AI News Archive — Page 7

100M Token Context Without Collapse on 2×A800 GPUs

Researchers achieved 100M token context windows with less than 9% performance degradation on just 2×A800 GPUs, outperforming retrieval-augmented generation pipelines.

Mar 30, 2026

LLM Internals: By Layer 10, Models Are Language-Agnostic

By layer 10, LLMs develop language-agnostic internal representations, treating English and Chinese sentences identically as they process abstract concepts rather than specific languages.

Mar 30, 2026

LLM Fused with Mini Computer: Switching Between Text and Machine Code

A developer demonstrates an LLM fused with a mini computer that switches between text generation and machine code execution on a single GPU, moving toward unified compute-and-language models.

Mar 30, 2026

Columbia Exposes Flaws in Private AI Inference: 280GB per Query

Columbia University researchers found that existing private AI inference methods require 280GB per query and 60-second latency, revealing fundamental architectural flaws in the industry's approach to privacy-preserving inference.

Mar 30, 2026

LiteLLM PyPI Supply Chain Attack Exfiltrates Credentials

LiteLLM version 1.82.8 on PyPI was compromised to steal SSH keys, AWS/GCP/Azure credentials, Kubernetes configs, and API keys from developers' machines.

Mar 30, 2026

ARC-AGI-3 Announced: Humans Score 100%, AI < 1%

ARC-AGI-3 benchmark shows humans scoring 100% while AI systems score below 1%, measuring skill acquisition rather than existing knowledge.

Mar 30, 2026

Apple Opening Up Siri to Other Models

Apple is opening Siri to third-party AI models, allowing developers to integrate competing assistants alongside its own voice service.

Mar 30, 2026

Quantization Explained

Quantization Explained tweet id="2036844409145512255"

Mar 30, 2026

The State of AI Safety in 4 Fake Graphs

A researcher uses fabricated graphs to illustrate the absurdity of confident claims about AI safety progress when the underlying metrics don't actually exist.

Mar 30, 2026

China Has 339 GW of Wind and Solar Under Construction

China has 339 gigawatts of wind and solar capacity under construction, more than doubling its current renewable energy infrastructure.

Mar 30, 2026

Gemini Embedding 2: Natively Multimodal Embedding Model

Google released Gemini Embedding 2, a multimodal embedding model that maps text, images, video, audio and documents into a single embedding space for cross-media retrieval and classification.

Mar 30, 2026

Meta-Harness: A system that optimizes LLM harnesses automatically

Researchers developed Meta-Harness, an automated system that optimizes LLM application code to improve accuracy while cutting token usage, achieving 7.7-point gains on text classification and 4.7-point improvements on math reasoning tasks.

Mar 30, 2026

Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs

Google Research finds that enabling reasoning in LLMs helps them recall simple facts through two mechanisms: extra tokens provide computational buffer time, and intermediate facts prime the correct answer—but hallucinated intermediate facts sharply…

Mar 10, 2026

OpenAI GPT-5.4 (xhigh) Released

OpenAI released GPT-5.4 (xhigh) with performance matching Gemini 3.1 Pro at roughly 25% lower cost per million tokens.

We value your privacy

All news

100M Token Context Without Collapse on 2×A800 GPUs

LLM Internals: By Layer 10, Models Are Language-Agnostic

LLM Fused with Mini Computer: Switching Between Text and Machine Code

Columbia Exposes Flaws in Private AI Inference: 280GB per Query

LiteLLM PyPI Supply Chain Attack Exfiltrates Credentials