最新

PokerBench is my attempt at a new LLM benchmark wherein frontier models play Texas Hold&#x27;em in an arena setting. It also features a simulator to view individual games and observe how the different models reason about poker strategy. Opus&#x2F;Haiku, Gemini Pro&#x2F;Flash, GPT-5.2&#x2F;5 mini, and Grok 4.1 Fast Reasoning have all been included.<p>All code -&gt; <a href="https:&#x2F;&#x2F;github.com&#x2F;JoeAzar&#x2F;pokerbench" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;JoeAzar&#x2F;pokerbench</a>

I forked a PyTorch DeepDream implementation and added video support with temporal consistency. It produces smooth DeepDream videos with minimal flickering, and is highly flexible including many parameters and supports multiple pretrained image classifiers including GoogLeNet. Check out the repo for sample videos! Features:<p>- Optical flow warps previous hallucinations into the current frame<p>- Occlusion masking prevents ghosting and hallucination transfer when objects move<p>- Advanced parameters (layers, octaves, iterations) still work<p>- Works on GPU, CPU, and Apple Silicon

Hi HN<p>I built diffchecker.dev, a simple online diff checker for comparing text&#x2F;code quickly in the browser.<p>I wanted something that is:<p>Fast (no signup, no clutter) Privacy-friendly (diff happens client-side) Useful for everyday developer workflows<p><a href="https:&#x2F;&#x2F;diffchecker.dev" rel="nofollow">https:&#x2F;&#x2F;diffchecker.dev</a><p>I’d really appreciate honest feedback:<p>Do you find the UI intuitive? Is anything confusing or missing?<p>Happy to iterate based on feedback.<p>Thanks in advance.

Hola Hackers!<p>I&#x27;m building Rice (docs.tryrice.com). Think of Rice as a managed state machine for AI agents with long term memory.<p>Rice is a platform that unifies long term memory and short term state management for AI agents. Effectively, Rice solves the context compounding issue in the immediate sense - by using Rice Slate (our state management service), the context consumption was down 60%. This makes the agents more efficient. The state management layer also allows agents to share context without the conventional &quot;message passing&quot; approach meaning you can run parallel AI agents.<p>The memory layer enables the agents to have a broader contextual understand of the data and relationships - personalisation and automation at scale for agents.<p>How we&#x27;re different (https:&#x2F;&#x2F;docs.tryrice.com&#x2F;rice-vs) and working on some cool aspects.<p>The core value prop -<p>1. Auditable Agentic executions out of the box 2. Shared state for AI agents (not using message passing approach) for efficient executions 3. Persistent memory for historical data and more.<p>Currently in beta phase, so looking for beta testers. Appreciate any thoughts and tests.<p>Please enter your email at tryrice.com if you&#x27;d like to get in the beta.