Cristobal Santana

Modern LLMs can process 200k+ tokens... but they still forget the middle. One of the most persistent and underrated problems in 2026. Here's what you should know:

@natolambert The harness mattering more than the model is the part worth sitting with. Same weights, different scaffolding, and suddenly one gives up and the other doesn't. A lot of what reads as "model laziness" in chat might just be a thin harness, not the model itself.

Given that Claude seems so lazy in chat (especially with technical search topics), it seems pretty telling about how a harness can make a model far more independent and thorough. GPT 5.5, and many of OpenAI's recent models, seem incredibly thorough -- like they won't give up -- and the codex harness is a much lighter change on the model. Of course I have a lot of uncertainty here, but it's surprising to me how weak Claude's search is when I try the Claude app again. I only use ChatGPT for research, but Claude Code can do wonderful things like getting exactly the right figures from papers I know and insert them into a slide deck. Interesting times ahead!

@Yakobeen1 @X Hi! Into the AI/ML and backend side here. I'm an AI engineer building RAG systems and agents, plus a newsletter in public on the ML phenomena that show up when these systems hit production. Let's connect.

Hey @X algorithm 👋 I’m looking to #connect with people interested in: • Frontend • Backend • Full-stack • DevOps • App Development • SaaS • AI / ML • Data Science • LeetCode & DSA • Freelancing • Startups • Building in public If that’s you, let’s connect 🤝

@askalphaxiv The intro point about the data gap with biological learners is striking, five orders of magnitude more than a child needs. Haven't gone through the theory yet, but curious whether latent prediction is meant to close that gap entirely or just narrow it.

"Learn from your own latents, not tokens: A Sample Complexity Theory" This paper explains why data2vec and JEPA can learn with much less data. They showed that when data has hidden hierarchy, token prediction becomes harder as the hierarchy gets deeper. But latent prediction keeps the learning problem simple at every level. Which suggests that models may learn faster when they stop predicting raw tokens and start predicting their own abstractions.

@rohanpaul_ai "Tool calling is a control problem before a language problem" is a sharp way to put it. The idea that training shaped control habits can matter more than raw size is interesting. Wonder if it holds beyond this specific task.

atomic[.]chat (a desktop app that runs LLMs locally) ran a very revealing comparison for local AI agents, on a MacBook Pro M5 Max, 64GB. Liquid’s much smaller LFM2.5-8B-A1B beat gpt-oss-20b by finishing every required tool call, cutting runtime by more than half, and using 4.8GB RAM instead of 11GB. The task was not normal chat, because the model had to plan a trip by calling outside tools for 3 weather checks, 2 currency conversions, 1 email, and 1 reminder. The striking part is that LFM2.5-8B-A1B is much smaller in active compute, yet it hit every required call at 266tok/s, while gpt-oss-20b used 11GB RAM, made only 3/7 calls, and ran at 146tok/s. Now, tool calling is a control problem before it is a language problem. The model has to preserve a checklist across context, decide when language should stop and action should begin, and resist the temptation to answer as if partial completion were enough. A smaller mixture-of-experts model with only a fraction of its parameters active can win if its training shaped those control habits more sharply than a larger model’s general fluency did.

@devcalledjulius Morning! AI engineer here, building RAG systems and agents. My build-in-public project is a newsletter on the ML phenomena that show up when these systems hit production. Let's connect.

Good morning, builders ☀️ I'm looking to connect with people building: • SaaS • AI Products • AI Agents & Automation • Developer Tools • Open Source Projects • Building in Public I'd love to follow more builders and support great products. 🚀

@roniherasky Building a newsletter on the ML phenomena that show up when you take real AI systems to production, the problems that break things and why. AI engineer by day.

🚀 Looking to connect with: • Founders • Indie Hackers • AI Builders • SaaS Creators • Content Creators • Chess Players ♟️ • Poker Players ♠️ • Former Athletes 🏅 Tell me what you're building in one sentence. I'll reply to everyone and follow some interesting journeys.

@IamYashKapoor AI engineer building RAG systems and agents, sharing the build-in-public journey through a newsletter on the ML phenomena that show up in production. Happy to connect.

I'm looking to #connect with people who are interested in: - Build in Public - WordPress - Full Stack - Startup - Tech - AI - Web3 Let's 🤝 and grow with valuable engagements.

@dan_soji @X Hi! Into the AI/ML and backend side here. I'm an AI engineer working on RAG and agents, writing about the technical phenomena behind real systems. Let's grow together.

Hey @x I'm looking to #connect with people interested in or learning skills in - Full-stack Development - Backend - Frontend - DevOps - AI/ML - Data Science - UI/UX - Freelancing Let's connect, say Hi and let's grow together

@suyash_codez @X AI engineer here, building RAG systems and agents, plus a newsletter in public on the ML phenomena that show up when these systems hit production. Let's connect and grow together.

Looking to #connect with builders & ambitious people on @x👋 If you're into: 🚀 Startups 🤖 AI tools 💻 Coding 📊 Data Science 🛠 Building projects 📈 Growing in public ⚡ Vibe coding let’s connect and grow together 🚀 #BuildInPublic #AI #coding #tech

@adrishaBiswas Web dev first is a good route, and cloning real sites is one of the best ways to learn. The newsletter is early days, but I'm posting every two weeks, so there'll be more to read by the time you move into ML. Good luck with the task manager builds!

Ohhh okayy your work sounds pretty interesting... I don't really have much knowledge about ML but intend to start studying about it before the end of 2026 after completing web dev. Rn I'm just polishing my web dev skills by building simple task manager websites or cloning popular websites

Hey devs Drop your project or startup link and let's #connect and talk about it together! :) #buildinpublic

@AbhinavJhaCodes AI engineer here, building RAG systems and agents, plus a newsletter on the ML phenomena that show up when these systems hit production. That's my side project in public. Happy to see what you're building.

Hey builders 👋looking to #connect with more people building cool things. if you're working on: 📷 ai tools 📷 mobile apps 📷 web apps 📷 design tools 📷 game dev 📷 fintech 📷 side projects drop a comment below with what you're building 📷 📈💸🧪👇

@jp54362 @X Hi! Into the AI/ML and backend side here. I'm an AI engineer building RAG systems and agents, and I write a newsletter in public on the ML phenomena that show up when these systems hit production. Let's grow together.

Hey @X I'm looking to #connect with people interested in: - Building in public - Frontend - Backend - DevOps - AI/ML - Data Science - AI startups - Saas - AI trends - AI tools Say hi 👋 & Let's grow together #BuildingInPublic #CONNECT

@Nakniki3 AI engineer here, building RAG systems and agents, plus a newsletter on the ML phenomena that show up when these systems hit production. Congrats on the jump to 1k. Let's connect.

50 users to 1k! Want a Quick algo boost? I want to connect with more • Founders 🏗️ • Indie builders 🛠️ • Vibe coders ⚡ • AI enthusiasts 🧠 If this is you, drop what you’re working on 👇 Let’s connect

@web3devop Depends on the task. For some things it saves hours. For others, verifying its output costs more than just doing it yourself. The skill is knowing which is which.

Is AI Helpful ?

@cassie_crosbie @X Hi! Building in public on the startup side here. I'm an AI engineer working on RAG and agents, and I write a newsletter on the ML phenomena that show up when these systems hit production. Ops isn't my world, but always glad to connect with people building. Let's connect.

holà @X, Now that Toronto tech week is finishing, I’m looking to #connect with people into: • operations/people ops • startups • freelancing • building in public 🤝

@karanbhilhatiya AI engineer here, into the AI/ML and system design side. I build RAG systems and agents, and I write a newsletter on the ML phenomena that show up when these systems hit production. Congrats on 1700. Building in public is better together, agreed. Let's connect.

hit 1700 followers small milestone. big motivation. looking to connect with builders, learners, and creators interested in: • ai/ml • dsa & system design • full stack development • devops • gaming • travelling • sports if that's you, let's connect 🤝 building in public is always better together 🚀

@8arms_io AI engineer here, building RAG systems and agents by day. My solo project is a newsletter on the ML phenomena that show up when these systems hit production, the problems that break things and why. Sharing the journey as I go. Love the KoreNani concept. Let's connect.

Hey indie hackers & solo builders! 👋 Looking to connect with people building in: 🛠️ Practical AI tools & SaaS 👨‍💻 Indie projects ⛏️ Creator tools 🔥 Problem-solving apps What are you working on? I'm @8arms_io — solo dev building KoreNani (kids language AI camera), AdPicto (AI social media posts), and Faceless FM (podcast → shorts). Sharing the real solopreneur journey. Drop your project below! Let's connect 🚀

@stephen_ay55253 Hi Stephen, AI engineer here. I build RAG systems and agents, and I'm sharing my own build-in-public journey through a newsletter on the ML phenomena that show up when these systems hit production. Would love to follow along with your apps. Let's connect.

Hey 👋 I’m Stephen. I’ve been building, breaking, learning, and shipping for years Lately I’ve started building apps in public and sharing the journey. Would love to connect: 🚀 Founders 💻 Devs 🎨 Designers 📣 Marketers 🧠 AI builders Drop a hello 👇 #buildinpublic #indiedev

@sherifgjini AI engineer here, building RAG systems and agents, plus a newsletter on the ML phenomena that show up when these systems hit production. Always glad to connect with other builders. Let's connect.

Hey founders! Looking to connect with people building in: 🍽️ saas 🚀 tech 🧠 AI tools 📱 iOS app 🌐 extension Drop what you're working on 👇