AI News Archive — Page 12

Google Flash 2.0 Image Generation, Gemma 3 27B

Google launched Gemini 2.0 Flash with native image generation on March 12, while releasing Gemma 3 27B, a 27-billion-parameter open-source model that tops small-model benchmarks with 128K token context and multilingual support.

Mar 16, 2025

Baidu ERNIE-4.5 — DeepSeek R1 level at half the price 👸

Baidu released ERNIE-4.5, matching DeepSeek R1's performance at half the cost while claiming benchmark wins over GPT-4o in multimodal tasks and GPT-4.5 in some text benchmarks.

Mar 16, 2025

Manus: DeepResearch + Operator agent from China

Chinese startup Butterfly Effect released Manus, an autonomous AI agent powered by Claude 3.5 and Qwen that performs real-world tasks like building websites and analyzing stocks, outperforming OpenAI's Deep Research on the GAIA benchmark.

Mar 16, 2025

Same.dev — One shot clone any website

Same.dev lets you clone any website in one shot, sparking debate over AI scraping defenses as developers explore hidden HTML prompts to protect their work.

Mar 16, 2025

The Ultra-Scale Playbook: Training LLMs on GPU Clusters 📈

Nanotron's Ultra-Scale Playbook on Hugging Face teaches LLM training on GPU clusters using 5D parallelism, ZeRO optimization, and fast CUDA kernels, backed by 4,000+ scaling experiments.

Mar 16, 2025

Deep Dive Into LLMs — Andrej Karpathy

Andrej Karpathy explains large language models from pre-training through inference in a comprehensive technical overview.

Mar 16, 2025

ML Street Talk — Transformers Need Glasses!

Federico Barbero explains why transformers fail at counting and copying long text, tracing the problem to architectural bottlenecks in how they compress and retrieve information.

Mar 16, 2025

Matt Segall — Is The Universe Conscious?

Philosopher Matthew Segall argues that current materialist models of reality are as limited as Ptolemy's astronomy, and that consciousness may be fundamental to the universe.

Mar 16, 2025

xAI launches Grok 3

xAI launched Grok 3, powered by 100k H100s, claiming o1-level performance in math, science, and coding with a 73.2% MMMU score; the $22/month subscription includes deep research and direct Twitter feed access.

Feb 21, 2025

Microsoft Launches Quantum Chip Majorana 1

Microsoft unveiled Majorana 1, its first quantum processor using topological qubits designed to be more stable and error-resistant than conventional approaches.

Feb 21, 2025

Hyperscaler Capex

Microsoft, Google, Meta, and Amazon spent $528 billion on AI infrastructure in 2024 as leaders predict AGI could arrive within 1-2 years.

Feb 21, 2025

Scaling up test-time compute with latent reasoning

https://x.com/MatthewBerman/status/1890081482104008920

Feb 21, 2025

AI Dinner 7.0 — DeepSeek Workshop

AI Socratic hosts AI Dinner 7.0 on February 20th in Manhattan to discuss DeepSeek R1 research papers and install the model on attendees' laptops.

We value your privacy

All news

Google Flash 2.0 Image Generation, Gemma 3 27B

Baidu ERNIE-4.5 — DeepSeek R1 level at half the price 👸