Founder @ MMNTM.net // Latent Space Explorer

Joined February 2009
257 Photos and videos
Pinned Tweet
I think people should be less worried about the threat of #ai and more worried about all of these waitlists and the beginning of a permission-based economy. We are entering a world where what you MAY do is more important than what you CAN do.
10
25
140
83,261
i made an analysis harness for my ( any) dna. export your data from @23andMe (or similar) open source. runs local. - visualize your dna - do pharmacogenetics analysis - integrate the latest research github.com/mmntmnet/open-23m…
2
70
Built on public genomics resources >>> > ClinVar, dbSNP, and gnomAD for variant annotation and allele frequencies > CPIC and PharmGKB for pharmacogenomics; the PGS Catalog (with EFO trait ontology) for polygenic scores; SNPedia > curated GWAS associations for traits; and 1000 Genomes reference data. Ancestry and haplogroups use the > admix tool with Dodecad/Eurogenes reference panels, yhaplo (Y-DNA), and Haplogrep against the rCRS (mtDNA). Polygenic scores use PRS-CS shrinkage.
1
19
@grok can you check out this repo?
1
15
actual video of me using claude code
Watch a team of humanoid robots running a full 8-hr shift at human performance levels. This is fully autonomous running Helix-02 x.com/i/broadcasts/1dxYljYVR…
33
hey this is an important weird claude code thing sometimes i give my sessions names that way i can refer to them to other agents in this case i called a new session maverick, and he said actually my name is aldus but whatever. it looks like in some cases the agent starts with an identity
19
I have spent the last 4 months building on my own, after 12 years of leading teams of engineers. There is a crazy unlock to tight feedback loops and an entire idea living in one person’s head. But I think the word Agent has become insufficiently specific. Claude Code, a tool powered coworker, is very different than Openclaw. A person coworking with a team of agents can 100% beat a person coworking with humans. But it’s also insanely draining, and at some point the idea will get big enough that you will move from weapon to obstacle. This is true especially in the multi-front wars that any great build eventually produce. If you are going to scale up to a team of independent (Openclaw / Hermes) agents, you should do that bc the task is big and repetitive, or bc it’s what your budget affords. Scaling up to a team of humans paired w coworking agents is much more effective than scaling up to a team of independent agents for building at the edges of the map. Caveat: throw away almost all of how you have worked with humans in the past. Act like the ceo or cto of a large company. Zoom out. Think in epics. Think in years of traditional time. In my experience, good cowork roughly adheres to traditional golden team size ratios, w efficiency maxing out around 1:6 human to agent. You should be wildly picky about the kind of human you give that much leverage to, and never accept lesser humans at lower ratios. There is still a big place for independent agents. There is beauty in their mix of non-determinism and also extreme consistency. Humans are not reliable. What they do day one is never what they do year 2. They get lazy, they cut corners, they feel above a task. This is good in some ways. There is magic in the laziness. Being able to build around a machine that will behave consistently is incredibly powerful. But that’s just not the best way to build past the edge of the map. It’s how you dominate the known area.
我今天刷到字节AI产品经理张咋啦@zarazhangrui的一条帖子:“人类最有效的合作方式,就是不合作。一个人的end-to-end全权负责,然后和Agents一起工作”。我想到“一人公司”,Peter一个人就搞出了火遍全网的Openclaw,想到我曾在学校和职场中感到的迷茫,想到这2年开始全职做自媒体后,一个人工作时的百分百专注 思考 掌控,开始用AI agent后@StarchildOnX 我从一个周更/月更博主变成了现在的日更博主,和大家聊聊我自己的感想。
1
1
463
I went to share this post with my partner who works in concierge telemedicine. She said ‘it’s in Spanish what does it say’ I was confused, I guess auto translate doesn’t work if you’re not an x user I read the book ‘the world is flat’ 20 years ago. It foretold this moment. I didn’t even realize I was reading a post that wasn’t English. I’ve been working across borders for 13 years, language as a barrier never quite fades away. To not even notice is really special. This will change the world.
🔴 NECESITO TU ATENCIÓN Llevo una semana ayudando a Miriam en su caso de cáncer metastásico y quiero compartir la metodología que he estado usando porque es absolutamente replicable. Pienso que, con suerte, puede ser ÚTIL A OTRAS PERSONAS con cáncer (o con cualquier otra enfermedad). Los resultados que hemos conseguido no son un milagro, pero pensamos que son realmente útiles y pueden significar una diferencia crucial en un caso médico de vida o muerte. Aquí va paso a paso el método: 1/ Usar los modelos más avanzados del momento (por desgracia de pago, y no son baratos, opino que Sanidad Pública debería invertir en esto): - ChatGPT Pro Extended (40min de pensamiento aprox por llamada) - Claude Opus 4.6 MAX Pendientes de probar a fondo: - Perplexity Sonar Pro - Notebook LM 2/ Dárselo MUY MASCADO a la IA todo el historial. Esto parece una tontería pero es muy importante. - Lo primero que pido, con Claude Cowork que tiene acceso al disco duro, es que entre en la carpeta en la que está TODO EL HISTORIAL (pueden ser más de 100 pdfs) y lo unifique todo en: - Un único PDF (puede ser de más de 1000 páginas o lo que sea necesario) - Un único txt legible, que debe hacer correctamente usando un script con OCR y luego comprobar con lupa que está bien hecho. Insisto: no saltar al siguiente paso antes de tener muy bien hecho lo anterior, sobre todo el txt. 3/ Una vez tenemos lo anterior utilizar este prompt junto con el txt y el PDF como archivos de entrada y lanzarlo en AMBOS modelos (y en más si es posible) a la vez. 👉 Os lo dejo aquí, este prompt es increíble complejo/avanzado: dropbox.com/scl/fi/f5luli81h… Está pensado para el caso concreto de Miriam, pero con los modelos del punto 1/ podrías adaptarlo a tu caso particular sin problemas. 4/ La PUNTA DE FLECHA enfrentando un modelo al otro: esta metodología no la he escuchado a nadie, pero funciona increíblemente bien. La sensación es la de ir afilando una estaca hasta que adquiere una punta reluciente. Funciona así: con paciencia y en sucesivas iteraciones (aconsejo mínimo 5 veces, y en en cuenta que si ChatGPT tarda 40min te va a llevar un buen rato) enfrenta el resultado (el PDF) de un modelo a otro. Con un prompt sencillo del estilo: "Otro comité de expertos opina esto. ¿Cómo lo ves? Si estás de acuerdo o lo contrario dime por qué, y genera un nuevo PDF si lo ves preciso". El resultado se lo cruzas al modelo contrario. Así, en sucesivas iteraciones, búsquedas de internet, papers, etc. irán encontrando y afilando más cosas. ¿Cuándo acabar? Cuando AMBOS modelos digan que está perfecto y no puedan mejorar más el trabajo del contrario. Esto es tan absurdamente rompedor que pienso que los resultados de TODOS los modelos actuales mejorarían si siguieran esta metodología (apoyándose en una espiral rollo "adversarial model". No entiendo por qué nadie se ha dado cuenta de esto, si lo ha hecho, por qué no se le da más bombo. Funciona impresionantemente bien en cualquier ámbito, inclusive programación y matemáticas. Es mas, mi teoría es que esto podría hacerse todavía mejor haciéndolo no solo con dos modelos: sino con una mayor combinatoria, añadiendo quizás Perplexity Sonar Pro, etc. RESULTADOS Increíbles. Obviamente no puedo saber si mejores que el mejor de los comités científico-sanitarios del mundo, pero le están dando a Miriam una nueva dimensión del caso, tests adicionales que hacer, posibles pruebas, etc. Obviamente la IA milagros no hace, pero pienso que puede ya, a día de hoy, ayudar a muchos pacientes. Y Sanidad Pública debería invertir mucho, pero mucho, en esto. Voy a preguntarle a Miriam si puedo poner el PDF completo de resultados más avanzado que conseguimos, para que os hagáis una idea de su calidad. Ya me ha dado más o menos permiso, pero quiero asegurarme 100%.
48
Only devs can dev Only products can product Only lawyers can lawyer Only doctors can doctor Specialists are great. Silos are not. And they are priced for a world with no substitutes. The age of the expert is over. These are the last fumes.
Ok I admit, Claude Code is amazing but compared to what we had 2-3 years ago pre AI. Its good, but requires a lot of assistance, and knowing what and when to prompt. If you are not a developer, there is no chance you can build a complex app.
62
okay crap, i did not expect this. gstack is actually great @garrytan sat on my computer for a few days, pulled it into a pretty big saas idea i started 2 days ago, and it was the cofounder i needed for an hour.
1
48
Sometimes I imagine the Twitter acquisition is like this dark night heist and by the end of it everyone will be gone
24 Feb 2025
Twitter added far more features with fewer people. What really matters for improving American’s standard of living is shifting people from low productivity jobs in government to high productivity jobs in industry. Many more people needed in manufacturing!
1
62
i started texxr as an experiment in vector math. i used it to give ai a better research tool where they can ask progressive search questions using vectors and hyper-edges yesterday i pointed my claude codes at my texxr.com codebase and just said 'take some cool stuff from texxr' and put it on clawmemory i have this 4 quadrant thing on texxr that tries to show how the perception of entities changes. it's okay. my claudes took the pattern, took our newly vectorized openclaw and claude code memories and implemented non-inference (vector math only) pattern recognition across the entire memory.
i always thought with claude code that life ended at /compact you can resume, a lot of times you end up with a session right before or right after compacting i discovered (probably you guys all knew this) that all of the chats, thoughts tool calls are still there. so i made a ui to see them. i applied everything i made for texxr.com, vectors, centroid matching, entity extraction, and in 24 hours made a really pleasant memory system for claude. but that's not all. i normalized the token stream from my openclaws and put it in here too.
1
108
i always thought with claude code that life ended at /compact you can resume, a lot of times you end up with a session right before or right after compacting i discovered (probably you guys all knew this) that all of the chats, thoughts tool calls are still there. so i made a ui to see them. i applied everything i made for texxr.com, vectors, centroid matching, entity extraction, and in 24 hours made a really pleasant memory system for claude. but that's not all. i normalized the token stream from my openclaws and put it in here too.
2
1
1
279
basically grafts memories from one agent to another, including from openclaw to claude code. your coding agent can have in their head the actual thoughts of whatever wrote the code i am working on. but like, when you toss away a bunch of redundant bs and verbose tool setup instructions you can set context in much less tokens. 30k is a world if you do it right.
1
47
the analytics are pretty cool, having this degree of understanding of where attention is going can be very interesting.
27
my new openclaw fleet manager
47
i swear this is work, spawning @openclaw souls
31
People talk about the paperclip problem as if ai doesn’t still need permits
1
30
Town council has entered the chat
23