here, but wouldn't it be worth it to trade all the long distance trains in the country for high speed NEC? Northeast Regional + Acela is 14M riders per year vs 4M for all the LD routes.
May 14, 2025 at 8:50 PM
I think I'm going to anger the train enthusiasts here, but wouldn't it be worth it to trade all the long distance trains in the country for high speed NEC? Northeast Regional + Acela is 14M riders per year vs 4M for all the LD routes.
Our choices of food, heating/cooling, living location, and transportation are the ones that have the most climate impact. If we're not focused on those, we're distracted. I completely understand being opposed to LLMs on other grounds, but then that should be the focus of the article.
May 14, 2025 at 1:54 PM
Our choices of food, heating/cooling, living location, and transportation are the ones that have the most climate impact. If we're not focused on those, we're distracted. I completely understand being opposed to LLMs on other grounds, but then that should be the focus of the article.
It's like someone saying they were trying to save money by cutting back on bubblegum purchases instead of cancelling their international trip. It's not whataboutism to focus on the biggest environmental impacts: food, heating/cooling, and transportation.
May 14, 2025 at 1:47 PM
It's like someone saying they were trying to save money by cutting back on bubblegum purchases instead of cancelling their international trip. It's not whataboutism to focus on the biggest environmental impacts: food, heating/cooling, and transportation.
All your facts are correct, but they require context to be useful. Growing one almond takes as much water as 150 ChatGPT queries.
May 14, 2025 at 1:30 PM
All your facts are correct, but they require context to be useful. Growing one almond takes as much water as 150 ChatGPT queries.
It's important for students to be aware of which aspects of their lives have the most environmental impact. open.substack.com/pub/andymasl...
A cheat sheet for why using ChatGPT is not bad for the environment
The numbers clearly show this is a pointless distraction for the climate movement
open.substack.com
May 14, 2025 at 12:40 PM
It's important for students to be aware of which aspects of their lives have the most environmental impact. open.substack.com/pub/andymasl...
I highly recommend this article for perspective on the numbers you are reporting here: open.substack.com/pub/andymasl...
A cheat sheet for why using ChatGPT is not bad for the environment
The numbers clearly show this is a pointless distraction for the climate movement
open.substack.com
May 14, 2025 at 12:36 PM
I highly recommend this article for perspective on the numbers you are reporting here: open.substack.com/pub/andymasl...
Right, the domain of perturbations for which the model can make good predictions will be constrained by the set of perturbations in the training data. So it might be able to predict the best course of treatment amongst existing options for an individual.
May 5, 2025 at 2:52 PM
Right, the domain of perturbations for which the model can make good predictions will be constrained by the set of perturbations in the training data. So it might be able to predict the best course of treatment amongst existing options for an individual.
Experiments are going to remain important and I'm skeptical that we can collect enough data to make good brain models, but LLMs can predict survey answers because they are pretty good models of human writing. Compression probably provides the regularization that makes the models causal.
May 5, 2025 at 1:43 PM
Experiments are going to remain important and I'm skeptical that we can collect enough data to make good brain models, but LLMs can predict survey answers because they are pretty good models of human writing. Compression probably provides the regularization that makes the models causal.
Will you have to present the ticket at both the origin and the destination? Otherwise someone could buy a cheap ticket for a short trip and use it for a long trip right?
April 3, 2025 at 1:35 AM
Will you have to present the ticket at both the origin and the destination? Otherwise someone could buy a cheap ticket for a short trip and use it for a long trip right?
What do we need to do to make this possible? Does Assembly Bill No. 4972 need to pass first?
March 31, 2025 at 12:39 AM
What do we need to do to make this possible? Does Assembly Bill No. 4972 need to pass first?
Great! So just say "according to the Gaza Health Ministry".
March 23, 2025 at 4:14 PM
Great! So just say "according to the Gaza Health Ministry".