Dante R. Pérez Méndez

Artificial Intelligence . Urban Mobility. Software Developer. PhD @UNAM_MX @IIMAS_UNAM @C3UNAM

[Sam Altman @sama]

o3 and o4-mini are super good at coding, so we are releasing a new product, Codex CLI, to make them easier to use.

this is a coding agent that runs on your computer. it is fully open source and available today; we expect it to rapidly improve.

[isabelle 🪐 @isabelleboemeke]

My photo kissing a nuclear waste cask went viral and people had… opinions about it.

There was outrage, with many claiming it was dangerous and that I was hurting my future children (lol).

Here’s why kissing a nuclear waste cask is actually fine🧵

[Naval @naval]

It’s going to take time to transition the world to a decentralized financial system.

Judging by market cap, we’re less than 5% of the way there.

[Bryan Johnson @bryan_johnson]

Health is becoming social status

[Dante R. Pérez Méndez @danteblink]

Y hablaba con las palomas también...

[MentalidadFeroz @MentalidadFeroz]

Todo el mundo está obsesionado con dormir 8 horas.

Pero Nikola Tesla sólo dormía 2 horas por noche y se convirtió en uno de los mayores inventores de todos los tiempos gracias a esta rutina.

Ahora, una investigación muestra por fin por qué funcionaba (y por qué no deberías intentarlo nunca): 🧵

[Dr Singularity @Dr_Singularity]

This is BIG. LEV is nearer. Researchers found that immunoglobulin G (IgG) accumulation is a key driving factor of aging.

Study is the first to map the spatial transcriptome of pan-organ aging in mammals, revealing tissue structural disorder and loss of cellular identity as key aging hallmarks and precisely locating the core regions and microenvironmental characteristics of aging sensitivity.

The Immunoglobulin-associated Senescence Phenotype (IASP) proposed by the study expands the frontiers of aging science and opens new avenues for delaying aging and preventing related diseases

https://t.co/eiTeL0g6aD

[Carlos Gershenson @cgershen]

New paper out!
Language Statistics at Different Spatial, Temporal, and Grammatical Scales
with @petitefer Rogelio Lozano @danteblink @ewancolman @ajmoralesguzman @trovadorboricua @carlospgmat03
https://t.co/JFhtVdDmID @MDPIOpenAccess @Entropy_MDPI

[Nacho Montes de Oca @nachomdeo]

Actualicemos la situación en Kursk, en donde la ofensiva ucraniana sigue progresando mientras se espera a llegada de los refuerzos rusos. Ahora se abrió otro frente en Belgorod. Putin comienza a hacer cuentas. Algo cambió en la guerra y alguien ya perdió. Sale hilo

[Chubby♨️ @kimmonismus]

The YouTube video "Kurzgesagt A.I. ‐ Humanity's Final Invention?" is currently circulating on Twitter / X and on the Internet. It gives a summary of the history of artificial intelligence, including AGI and ASI. Many people are impressed.
However, I have to say that I have rarely seen such an incredibly stupid video. I'll be happy to explain why.
I always try to be polite and friendly. However, there are so many false claims in the video that it annoys me.
First of all, it is typical for a German video, because the founders are German (and I myself am German). The basic tenor of the video is: German Angst. Almost the entire time of the video is spent on what could go wrong with AGI. How we as humans could ultimately be inferior to AI and how animals are controlled (that annoys me about German podcasts: instead of seeing the advantages, the positive aspects of AI, they constantly talk about the worries and dangers. I hate German podcasts about AI)
It takes on the assumption that AI might have an intrinsic desire to destroy and enslave for whatever reason. I hear this fear a lot and I don't understand it. No AI created Auschwitz, no AI created racism, no AI developed colonialism. That was man himself. So what is being done here is to project man's fear of man-made disaster onto a technology (a psychic meachsnism, which is a psychological defense https://t.co/z313aLlw8i ). A childish endeavor. First of all, an AI has no intrinsic motivation for power. That is a human assumption. AI is trained with the data we have. This consists of philosophy about altruism and charity as well as theories of subjugation and abuse of power. AI is trained with balanced material.
So if the AI becomes as smart as we always assume, the ASI the most intelligent entity that ever existed, why would it want to enslave humans? Why? It will be so powerful that it will have no interest in such banal, trivial and base urges. (people watch too much Matrix, so they think humans need to be exploited like a battery by AI)
The video also warns that whoever controls AI could also control the entire economy. Let's be honest: to a large extent this is reality already! Billions of people are starving in the world, there is no need for an AGI through which someone controls the global economy. People are already starving!
In addition, many basic anthropological assumptions are wrong and I recommend that the authors read Yuval Harari (Sapiens: A Brief History of Humankind: (Patterns of Life)) again.

Rather, we should be rational optimists. As a materialist Hegelian, I assume that the world develops in dialectical cycles from the lower to the higher.The world has always changed and humans have always adapted to new circumstances. We have come into symbiosis with the world through change and the same will happen with AI / AGI / ASI. Maybe we will merge, maybe something else will happen. But the German fear that an AI will enslave or kill us like in Terminator is essentially unfounded and childish.
Rather read Kurzweil, Zizek and Harari than watch such videos.
https://t.co/z3AQ7zGm2Y

[. @mc_10v1n]

La probabilidad de que los porcentajes de la eleccion venezolana sean ciertos: 0,00000000988

[Dante R. Pérez Méndez @danteblink]

Imágenes que puedes oler 🤢

[Science girl @sciencegirl]

Cinema snacks in China

[Danny Postma @dannypostma]

If you don’t sniff creatine, eat steak from an airfryer and lift weights 8 times a week you’re not gonna make it as indie hacker

[Dante R. Pérez Méndez @danteblink]

@earcos Ni caso Eduardo. Como casi con cualquier tema, las redes sociales están muy polarizadas, u odias o amas los autos eléctricos. Hay gente que basa su identidad en tener un auto a combustión y cualquier mínimo argumento en contra los hace estallar. No te lo tomes personal.

[Dante R. Pérez Méndez @danteblink]

Acá la explicaci��n:
https://t.co/CEIbzN73Cf

[Datadicto📊 @datadicto_es]

🇻🇪Estos son los resultados oficiales de las elecciones de Venezuela ofrecidos por el CNE.

🗳️Sin embargo, si transformamos el número de votos en porcentaje con decimales obtenemos lo siguiente:

➡️Maduro: 51,20000%
➡️Edmundo: 44,20000%

🧵Suena raro, ¿no? Abro mini hilo

[Dante R. Pérez Méndez @danteblink]

Saber matemáticas básicas podría suponer la diferencia entre salirte con la tuya o que se confirme tu fraude electoral. Qué fuerte.

[Juan Ramón Rallo @juanrallo]

Esto sí que es revelador. El número y la distribución de votos con el que el CNE venezolano proclamó a Maduro vencedor (con el 80% de las mesas escrutadas) te da unos porcentajes de voto redondos (hasta el sexto decimal). ¿Resultado casual o resultado manufacturado?

[Sam Altman @sama]

we think there is room to make search much better than it is today.

we are launching a new prototype called SearchGPT: https://t.co/A28Y03X1So

we will learn from the prototype, make it better, and then integrate the tech into ChatGPT to make it real-time and maximally helpful.

[Carlos Santana @DotCSV]

El resumen para los no techies es que Meta ha hecho la inversión de dinero de entrenar un modelo casi a la par que sus competidores (GPT-4, Claude Sonnet 3.5) y lo ha regalado para que cualquier lo descargue y ejecute.

Jugando a romper el mercado.

[Dante R. Pérez Méndez @danteblink]

@platom___ @RamonNogueras

[Carlos Santana @DotCSV]

🍓 NUEVAS FILTRACIONES de OPENAI

Un artículo de Reuters filtra información de una investigación interna llamado Strawberry, anteriormente conocido como Q* -chann channn!-

📰 El artículo la verdad que pone palabras a mucho de lo que se ha venido especulando en el último año:

→ Hablan de modelos "strawberries" para referirse a IAs con capacidad mejorada de razonamiento.

→ Ya se había filtrado anteriormente sobre Q* su capacidad de resolver problemas complejos de ciencia y matemáticas.

→ La técnica sería de post-entrenamiento. Es decir, una vez entrenas un LLM o LMM, puedes aplicarla para mejorar sus capacidades (como sucede con RLHF)

→ Apuntan a un paper de Stanford del que podría estar basado (Self-Taught Reasoner - STaR). Quizá de ahí la estrella de Q* y también el comienzo de STRawberry.

→ OpenAI busca con esta técnica lograr que las IAs puedan resolver long-horizon tasks. Es decir, tareas que recorrieran de planificación y ejecución de múltiples pasos y decisiones.

→ Investigan el usar estas capacidades de ejecución de tareas complejas conectándolo a un agente que ejecute en un ordenador instrucciones, y realizando trabajo de ingeniero de software y ML.

Wow. 😯