June 2026

Evaluating long-term memory limits in stateless LLM chatbots — feedback needed [D]

⋅

Hi all, I’m working on a research project exploring how stateless LLM-based chatbots handle long conversations and whether important earlier information is still reliably retained over time. My idea is to: Run a chatbot using an LLM API without any external memory system Introduce key facts early in a long conversation Continue with many unrelated messages (hundreds of turns) Later test whether the model can still correctly recall those facts at different intervals I’m planning to measure recall […]

Ver mais

Liked Liked
RL for robotics: Where do you spend the most time outside of training?

RL for robotics: Where do you spend the most time outside of training?

⋅

Hey 👋 I’m doing a Cornell Master’s project to understand what actually slows down robotics teams once a policy is trained. Most papers focus on training. In practice, a lot of engineering seems to go into things like: – sim-to-real transfer – policy evaluation – debugging failures on hardware – testing and validation – building simulation environments – reproducing bugs – benchmarking new policies I’m collecting data from researchers and engineers working with RL, behavior cloning, VLA/VLMs, or […]

Ver mais

Liked Liked
Your Automation Followed Instructions. Your AI Agent Makes Decisions.

Your Automation Followed Instructions. Your AI Agent Makes Decisions.

⋅

That Changes Everything Automation was created to do what it was told. It followed rules. Did the same things over and over again until something changed. Then there is Agentic AI. It gets what you want thinks about the good and bad and makes choices. This change is not about computers; it is, about how we work. It is when businesses stop asking, “Can the system do this?”. Start asking, “Can we trust what it decides?” In 2019 I worked with […]

Ver mais

Liked Liked

I shrank a transformer until every number fitted on the screen and made the weights editable [R]

⋅

I’ve been teaching myself how LLMs actually work, not at the API level, but down to the matrix multiplications. To force myself to really understand the forward pass, I first built a complete transformer by hand in a spreadsheet from embeddings through to the loss. Then I turned the forward pass into a web page so it’s easier to share. It’s a full transformer (single attention head, single block) shrunk to the smallest size where every single number […]

Ver mais

Liked Liked
SynthID, marcas de agua y otras herramientas de la paranoIA

SynthID, marcas de agua y otras herramientas de la paranoIA

⋅

Sofía Pérez, de Newtral, me envió un correo para interesarse por mi opinión sobre la tecnología SynthID, creada por Google DeepMind como una forma de generar marcas de agua invisibles en sus contenidos de texto, imagen o de otros tipos, con el fin de poder detectarlas posteriormente y establecer si fueron o no creados mediante inteligencia artificial. Tras contestar a su correo, me puse a juguetear con la herramienta y la verdad es que me gustó lo que […]

Ver mais

Liked Liked

Exile Vector

⋅

:::info Astounding Stories of Super-Science January 1998, by Astounding Stories is part of HackerNoon’s Book Blog Post series. You can jump to any chapter in this book here. THE COUNT OF MONTE CRISTO – VOLUME 1 Astounding Stories of Super-Science January 1998: THE COUNT OF MONTE CRISTO – VOLUME 1 Chapter 9 – The Evening of the Betrothal By: Alexandre Dumas & Auguste Maquet ::: Villefort had, as we have said, hastened back to Madame de Saint-Méran’s in […]

Ver mais

Liked Liked

Best Usage-Based Billing Platforms for AI Companies in 2026: Metronome, Orb, Stripe, & Alternatives

⋅

The best usage-based billing platforms for AI companies in 2026 include Credyt for real-time billing that authorizes and debits a customer’s balance as usage happens, Metronome for high-volume enterprise metering and contracts, Orb for invoice-based billing with custom SQL metrics and pricing simulation, Stripe Billing for subscription-first products that want billing, payments, and tax on one account, Lago for open-source self-hosted billing, and Chargebee for subscription lifecycle management with metered add-ons. These platforms split on one architectural axis: […]

Ver mais

Liked Liked
Spot walking procedural terrain: Isaac base policy + transfer learning, all driven from Python on my own Vulkan renderer

Spot walking procedural terrain: Isaac base policy + transfer learning, all driven from Python on my own Vulkan renderer

⋅

Short clip of Spot crossing procedural terrain. The base policy came from Isaac Lab, everything else is mine. Policy: chain of PPO transfers, not zero-shot. Isaac Lab flat walker → fine-tuned for rough terrain → fine-tuned again for discrete stairs. Keeping the Isaac gait while learning new terrain: I keep a frozen copy of the original Isaac policy as a teacher and add a scan-gated imitation reward. On flat ground the policy is penalized for drifting from the […]

Ver mais

Liked Liked

Load More

↑

Digitado © 2026