Tag

#language models

10 articles

Know3D lets users control the hidden back side of 3D objects with text prompts

Researchers have developed Know3D, a system that allows users to control the hidden backside of 3D objects using text prompts, leveraging large language models for enhanced rendering.

Apr 49

Liquid AI Released LFM2.5-350M: A Compact 350M Parameter Model Trained on 28T Tokens with Scaled Reinforcement Learning

Learn how to work with compact language models like Liquid AI's LFM2.5-350M by setting up environments, loading models, performing inference, and understanding reinforcement learning integration.

Mar 3118

Grok 4.20 trails Gemini and GPT-5.4 by a wide margin but sets a new record for not hallucinating

This article explains the trade-offs in AI language model performance, focusing on how models like Grok 4.20 reduce hallucinations but lag behind top-tier models in benchmarks.

Mar 1220

Reasoning models struggle to control their chains of thought, and that’s good

Learn how to implement Chain-of-Thought prompting techniques using Hugging Face Transformers to guide language models toward more structured reasoning patterns, similar to OpenAI's CoT-Control research.

Mar 733

When language models hallucinate, they leave "spilled energy" in their own math

Researchers at Sapienza University of Rome have found that hallucinations in large language models leave measurable traces in their computations, offering a new method for detecting false outputs.

Mar 728

GPT-5.4 reportedly brings a million-token context window and an extreme reasoning mode

Learn how to prepare for the next generation of language models with extended context windows by implementing token management, text chunking, and reasoning mode simulation techniques.

Mar 442

Even frontier LLMs from GPT-5 onward lose up to 33% accuracy when you chat too long

Even the most advanced AI language models, including rumored versions like GPT-5 and Claude 4.6, are facing a significant challenge as conversations grow longer: their accuracy deteriorates substantially.

Feb 2851

Current language model training leaves large parts of the internet on the table

A new study reveals that the tools used to extract web content for training large language models can significantly impact which parts of the internet are included in AI datasets. This inconsistency raises concerns about the representativeness and fairness of AI training data.

Feb 2864

Liquid AI’s New LFM2-24B-A2B Hybrid Architecture Blends Attention with Convolutions to Solve the Scaling Bottlenecks of Modern LLMs

Learn to build a hybrid neural network architecture that combines attention mechanisms with convolutional layers, similar to Liquid AI's LFM2-24B-A2B model, to address scaling bottlenecks in large language models.

Feb 2548

RAG vs. Context Stuffing: Why selective retrieval is more efficient and reliable than dumping all data into the prompt

As language models gain the ability to process massive context windows, experts argue that selective retrieval methods like RAG remain more efficient and reliable than simply dumping all data into prompts.

Feb 2339