@StudioLHC

▫️LHC • Monthly magazine on AI & AI-assisted since 2023 17 issues and 2 special issues online. #lhcmag

 France
linktr.ee/lhc
Joined February 2022
  • Tweets1.7K
  • Following2.1K
  • Followers443
403Media
Pinned Tweet
LHC
@StudioLHCOct 21
The Forbidden Toys Catalog returns for a third volume. Imagine a toy that should never have been marketed and send your proposal to LHC. lhc.media/en/media/trends/op…

[Open Call] Forbidden Toys Catalogue • LHC

Vol. 3 of LHC's Forbidden Toys Catalogue: the fake Christmas catalogue turns into a product recall. Generate AI toys that don't exist, vile and flawless, as pure shelf tags (name, fake price, lying...

lhc.media
39
Understanding a DNA sequence and explaining a mutation in the same response: that is Carbon-X's ambition. @huggingface pits it against general-purpose models, with tight results. lhc.media/en/media/ai-watch/…

Carbon-X: Hugging Face trains an AI model to understand DNA and explain mutations • LHC

Hugging Face releases Carbon-X, an open 14B DNA-language model trained on 300 billion tokens to analyze genetic sequences and mutation effects.

lhc.media
4
In pharma R&D, evidence often contradicts itself more than it adds up. @perceptis_ai OS aims precisely to structure this sorting process between researchers and AI agents. lhc.media/en/media/ai-watch/…

PercepticOS organizes the work of researchers and AI agents around scientific evidence. • LHC

PercepticOS pairs researchers and AI agents in a shared research space to analyze hypotheses, cross-examine evidence, and document decisions.

lhc.media
4
Private data, code, and training environments are being exposed for all to see, raising a simple question: who is paying for what, and for what purpose? @datavendor_ai attempts to answer it. lhc.media/en/media/ai-watch/…

DataVendor wants to create a marketplace for private data used by AI labs. • LHC

DataVendor is launching a marketplace to sell private data, code, and training environments to AI labs.

lhc.media
1
8
Real-time systems do not leave much room for wordy models. @liquidai wants to slip RIPL into them, an inference engine tailored for search, payments, and advertising. lhc.media/en/media/ai-watch/…

Liquid AI tackles the last milliseconds of inference • LHC

Liquid AI introduces RIPL, an inference engine for real-time search, commerce, and financial workloads, with selected responses completed in milliseconds.

lhc.media
6
Backed by Verdict Capital, @distcog is betting on multiple models rather than a single giant. The idea: making specialized AIs cooperate, with a result that remains highly theoretical. lhc.media/en/media/ai-watch/…

Distributed cognition wants to build collective Superintelligence from multiple AI models • LHC

Backed by Verdict Capital, Distributed Cognition explores a collective AI based on the cooperation, diversity, and communication of small models.

lhc.media
7
Publishing weights isn't enough: inference remains the blind spot. @MiniMax_AI entrusts M3.1 Flash to @inferact for production and pushes its optimizations to vLLM. lhc.media/en/media/ai-watch/…

MiniMax offloads part of M3.1 Flash inference to Inferact and strengthens its partnership with vLLM...

MiniMax and Inferact are collaborating to leverage M3.1 Flash in production and integrate their optimizations into the vLLM open-source inference engine.

lhc.media
8
When a simulation reacts too late, a beautiful video is no longer enough. @kyutai_labs introduces CADRE, seven open autoencoders to bring interactivity back to 20 fps on a single GPU. lhc.media/en/media/ai-watch/…

CADRE: kyutai reduces the response time of AI-generated virtual worlds • LHC

Kyutai releases CADRE, seven open video autoencoders for building interactive worlds at 20 frames per second on a single GPU.

lhc.media
11
When a PDF mixes tables, charts, and formulas, classic OCR quickly loses its way. @llama_index relies on OCRia-1, rated at 87.4 on ParseBench to put things back in order. lhc.media/en/media/ai-watch/…

LlamaIndex enhances the reading of complex tables and charts with a 4-billion-parameter model • LHC

OCRia-1 improves table, chart, and formula extraction from PDFs. The LlamaIndex model achieves 87.4 on ParseBench.

lhc.media
8
With 352,000 human votes across 14 vector generation models, a benchmark separates pure aesthetics from technical rigor. lhc.media/en/media/ai-watch/… @datapointai

SVG-Bench: 352,000 human votes compare AI models for vector graphics • LHC

Datapoint AI releases SVG-Bench, an open dataset gathering 352,270 human votes to compare 14 AI vector creation models.

lhc.media
10
Canva’s first in-house model: Craft arrives in Dream Lab featuring image generation designed for graphic composition and typography. lhc.media/en/media/ai-watch/… @canva

Canva develops its first image model and breaks into DesignArena’s top 4 • LHC

Canva introduces Craft, its first AI image generation model, capable of planning compositions, rendering text, and creating in 2K.

lhc.media
25
Image generators usually produce static pixels. @ideogram_ai Vector directly generates editable SVG files or converts existing visuals into editable paths. lhc.media/en/media/ai-watch/…

From generated image to true vector file: Ideogram introduces Vector • LHC

Ideogram introduces Vector, its first AI model capable of generating editable SVGs or vectorizing existing images for graphic design.

lhc.media
25
Accelerating code production does not protect against unexpected production outages. @brekenai , a startup, is attempting to solve this problem by simulating months of real-world use and degraded scenarios. lhc.media/en/media/ai-watch/…

Breken Wants to Find Software Bugs Before Users Do by Simulating Real-World Conditions • LHC

Breken simulates production environments to find software bugs before users encounter them, identifying 76 of 235 defects in its ESCAPES benchmark.

lhc.media
25
Load more