Stories

425 stories · updated Oct 28

Ongoing AI stories, each with every update in order and who confirmed it.

Product1 update · last updated Oct 23

AQL-Lite

On Oct 23, arXiv paper proposes AQL-Lite for data-centric Arabic LLM knowledge acquisition.

LatestOfficialarXiv paper proposes AQL-Lite for data-centric Arabic LLM knowledge acquisitionOct 23

Product1 update · last updated Oct 23

MoEZip

On Oct 23, arXiv paper presents MoEZip routing-aware KV cache compression for MoE LLMs.

LatestOfficialarXiv paper presents MoEZip routing-aware KV cache compression for MoE LLMsOct 23

Product1 update · last updated Oct 23

SWE-VERIFIER

On Oct 23, arXiv paper introduced SWE-VERIFIER selective verification for coding agents.

LatestOfficialarXiv paper introduces SWE-VERIFIER selective verification for coding agentsOct 23

Product1 update · last updated Oct 23

GAV

On Oct 23, arXiv paper presents GAV, a scientific experimentation environment for PDE agents.

LatestOfficialarXiv paper presents GAV, a scientific experimentation environment for PDE agentsOct 23

Product1 update · last updated Oct 23

As of Oct 23, TinyFish published two engineering posts on agentic search: one defines a 7-phase production architecture with evidence contracts and bounded retries, and the other explains why its deep research system dropped a separate planner and per-query subagents, per MarkTechPost.

LatestReported by MarkTechPostTinyFish details a 7-phase agentic search architecture and drops a separate planner, per MarkTechPostOct 23

Model5 updates · last updated Oct 23

Ling

On Oct 23, InclusionAI released Ling-3.0-flash-Sante, a healthcare MoE with 124B total and 5.1B active parameters.

LatestOfficialInclusionAI releases Ling-3.0-flash-Sante, a healthcare MoE with 124B total and 5.1B active parametersOct 23

Product2 updates · last updated Oct 22

TorchNeuron

On Oct 22, Hugging Face published a TorchNeuron lookup for which models run on AWS Trainium.

LatestOfficialHugging Face publishes a TorchNeuron lookup for which models run on AWS TrainiumOct 22

Product2 updates · last updated Oct 22

Atria Dawn

On Oct 22, InternLM published Atria-Dawn, a GlmMoeDsa open-weight model on Hugging Face.

LatestOfficialInternLM publishes Atria-Dawn, a GlmMoeDsa open-weight model on Hugging FaceOct 22

Product4 updates · last updated Oct 22

Muse Spark Open

On Oct 22, Alexandr Wang offered $100k to the fastest Muse Spark Open implementation.

LatestOfficialAlexandr Wang offers $100k to the fastest Muse Spark Open implementationOct 22

Company1 update · last updated Oct 22

Canva

As of Oct 22, Canva has released its first in-house image generation model and is making it available to all users, promising better, more customizable results at lower cost, TechCrunch reports.

LatestReported by TechCrunchCanva releases its first image generation model and opens it to all users, per TechCrunchOct 22

Company1 update · last updated Oct 22

Nirva

As of Oct 22, Nirva raised $33 million for a $249 ambient AI pendant designed to listen only to its wearer, and plans to ship by year-end, TechCrunch reports.

LatestReported by TechCrunchNirva raises $33M for an ambient AI pendant that only listens to its wearer, per TechCrunchOct 22

Product1 update · last updated Oct 22

RIPL

On Oct 22, Liquid AI launched RIPL, a real-time inference engine it said returns LFM answers in under 50 ms.

LatestOfficialLiquid AI launches RIPL, a real-time inference engine it says returns LFM answers in under 50 msOct 22

Product1 update · last updated Oct 22

OlmoEarth

On Oct 22, Ai2 released OlmoEarth v1.3 satellite embeddings covering the whole planet at 10 m.

LatestOfficialAi2 releases OlmoEarth v1.3 satellite embeddings covering the whole planet at 10 mOct 22

Company2 updates · last updated Oct 22

MIT

On Oct 13, MIT researchers show randomly connected neural networks can self-assemble gradient-style learning from local plasticity rules.

LatestReported by TechCrunchMIT marooned 13 AI agents on a virtual island and they invented games, per TechCrunchOct 22

Product1 update · last updated Oct 22

Theorem

On Oct 22, Anthropic shows LLMs can produce formal software-verification proofs at large scale with Theorem.

LatestOfficialAnthropic shows LLMs can produce formal software-verification proofs at large scale with TheoremOct 22

Product2 updates · last updated Oct 22

Anthropic-managed Claude

On Oct 22, Google Cloud made an Anthropic-managed Claude deployment option generally available.

LatestOfficialAnthropic launches Claude Models on Google Cloud as an Anthropic-managed Claude APIOct 22

Product1 update · last updated Oct 22

Google Cloud Knowledge Catalog

On Oct 22, Google Cloud Knowledge Catalog preview added self-learning enrichment, context set and a new UI.

LatestOfficialGoogle Cloud Knowledge Catalog preview adds self-learning enrichment, context sets and a new UIOct 22

Product1 update · last updated Oct 22

ForTy

On Oct 22, Google DeepMind released ForTy, a 10 m map of natural, planted and farmed forests.

LatestOfficialGoogle DeepMind releases ForTy, a 10 m map of natural, planted and farmed forestsOct 22

Product1 update · last updated Oct 22

WorkForge

On Oct 22, Tencent Hunyuan introduced WorkForge for verifiable long-horizon agent training environments.

LatestOfficialTencent Hunyuan introduces WorkForge for verifiable long-horizon agent training environmentsOct 22

Product13 updates · last updated Oct 22

Kiro

On Oct 22, Kiro describes a trap in self-improving agents that grade their own fixed.

LatestOfficialKiro describes a trap in self-improving agents that grade their own fixesOct 22

How we label

Official: An institution announces something that has already happened, through its own channel or a founder or executive account.
Statement by …: A named person's views, predictions, plans or comments, including remarks reported by media when the original is unavailable.
Reported by …: Only media or another third party is saying it. We name the outlet.
Unconfirmed: Circulating online (posts, comments, leaks). No company, named person or outlet has confirmed it.

Items older than 72 hours never go in Top stories and show their original date.