🧠 Large Language Models

News about foundation models, LLMs, multimodal models, benchmarks, and model releases from OpenAI, Anthropic, Google, Meta, Mistral, and others.

1,682 articles

RAG or fine-tuning: which LLM strategy fits enterprise AI?

An explainer compares retrieval-augmented generation with LLM fine-tuning for enterprise use. It covers differences, benefits, use cases and costs to help teams pick the right approach. Source: hackernoon.com

AI for Science program widens to foundations and institutions

The AI for Science program, which hands out free Claude credits to researchers on high-impact scientific projects, is expanding. It now extends to foundations and institutions. Source: anthropic.com

DiffuSpace open-sources Clover-35B-A3B, a diffusion LM for agents and coding

DiffuSpace has released Clover-35B-A3B-Preview, a 35B-parameter MoE diffusion language model, as open source. The team reports a score of 73.6 on SWE-bench Verified. Source: pandaily.com

Mistral Large 4 arrives on Hugging Face as "Le Chonk"

Mistral Large 4, nicknamed Le Chonk, is a multimodal, multilingual Mixture-of-Experts (MoE) model. It accepts text and images as input and generates text output. Source: huggingface.co

Gemini now lets you edit and preview documents in-app

Gemini can now generate, edit and preview PDFs and Docs directly in the app. A new full-screen editor makes it easy to refine files on the spot. Source: blog.google

Survey charts how narrative theory informs LLM storytelling and comprehension

A survey in Artificial Intelligence Review sorts narrative theory-driven LLM methods into categories. It finds that story generation tasks lag behind understanding tasks. Source: bioengineer.org

LLMs parse public health policy debates in German news

Researchers show that GPT-5 and Gemini can extract and classify policy-framing quotes from German news articles. Their F1-scores come close to 80 percent when measured against a human benchmark. Source: bioengineer.org

Altman admits OpenAI was 'pretty wrong' on AI jobs as Goldman flags job losses

Sam Altman says OpenAI was 'pretty wrong' about AI's impact on jobs. Meanwhile, Goldman Sachs warns that AI is still eliminating around 11,000 jobs per month. Source: finance.yahoo.com

Yudkowsky predicts overconfidence after partial AI progress

A statement from MIRI co-founder Eliezer Yudkowsky, posted by @allTheYud, circulated in AI safety discussions after being retweeted by @repligate. It warns about overconfidence setting in once AI has made only partial progress. Source: digg.com

Altman claims the AI singularity has arrived

OpenAI CEO Sam Altman says AI has crossed a long-sought threshold: the technology can now improve itself. Source: abc7news.com

Altman's 'scariest warning yet' draws attention on YouTube

A video from the channel TheAiGrid, titled 'Sam Altman Just Issued His Scariest Warning Yet', has gathered about 96K views and 1.8K likes within hours. The snippet gives no details of the warning itself. Source: youtube.com

Google launches flagship Gemini 4 Carbon with stronger cybersecurity features

Google has unveiled Gemini 4 Carbon, a frontier LLM that excels at complex programming tasks. The model arrives just two weeks after the previous release. Source: siliconangle.com

NVIDIA releases open-weight Nemotron 3.5 Transcribe speech model

NVIDIA has published Nemotron 3.5 Transcribe, an open-weight, promptable model that converts recorded speech into usable text. It is available on Hugging Face and ModelScope, alongside the Granary v2 dataset. Source: huggingface.co

OG: an AI tool that finds the genuine Solana coin

OG lets users enter a ticker, then scans every Solana coin sharing that name, reads on-chain data, and uses AI to say which one is the real token. The listing shows a $REPLIGATESOL token as an example. Source: ogcheck.fun

Revamped Artificial Analysis index widens gap with open models

Artificial Analysis has overhauled its Intelligence Index, reshuffling the leaderboard. As a result, Kimi K3.1 Pro slipped five places in the rankings. Source: the-decoder.com

Solana Social Explorer token tied to new AI terminal 'Nightjar'

A market page for the Solana Social Explorer (SSE) token mentions that a figure connected to Repligate and other big accounts released a new AI terminal backrooms project called 'Nightjar Terminal'. All fees are reportedly directed to his GitHub. Source: pump.fun

Nvidia's open Nemotron 3.5 Transcribe bundles several speech tasks

Nvidia has released Nemotron 3.5 Transcribe, an open speech model that follows text instructions. It combines transcription, translation, and speaker recognition in a single model. Source: the-decoder.com

Karpathy's ASD-STE100 idea: slash token counts with simplified English

Andrej Karpathy has floated applying ASD-STE100, a controlled simplified-English standard, to cut token usage dramatically. The idea comes as AI-generated content becomes ever more abundant. Source: medium.com

Altman voices 'chronic disappointment' in Gen Z, urges entrepreneurship

OpenAI CEO Sam Altman says AI will upend jobs, but argues that young workers who embrace entrepreneurship could turn the disruption into opportunity. Source: fortune.com

Anthropic's potential $2T IPO and neocloud predictions: Emad Mostaque weighs in

In a podcast episode, the hosts talk with Emad Mostaque about the latest in AI, including Anthropic's fast-growing revenue and a potential $2 trillion IPO. The discussion also touches on Elon Musk's predictions about neoclouds. Source: youtube.com