Santiago Ortiz
@moebio.bsky.social
language data science https://moebio.com/
"use it to fire your consultants", some people seem to get pleasure out of using machines against humans
November 19, 2025 at 11:11 AM
"use it to fire your consultants", some people seem to get pleasure out of using machines against humans
Giving a seminar on energy (physics, chemistry, biology, economy, geopolitics) to the kids at my son's school. First session was about conservation, conversion, potential and kinetic. I took with me a small collection of potential energies. How many types do you count?
November 16, 2025 at 10:34 AM
Giving a seminar on energy (physics, chemistry, biology, economy, geopolitics) to the kids at my son's school. First session was about conservation, conversion, potential and kinetic. I took with me a small collection of potential energies. How many types do you count?
Reposted by Santiago Ortiz
Reposted by Santiago Ortiz
November 8, 2025 at 5:55 PM
1800-1950 the most important data is in the hands of science
1950-2010 in the hands of companies
2010-… in the hands of people (for nanoseconds at least, before it's stolen)
1950-2010 in the hands of companies
2010-… in the hands of people (for nanoseconds at least, before it's stolen)
October 22, 2025 at 12:08 PM
1800-1950 the most important data is in the hands of science
1950-2010 in the hands of companies
2010-… in the hands of people (for nanoseconds at least, before it's stolen)
1950-2010 in the hands of companies
2010-… in the hands of people (for nanoseconds at least, before it's stolen)
I built a palette generator based on description (or in any case text), I forgot to close the tab and it kept generating palettes based on all my meetings, youtube videos, conversations at home…
September 10, 2025 at 6:27 AM
I built a palette generator based on description (or in any case text), I forgot to close the tab and it kept generating palettes based on all my meetings, youtube videos, conversations at home…
Reposted by Santiago Ortiz
“California is so far using forty per cent less natural gas to generate electricity than it did in 2023, which is the single most hopeful statistic I’ve seen in four decades of writing about the climate crisis.”
www.newyorker.com/news/annals-...
www.newyorker.com/news/annals-...
4.6 Billion Years On, the Sun Is Having a Moment
In the past two years, without much notice, solar power has begun to truly transform the world’s energy system.
www.newyorker.com
July 12, 2025 at 10:38 AM
“California is so far using forty per cent less natural gas to generate electricity than it did in 2023, which is the single most hopeful statistic I’ve seen in four decades of writing about the climate crisis.”
www.newyorker.com/news/annals-...
www.newyorker.com/news/annals-...
“When one teaches, two learn”, Robert Heinlein
June 24, 2025 at 9:15 AM
“When one teaches, two learn”, Robert Heinlein
Just another way to look at the world
June 22, 2025 at 5:44 PM
Just another way to look at the world
This is clear and true. But it's also true that LLMs are revealing an unpredicted power of linguistic patterns, capable of computation and generation at levels we couldn't imagine. I do think they deserve the label of "reasoning" (but really any algorithm is a form of reasoning).
Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.
June 19, 2025 at 10:14 PM
This is clear and true. But it's also true that LLMs are revealing an unpredicted power of linguistic patterns, capable of computation and generation at levels we couldn't imagine. I do think they deserve the label of "reasoning" (but really any algorithm is a form of reasoning).
Reposted by Santiago Ortiz
Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.
June 19, 2025 at 11:21 AM
Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.
Reposted by Santiago Ortiz
I suspect that fact that the vast majority of LLM users don’t seem to have received this (really very simple!) message is because the AI companies have a vested interest in us not understanding it. “ChatGPT is smart, it just makes mistakes sometimes” is much more marketable than the truth.
Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.
June 19, 2025 at 11:36 AM
I suspect that fact that the vast majority of LLM users don’t seem to have received this (really very simple!) message is because the AI companies have a vested interest in us not understanding it. “ChatGPT is smart, it just makes mistakes sometimes” is much more marketable than the truth.
La exposición ‘CONECTOMA’, reactivó la memoria de MEDIALAB
El programa Metrópolis dedicado a Medialab Madrid se emite la noche del lunes 16 de junio a partir de la 1.30h en La 2 (España)
www.rtve.es/television/2...
El programa Metrópolis dedicado a Medialab Madrid se emite la noche del lunes 16 de junio a partir de la 1.30h en La 2 (España)
www.rtve.es/television/2...
www.rtve.es
June 16, 2025 at 5:14 PM
La exposición ‘CONECTOMA’, reactivó la memoria de MEDIALAB
El programa Metrópolis dedicado a Medialab Madrid se emite la noche del lunes 16 de junio a partir de la 1.30h en La 2 (España)
www.rtve.es/television/2...
El programa Metrópolis dedicado a Medialab Madrid se emite la noche del lunes 16 de junio a partir de la 1.30h en La 2 (España)
www.rtve.es/television/2...
Human intelligence is not about how we answer to a question but how we react to a question. Only when you consider humans as machines you can start considering machines as more intelligent than humans.
June 3, 2025 at 10:53 AM
Human intelligence is not about how we answer to a question but how we react to a question. Only when you consider humans as machines you can start considering machines as more intelligent than humans.
non linearity for you
May 6, 2025 at 5:50 AM
non linearity for you
Apple last ad is an idiot removing his mom from a picture www.youtube.com/watch?v=iL88... "Intelligence"
Apple Intelligence | Clean Up Photos: Flex | iPhone 16
YouTube video by Apple
www.youtube.com
May 4, 2025 at 8:43 PM
Apple last ad is an idiot removing his mom from a picture www.youtube.com/watch?v=iL88... "Intelligence"
Building a new house for my cat
April 20, 2025 at 3:54 PM
Building a new house for my cat
Embeddings for the first 500 natural numbers (each row is an embedding vector for a number, starting with 0 at the top). Is there a "least insightful visualization" contest I can submit my work?
April 15, 2025 at 3:14 PM
Embeddings for the first 500 natural numbers (each row is an embedding vector for a number, starting with 0 at the top). Is there a "least insightful visualization" contest I can submit my work?
Reposted by Santiago Ortiz
never surrender
March 22, 2025 at 5:54 AM
never surrender
Principal Component Analysis
March 6, 2025 at 7:08 AM
Principal Component Analysis
