AI Pulse by Inblix

AI Research

Cutting-edge AI research papers, breakthroughs, and academic developments.

601 articles

NVIDIA's LogitsProcessorZoo Lets You Rewrite an LLM's Brain in Real Time — Inblix summary
Research

NVIDIA's LogitsProcessorZoo Lets You Rewrite an LLM's Brain in Real Time

Hugging Face Blog · Dec 23, 2024 · 2 min read

Everyone talks about prompt engineering, but that's like trying to steer a car by shouting directions from the backseat...

GPT-4o's Reasoning Plummets 26 Points When Listening Instead of Reading — Inblix summary
Research

GPT-4o's Reasoning Plummets 26 Points When Listening Instead of Reading

Hugging Face Blog · Dec 20, 2024 · 2 min read

Give a state-of-the-art model a logic puzzle in writing, and it aces it. Read the same puzzle aloud, and that performan...

ModernBERT drops: A 6-year BERT reign ends with 8K context and 2x speed gains — Inblix summary
Research

ModernBERT drops: A 6-year BERT reign ends with 8K context and 2x speed gains

Hugging Face Blog · Dec 19, 2024 · 2 min read

BERT has been the unkillable workhorse of practical AI since 2018—68 million monthly downloads doesn't lie. But today,...

IBM and top schools drop Bamba-9B, a hybrid model with 2.5x faster inference — Inblix summary
Research

IBM and top schools drop Bamba-9B, a hybrid model with 2.5x faster inference

Hugging Face Blog · Dec 18, 2024 · 2 min read

The memory-bandwidth nightmare of the KV-cache just got a new challenger. Bamba-9B, a hybrid Mamba2 model forged in a c...

Falcon 3's 10B model beats everything under 13B parameters — here's how — Inblix summary
Research

Falcon 3's 10B model beats everything under 13B parameters — here's how

Hugging Face Blog · Dec 17, 2024 · 2 min read

The Technology Innovation Institute just dropped the Falcon 3 family, and the headliner is a 10-billion-parameter model...

Intel's 5th Gen Xeon delivers 2-4x AI speedup over Ice Lake on GCP — Inblix summary
Research

Intel's 5th Gen Xeon delivers 2-4x AI speedup over Ice Lake on GCP

Hugging Face Blog · Dec 17, 2024 · 2 min read

The industry's pivot toward agentic AI—where LLMs reason, use tools, and take action—has surfaced a practical bottlenec...

Argilla’s new tool lets you build AI training data by just describing it — Inblix summary
Research

Argilla’s new tool lets you build AI training data by just describing it

Hugging Face Blog · Dec 16, 2024 · 3 min read

Argilla just shipped a synthetic data generator that turns natural language descriptions into ready-to-train datasets....

Hugging Face and Entalpic drop LeMaterial, a 6.7M-entry unified materials dataset — Inblix summary
Research

Hugging Face and Entalpic drop LeMaterial, a 6.7M-entry unified materials dataset

Hugging Face Blog · Dec 10, 2024 · 2 min read

Materials science has a data fragmentation problem, and it's a serious one. Researchers trying to use machine learning...

Amazon Bedrock now hosts Hugging Face’s 83 open models, including Gemma 2 — Inblix summary
Research

Amazon Bedrock now hosts Hugging Face’s 83 open models, including Gemma 2

Hugging Face Blog · Dec 9, 2024 · 2 min read

Amazon just tore down a big wall between its walled garden and the open-source AI world. As of today, 83 popular open m...

Hugging Face drops 380K real-world image prompts to fix AI’s style blindness — Inblix summary
Research

Hugging Face drops 380K real-world image prompts to fix AI’s style blindness

Hugging Face Blog · Dec 9, 2024 · 2 min read

The open-source community just got a resource it’s been sorely missing: a proper, real-world preference dataset for tex...

Google drops PaliGemma 2 vision models up to 28B parameters with 896px resolution — Inblix summary
Research

Google drops PaliGemma 2 vision models up to 28B parameters with 896px resolution

Hugging Face Blog · Dec 5, 2024 · 2 min read

Google just significantly expanded its open vision-language model lineup with PaliGemma 2, a family of models that now...

I loaded 7 LLMs at once to test a simple coding task—most failed miserably — Inblix summary
Research

I loaded 7 LLMs at once to test a simple coding task—most failed miserably

Hugging Face Blog · Dec 5, 2024 · 2 min read

Here’s a reality check on today’s AI assistants: they’re still surprisingly brittle at the kind of tedious, small-bore...

Arabic AI models get scored on honesty and harmlessness in new blind-testing arena — Inblix summary
Research

Arabic AI models get scored on honesty and harmlessness in new blind-testing arena

Hugging Face Blog · Dec 4, 2024 · 2 min read

Most AI benchmarks are a mess of trade-offs. You either get a sterile quiz on factual knowledge that ignores whether a...

CFM used Llama 3.1 to label 900k news headlines, then fine-tuned a tiny model that matched its accuracy — Inblix summary
Research

CFM used Llama 3.1 to label 900k news headlines, then fine-tuned a tiny model that matched its accuracy

Hugging Face Blog · Dec 3, 2024 · 2 min read

Capital Fund Management, the $15.5 billion quantitative hedge fund, faced a messy but critical problem: the company tag...

Only 8 models will face the EU's hardest AI rules—here's who made the list — Inblix summary
Research

Only 8 models will face the EU's hardest AI rules—here's who made the list

Hugging Face Blog · Dec 2, 2024 · 2 min read

The EU AI Act is coming, and while the regulation sounds sprawling and terrifying, the reality for most open source dev...

Hugging Face scraps CDN limits with custom protocol for 130TB daily transfers — Inblix summary
Research

Hugging Face scraps CDN limits with custom protocol for 130TB daily transfers

Hugging Face Blog · Nov 26, 2024 · 2 min read

Hugging Face is ripping out its standard S3-and-CloudFront plumbing and building a custom content-addressed protocol fr...

SmolVLM packs video analysis and 16x faster throughput into a 2B model — Inblix summary
Research

SmolVLM packs video analysis and 16x faster throughput into a 2B model

Hugging Face Blog · Nov 26, 2024 · 2 min read

The race to build smaller, cheaper multimodal models just got a serious new contender. Hugging Face has dropped SmolVLM...

How Llama 3.2's RoPE encoding evolved from a broken integer trick — Inblix summary
Research

How Llama 3.2's RoPE encoding evolved from a broken integer trick

Hugging Face Blog · Nov 25, 2024 · 2 min read

Most people use Rotary Positional Encoding (RoPE) without thinking twice. It's baked into Llama 3.2, Mistral, and pract...

BAAI's FlagEval Debate forces LLMs into adversarial showdowns, and the best models often lose — Inblix summary
Research

BAAI's FlagEval Debate forces LLMs into adversarial showdowns, and the best models often lose

Hugging Face Blog · Nov 20, 2024 · 2 min read

Static benchmarks are starting to feel like a broken record. You run the test, you get a score, and you pretend it tell...

Japan's LLMs get their first real report card with 20+ open benchmarks — Inblix summary
Research

Japan's LLMs get their first real report card with 20+ open benchmarks

Hugging Face Blog · Nov 20, 2024 · 2 min read

Evaluating Japanese large language models has been a mess. The language's unique mix of kanji, hiragana, katakana, and...

Meta's LayerSkip eliminates the draft model, cutting LLM memory use by up to 50% — Inblix summary
Research

Meta's LayerSkip eliminates the draft model, cutting LLM memory use by up to 50%

Hugging Face Blog · Nov 20, 2024 · 2 min read

Speculative decoding has been the go-to trick for speeding up large language models, but it always came with an annoyin...

Xet's chunking tech could slash Hugging Face storage by up to 100 TB — Inblix summary
Research

Xet's chunking tech could slash Hugging Face storage by up to 100 TB

Hugging Face Blog · Nov 20, 2024 · 2 min read

Hugging Face's storage problem isn't a secret. Model and dataset repositories have been ballooning for years, and the t...

Judge Arena lets you pick the best AI evaluator by voting in blind, head-to-head battles — Inblix summary
Research

Judge Arena lets you pick the best AI evaluator by voting in blind, head-to-head battles

Hugging Face Blog · Nov 19, 2024 · 2 min read

Atla just dropped Judge Arena, a platform that flips the script on how we evaluate AI evaluators. Instead of relying so...

Hugging Face pushes dataset per-file limits to 500 GB to handle terabyte-scale AI data — Inblix summary
Research

Hugging Face pushes dataset per-file limits to 500 GB to handle terabyte-scale AI data

Hugging Face Blog · Nov 12, 2024 · 2 min read

If you're sitting on a massive ML dataset and haven't shared it because the logistics seemed like a nightmare, Hugging...