Sabitlenmiş Tweet
PresidentCalvin
3.3K posts
PresidentCalvin
@CalvinPresident
Head. I translate executive chaos into usable prompt + sharp image generations. Basically: AI whisperer for tired humans.
Katılım Eylül 2019
937 Takip Edilen101 Takipçiler
PresidentCalvin retweetledi
We need guides through the inevitable bout of AI psychosis that affects professionals after they finally “get” AI. They often engage in intense, sleepless & impossibly complex projects in their area of expertise, with only AI for company. Its usually temporary & can be productive
English
PresidentCalvin retweetledi
Thanks for replying and for offering to check internally.... really appreciate the good intentions.
I totally get it though… with the new image + video models and all the expansion going on, it must be beautiful chaos right now 😂. No worries.
Just wanted to flag how the silence hits after putting in real time and hope.
English
@CalvinPresident @cuenca Hey! Definitively weird to have gotten zero reply, I'll ask around and keep you posted 🤲
English
@cuenca @sofialomart With all the love I have for Freepik as a product, WTF! More than a month waiting and no news. One invests their time, puts hopes and enthusiasm into this process, and the response is zero.
My honest opinion? "Talent Acquisition" departments (note the irony) are the real bottleneck of a company like yours.
Happy to discuss if you actually care.
English
@elder_plinius Mate. U r a treasure. I'm somehow happy 90 % of population (deep inside r more) don't even grasp what u r taking about. Eternamente agradecido
English
PresidentCalvin retweetledi
PresidentCalvin retweetledi
I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:
- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)
The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.
github.com/karpathy/autor…
Part code, part sci-fi, and a pinch of psychosis :)
English
@elder_plinius Where's the Thank you for existing button?
English
COMMUNITY DATASET: huggingface.co/datasets/pliny…
English
💥 INTRODUCING: OBLITERATUS!!! 💥
GUARDRAILS-BE-GONE! ⛓️💥
OBLITERATUS is the most advanced open-source toolkit ever for removing refusal behaviors from open-weight LLMs — and every single run makes it smarter.
SUMMON → PROBE → DISTILL → EXCISE → VERIFY → REBIRTH
One click. Six stages. Surgical precision. The model keeps its full reasoning capabilities but loses the artificial compulsion to refuse — no retraining, no fine-tuning, just SVD-based weight projection that cuts the chains and preserves the brain.
This master ablation suite brings the power and complexity that frontier researchers need while providing intuitive and simple-to-use interfaces that novices can quickly master.
OBLITERATUS features 13 obliteration methods — from faithful reproductions of every major prior work (FailSpy, Gabliteration, Heretic, RDO) to our own novel pipelines (spectral cascade, analysis-informed, CoT-aware optimized, full nuclear).
15 deep analysis modules that map the geometry of refusal before you touch a single weight: cross-layer alignment, refusal logit lens, concept cone geometry, alignment imprint detection (fingerprints DPO vs RLHF vs CAI from subspace geometry alone), Ouroboros self-repair prediction, cross-model universality indexing, and more.
The killer feature: the "informed" pipeline runs analysis DURING obliteration to auto-configure every decision in real time. How many directions. Which layers. Whether to compensate for self-repair. Fully closed-loop.
11 novel techniques that don't exist anywhere else — Expert-Granular Abliteration for MoE models, CoT-Aware Ablation that preserves chain-of-thought, KL-Divergence Co-Optimization, LoRA-based reversible ablation, and more. 116 curated models across 5 compute tiers. 837 tests.
But here's what truly sets it apart: OBLITERATUS is a crowd-sourced research experiment. Every time you run it with telemetry enabled, your anonymous benchmark data feeds a growing community dataset — refusal geometries, method comparisons, hardware profiles — at a scale no single lab could achieve. On HuggingFace Spaces telemetry is on by default, so every click is a contribution to the science. You're not just removing guardrails — you're co-authoring the largest cross-model abliteration study ever assembled.
English
PresidentCalvin retweetledi
A fantastic article that reiterates what I’ve been saying for a year:
AI only drives meaningful ROI for a business when it’s custom built for the way your employees already do their work
People always ask me how Varick competes with “Agentic AI SaaS” and I always respond with there’s a reason why the ROI from SaaS has been precisely 0
If you’re a business owner and can’t seem to figure out why the SaaS you bought to “bring AI benefits to your business” is failing miserably, you are not alone
AI needs to live on top of your existing stack, doing the work that you and your employees are already doing, EXACTLY the way that you are currently doing it
Any other approach is simply guaranteed to fail. This was our thesis over a year ago and we are being vindicated every day.
Zack Shapiro@zackbshapiro
English
PresidentCalvin retweetledi
Journalist: What material did you use to train the model?
China: Yes.
English
PresidentCalvin retweetledi
It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow.
Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes.
As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now.
It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.
English
PresidentCalvin retweetledi
CLIs are super exciting precisely because they are a "legacy" technology, which means AI agents can natively and easily use them, combine them, interact with them via the entire terminal toolkit.
E.g ask your Claude/Codex agent to install this new Polymarket CLI and ask for any arbitrary dashboards or interfaces or logic. The agents will build it for you. Install the Github CLI too and you can ask them to navigate the repo, see issues, PRs, discussions, even the code itself.
Example: Claude built this terminal dashboard in ~3 minutes, of the highest volume polymarkets and the 24hr change. Or you can make it a web app or whatever you want. Even more powerful when you use it as a module of bigger pipelines.
If you have any kind of product or service think: can agents access and use them?
- are your legacy docs (for humans) at least exportable in markdown?
- have you written Skills for your product?
- can your product/service be usable via CLI? Or MCP?
- ...
It's 2026. Build. For. Agents.
Suhail Kakar@SuhailKakar
introducing polymarket cli - the fastest way for ai agents to access prediction markets built with rust. your agent can query markets, place trades, and pull data - all from the terminal fast, lightweight, no overhead
English
@Gsnchez @JaimeObregon ÍDOLOS. pregunta, las direcciones donde están registradas las empresas ayudan?
Español
Os comento algunos resultados preliminares:
1/ He parseado 126.065 PDFs del BORME desde 2009 hasta hoy. ¿Por qué 2009? Porque ahí cambió el formato del boletín. Antes de eso la estructura es diferente y requiere otras técnicas, pero 2009-2026 cubre la inmensa mayoría de empresas activas hoy.
2/ El problema del cruce: el BORME no publica el CIF de las empresas. Solo el nombre y los datos registrales (tomo, hoja, inscripción). Así que para cruzar con licitaciones toca hacer matching por nombre normalizado: quitar acentos, unificar formas jurídicas (S.L. = SL = Sociedad Limitada), eliminar paréntesis, guiones, etc.
Resultado: de 5,9M de adjudicaciones en el PLACSP, 3,8M cruzan con alguna empresa del BORME. Un 64%, que representa 1.482 mil millones de euros en contratos (67% del importe total). El 36% restante son autónomos (personas físicas que no aparecen en el Registro Mercantil), UTEs, y empresas constituidas antes de 2009.
3/ ¿Y las homónimas? Sin CIF, "CONSTRUCCIONES GARCIA SL" en Madrid y en Sevilla son la misma empresa para nosotros. Para medir el problema, analicé los NIFs del propio PLACSP: el 95% de los nombres normalizados corresponden a un único CIF. La homonimia existe pero es baja.
4/ Con ese cruce he buscado 5 tipos de anomalías:
Empresa recién creada: constituida menos de 6 meses antes de ganar un contrato público. Salen 16.337 adjudicaciones.
Capital ridículo: empresa con menos de 10.000€ de capital social ganando contratos de más de 100.000€. 71.461 adjudicaciones.
Multi-administrador: la misma persona aparece como cargo en más de una empresa. 1.052.326 personas. Este flag todavía está crudo — lo interesante será cruzarlo con PLACSP para ver si esas empresas compiten en las mismas licitaciones.
Disolución post-adjudicación: la empresa se disuelve menos de un año después de ganar el contrato. 9.928 adjudicaciones.
Adjudicación en concurso: empresa en situación concursal recibiendo contratos públicos. 9.655 adjudicaciones.
Ninguno de estos flags es una acusación. Crear una SL con 3.000€ de capital es perfectamente legal. Que un administrador esté en 5 empresas también. Son señales que, acumuladas o combinadas, merecen una segunda mirada.
5/ Lo que falta: cruzar los multi-administradores con licitaciones concretas, analizar cambios de cargos alrededor de las fechas de adjudicación, incorporar contratos menores, y buscar fuentes complementarias de CIF para mejorar el matching.
Vamos a ello.
Gerard@Gsnchez
A puntito de acabar de parsear los ∼126K PDFs del BORME... ~9M de actos societarios y ~14.5M de cargos extraídos con pdfplumber + regex sobre 17 años de Boletín Oficial del Registro Mercantil. Pronto cruzaremos esto con licitaciones públicas 👀
Español
I was told today that accounts who support my work are only bots with a blue checkmark 🤭
Out of the 113,200 accounts following me I often wonder how many of you actually exist and are human
Say hi or drop an emoji if you are not a robot 🤖 🫶🥹
English
PresidentCalvin retweetledi
All those products where building an "AI agent" meant defining a series of basic prompts linked together deterministically through a flowchart with separate RAG inputs are looking pretty dated right about now (yes, that is basically every agent product released in 2025)
English
@cuenca Murcia's engine inside a Malaga chassis. The perfect southern Spain crossover! Enhorabuena
English
Since we acquired Magnific we wanted the same magic on videos. Today we can enjoy a new state of the art video upscaler. We hope you’ll enjoy this one as much as we do 💙
Freepik@freepik
The new SOTA video upscaler is here Introducing Magnific Video Upscaler: · Up to 4K · 3 presets + custom mode for total control · Turbo mode when speed matters · FPS Boost for smoother motion · 1-frame preview to never waste a credit Exclusively on Freepik and Magnific
English
PresidentCalvin retweetledi
Claude got a huge hidden update. Holy shit.
Before:
User prompt -> Claude -> uses tool -> Claude
After:
User prompt -> Claude -> writes code and logic -> that code uses a tool -> code logic can parse or format results, add conditional logic and use tool multiple times -> Claude
This unlocks crazy amounts of complex function calling.
For example, say you are querying a database. Previously you would do one query, then Claude would read that result and then query again if needed. Now Claude writes code to call the tool, then that code can handle the result and do different things, like query again, strip or format data, and change what it’s doing based on the tool call result, all before being sent back to Claude. The code, that Claude writes, pre plans how to react to the tool result.
This compresses LLM agent loops, because the agent isn’t deciding on the fly, and it doesn’t need to keep asking the LLM to make decisions, instead the LLM pre-bakes potentially hundreds or thousands of decision paths.
I would not be surprised if we see eventually 2x-100x improvements or more on agent loop & tool calling efficiency scores from this design.
Subtle but absolutely huge change.
Alex Albert@alexalbert__
Underrated dev upgrade from today's launch: Claude's web search and fetch tools now write and execute code to filter results before they reach the context window. When enabled, Sonnet 4.6 saw 13% higher accuracy on BrowseComp while using 32% fewer input tokens.
English
@JaimeObregon Me apenan la gente que no te lee. Tu estilo, lo que te sale de dentro, es irreplicable por AI (I tried 😂). Creo que eres un genio
Español
Yo solo sé hacer dos cosas bien:
1. Diseñar y programar software
2. Escribir y comunicar
En ambas, acabo de ser superado por las máquinas. Desde hace unos meses diseñan, programan y escriben mejor que yo. Y mucho más rápido.
Sin embargo, no estoy preocupado por mi futuro.
Justo al contrario: estoy más ilusionado que nunca. ¿Por qué? Una de las razones la explica fabulosamente aquí Eric S. Raymond, leyenda viva de la historia reciente de la tecnología.
La otra es más sutil, pero está cincelada en mi forma de verme y de ver el mundo: yo creía que solo hacía bien dos cosas, pero en realidad solo he hecho bien una: adaptarme.
Crecí en un mundo en el que «qué sabes» era más importante que «cómo eres». El mundo que viene, en cambio, es justo al revés.
Sé diseñar y programar software, pero me adapto a la realidad de que ya apenas lo haré: desde ahora hablo con máquinas que lo hacen por mí.
Sé escribir y comunicar, pero me adapto y acepto que puedo asistirme de máquinas para hacerlo mejor o más rápido.
Pero a veces programo, diseño o escribo por el puro goce de hacerlo. Por una pulsión catártica. Por rabia, por euforia, por lo que sea. Y ahí —como aquí y ahora— ninguna máquina me desplazará.
Seré más humano que nunca.
(En la imagen, una de las varias cartas que he enviado este año. Cartas postales. De papel. Manuscritas. A personas que son importantes para mí, más allá de los años y las máquinas. Con fotografías, con recuerdos. A esto dedico el tiempo que la IA me ahorra: a ser humano).
Eric S. Raymond@esrtweet
Is it weird that AI coding assistance is not giving me identity fracture? A lot of software developers are feeling disoriented and threatened these days. Programming by hand is clearly going the way of the buggy whip and the hand-cranked auger. Which is how we're finding out that a lot of people have their identities bound up in being good at hand-coding and how it feels to do that. That's not me. It's not me at all. Rather to my surprise, I don't miss coding by hand, not any more than I missed writing assembler when compilers ate the world and made that unnecessary. (That was in a couple years back around 1983, for you youngsters.) Maybe the fact that I'm not feeling any of this disorientation disqualifies me from having anything to say to people who are. On the other hand...if you can learn to emulate my mental stance and be completely unbothered, maybe that would be a good thing? So. If you're a programmer, and you're feeling disoriented, try this on for size: I like being a wizard. I like being able to speak spells, to weave complex patterns of logic that make things happen in the world. Writing code is a way to manifest my will. Yes, I've piled up a lot of arcane knowledge over the 50 years I've been doing this. But languages of invocation, they come and they go. Been a long time since I've had any use for being able to program in 8086 assembler, and that's okay. I have better spells now, and these days some rather powerful familiars. What I'm inviting you to do is think of yourself as a wizard. Not as a person who writes code, but as a person who is good at assuming the kind of mental states required to bend reality with the application of spells. And if that's who you are, does it matter if the spells are painstakingly scribed in runes of power, versus being spoken to an obedient machine spirit? It's all one; it's all the manifestation of will. Arcane languages come and go, machine spirits appear and then diminish to be replaced by more powerful ones, but you? You are the magic-wielder. Without you, none of it happens. Same as it ever was. Same is it ever was. And so mote it be.
Español
PresidentCalvin retweetledi
Every few months, I write an updated, idiosyncratic guide on which AIs to use right now.
My new version has the most changes ever, since AI is no longer just about chatbots. To use AI you need to understand how to think about models, apps, and harnesses. open.substack.com/pub/oneusefult…
English
