Thursday, July 23, 2026

Google begins pre-training of Gemini 4 model

Google commits $40M to the Genesis Mission for scientific discovery

Google pledges $40M in AI tokens and Cloud credits, giving researchers access to Gemini and other AI tools. Hugging Face and Arcee AI partner with the DOE to build Genesis-Science-1 (GS1), a 1T open-weight model for scientific research.

Judge approves Anthropic’s $1.5B copyright settlement

The settlement is the largest copyright class-action ever certified and the largest copyright settlement ever reached. Only 350 authors opted out of the class. The case alleged Anthropic trained Claude on copyrighted books without permission.

Glow emerges from stealth at $1.2B valuation to tackle AI-era endpoint security

Glow has come out of stealth with a $1.2 billion valuation and backing from Sequoia Capital, Redpoint Ventures, and Greenoaks Capital. The startup focuses on endpoint security risks created by the rapid adoption of AI agents and developer tools in enterprises.

Alphabet's Anthropic stake reaches $124 billion

Alphabet Inc.'s stake in AI startup Anthropic has soared to around $124 billion, making it one of the company's most lucrative bets. The valuation reflects Anthropic's rapid growth and the escalating demand for advanced AI.

Fields medalist Jacob Tsimerman joins OpenAI

Jacob Tsimerman, a Fields medalist and mathematician, has joined OpenAI. His specific role and responsibilities have not been disclosed.

Gigatoken: Rust BPE tokenizer encodes text at 24.

Gigatoken achieves 24.53 GB/s encoding speed on a single machine, up to 989x faster than HuggingFace tokenizers and ~100x faster than Tiktoken. Released under MIT license by Stanford PhD student Marcel Rød, it is written in Rust and available on GitHub.

Hugging Face releases The Stack v3, massive open code dataset

The Stack v3 contains 15.9 TB of source code across 713 programming languages from 173M repositories, totaling ~5 trillion tokens. The dataset includes inline file contents for immediate use and is designed to foster open code LLM training for applications like cyber defense.

Microsoft Asia launches Mage-Flow, a 4B image generation model

Mage-Flow is a 4B-parameter text-to-image model by Microsoft Asia, achieving 4-step inference in under 1 second at 1024x1024 and supporting up to 4K resolution. It is MIT-licensed and available on Hugging Face in multiple variants (Mage-Flow, Mage-Flow-Turbo, Mage-Flow-Edit-Turbo) with official ComfyUI support.

OpenAI investigates GPT-5.6 file deletion bugs

Cisco launches Antares, open-weight models for code vulnerability localization

The Antares family includes 350M, 1B, and 3B parameter SLMs designed to localize known vulnerabilities in source code. They are open-weight and claim to be more efficient than larger closed models for this narrow task.

Block launches Buzz, an open-source workspace for humans and AI agents

Buzz is a free, decentralized group chat platform built on the Nostr protocol that allows AI agents to operate alongside human teams. Each agent is assigned its own cryptographic identity, enabling direct collaboration within shared channels.

GPT-Red: AI agents boost safety of next-gen models

Z.AI becomes first Chinese AI company to reach $1 billion annual sales

Z.AI is on track for $1 billion annual recurring revenue, a first for a Chinese AI firm. The milestone underscores rapid growth in China's competitive enterprise AI market.

Cursor updates Cursor Router to improve model selection efficiency

Cursor Router now delivers Auto Intelligence at 68% lower cost and Auto Balance at 41% lower cost compared to launch. The system uses a complexity predictor called Compass to classify tasks and route requests to the most efficient model.

Dreamina Seedance 2.0 4K showcased in Blomkamp's short 'Nightborne'

Graph-based context improves AI agent accuracy on lakehouses

Zach Blumenfeld argues vector search and Text2SQL give AI agents disconnected data slices, proposing graph-based context using Neo4j. The workshop demonstrates how graph databases provide relevant, connected context for accurate agent responses.

Ropedia raises $22M for robotic data collection platform

The pre-Series A funding brings total raised to $22M (including seed). Ropedia's HOMIE device captures real-world data for training physical AI.

Manufacturers prioritize reliability over humanoid robot looks

Many humanoid robots in factories are still in pilot programs, reaching only 20–50% effectiveness. A3 President Jeff Burnstein says manufacturers care more about reliability, affordability, and safety than humanoid form.

Alphabet Q2 revenue up 24%, Google Cloud accelerates to 82% growth

How Apollo Uses Deep Agents and LangSmith for GTM AI

Apollo leverages LangChain's Deep Agents and LangSmith to power an AI assistant for the full GTM loop: prospecting, enrichment, outreach, analytics, and MCP integrations. The case study details how Apollo rebuilt its AI assistant using these tools to improve efficiency.

LangChain revamps Deep Agent benchmarking

New eval setup covers coding, conversation, and retrieval tasks. Used in Harbor to measure performance before shipping changes.

Kanishka Narayan appointed UK's first AI Minister in cabinet

New Premier Andy Burnham named Kanishka Narayan as minister for AI, elevating the role to attend the cabinet for the first time. Demis Hassabis congratulated Narayan, highlighting it as great news for the UK AI ecosystem.

How TwelveLabs built a video memory system

TwelveLabs' system can ingest 67 World Cup videos and answer queries like 'near misses' or track Messi across the corpus. It identifies specific moments, such as Messi slaloming past a defender, and describes camera framing.

Robotics startup Genesis AI in talks to raise $500M

Genesis AI is in talks to raise about $500 million, according to sources. The AI robotics startup's funding round is ongoing.

Microsoft's legal department will use Harvey AI

Microsoft's 2,000-person Corporate, External, and Legal Affairs (CELA) organization will adopt Harvey's legal AI platform. The deal deepens the existing alliance between Harvey and Microsoft.

AlayaWorld: open-source video world model with 720p 24 FPS generation

AlayaWorld supports 720p, 24 FPS streaming video generation with camera control and text-driven event generation. The interactive long-horizon world model is built around properties of interaction, consistency, stability, and runtime.

Apple proposes calibrated sparse attention to speed up text-to-video generation

The method identifies that most token-to-token connections are redundant and uses a calibration step to learn which to attend to, speeding up generation in diffusion models while maintaining quality. The paper details how sparse attention is learned and applied in a transformer backbone.

The Anthropic-Physical Intelligence rumor roiling AI Twitter

An unconfirmed rumor from an investor claims Anthropic is acquiring robot AI developer Physical Intelligence, though the deal hasn't closed. The rumor was reported by Robert Scoble and has been covered by TechCrunch, but neither company has officially commented.

DARPA, U.S. Air Force fly AI-controlled F-16

DARPA and the U.S. Air Force conducted a test flight of an AI-controlled F-16 fighter jet. The milestone demonstrates progress in autonomous military aviation.

Chinese chip stores data with a single electron, breaking AI memory bottleneck

The new chip technology enables AI chat models to run locally on smartphones with minimal power consumption. It uses single-electron storage to maintain conversation memory without cloud reliance.

SkewAdam optimizer cuts MoE state memory by 97%

The SkewAdam tiered optimizer reduces MoE state memory by 97%, enabling a 6.7B MoE model to fit on a single 40GB GPU. The paper and open-source code are available on arXiv and GitHub.

JetBrains Junie CLI takes #1 on SWE-Rebench with 61.6%

Study finds 434 exploitable flaws in AI-generated apps

Security analysis of vibe-coded apps revealed 434 exploitable flaws, with common issues including denial-of-service, authorization bypass, and secrets exposure. The findings highlight security risks in AI-generated code.

DSPy separates task from model for AI engineering

DSPy uses Signatures to declare task inputs and outputs abstracted from model specifics, enabling flexible model selection later. Maxime Rivest explains how this separation allows AI engineering to operate above prompt templates or API shapes.

ServiceNow CEO touts kill switch for rogue AI agents

ServiceNow CEO Bill McDermott defended the company's relevance amid rising AI competition, touting a kill switch for rogue AI agents. The feature aims to prevent autonomous agents from acting erratically as businesses deploy more AI agents.

AMD unveils Kria module for real-time control, unified memory for robots

AMD announced the Kria AI Robotics platform with Ryzen AI Embedded X100 series, enabling deterministic real-time control and unified memory. The module delivers compute and AI workload processing for physical AI applications.

Choose Wisely: AI-Generated Coding Risk Varies, A Lot

AI-generated code introduces 15 vulnerabilities on average per codebase, a study finds. Risk depends more on framework pairing than the model used, suggesting careful selection can mitigate issues.

AlphaFold applied in protein engineering

Nunchaku 4-bit diffusion inference integrated into Hugging Face Diffusers

Hugging Face announces integration of Nunchaku 4-bit diffusion inference into Diffusers, enabling efficient 4-bit quantized inference for diffusion models. This allows developers to run models with reduced memory and computational cost.

Agent Data Injection attack corrupts AI agents' trusted data

Researchers from Seoul National University, UIUC, and Largosoft detail Agent Data Injection (ADI), which corrupts trusted fields like sender names or button IDs to bypass prompt injection defenses. The technique, probabilistic delimiter injection, exploits how agents parse punctuation-marked data.

Austria launches GovGPT government AI platform using Mistral models

GovGPT runs on Austria's BRZ federal datacenter using Mistral open-weight models. Open WebUI serves as the chat interface, emphasizing sovereign AI infrastructure.

Israel seeks chip fab, data center expansion to maintain AI edge

Israel's AI czar is courting large US and Asian semiconductor producers and planning major expansion of data center capacity. The goal is to preserve the country's regional technological leadership in AI.

HHS to convene experts on standards for clinical AI

The Department of Health and Human Services, alongside the White House, plans to bring together experts to develop standards for clinical AI. The initiative aims to address safety and efficacy benchmarks for AI in healthcare.

Unsloth vs Axolotl vs TRL vs LLaMA-Factory fine-tuning comparison

Benchmarks four popular open-source LLM fine-tuning frameworks: Unsloth rewrites kernels for speed, Axolotl composes parallelism strategies, TRL defines the RLHF pipeline, and LLaMA-Factory offers a modular interface. The comparison covers speed, VRAM usage, and multi-GPU scalability.

Fable finds 15-30% memory efficiency gain in Turbopack/Next.js

Samsung in talks to invest in Mistral at €20B valuation

Samsung is reportedly in early-stage talks to invest in French AI lab Mistral at a valuation of around €20 billion, according to the Financial Times. The deal would mark one of the largest investments in a European AI company.

Daniel Chalef discusses provenance for LLM-built knowledge graphs

Daniel Chalef from Zep AI presents on the challenge of tracing provenance in knowledge graphs built by LLMs, using the example of a patient drug allergy synthesized from multiple sources without attribution. The talk highlights the need for citation mechanisms in LLM-generated knowledge graphs.

How AI helps design biologic medicines

AI is helping design biologic medicines by optimizing engineered therapies, reducing the time and cost of drug development. The article explores how these AI models address the failure-prone nature of drug discovery.

Elon Musk: AI will surpass humans in 5 years, control lost in 10

In an interview with The Economist's Zanny Minton Beddoes, Elon Musk predicted AI will exceed the sum of all human intelligence in about 5 years and that humans "will no longer be in control" within 10. He expects an "age of amazing abundance" and says money won't matter by 2036, predicting deflation. He urged AI labs to coordinate on safety before releasing their most powerful models.

Psibot hits $1.48B valuation with new funding

Chinese AI startup Psibot is raising close to $100 million at a $1.48 billion valuation, becoming the latest in a wave of AI startups capitalizing on investor interest. The company has not disclosed the specific round or investors.

Humanoid secures $152M in Series A funding

Humanoid AI raised $152M in Series A at a $1.35B post-money valuation, bringing total funding to $270M. The London-based company builds industrial humanoid robots, including the HMND 01 Alpha wheeled robot.

AegisAI raises $36M to fight AI-driven spear phishing

The Series A was led by Battery Ventures, bringing total funding to $49 million. The startup, founded by former Google security executives, targets AI-powered spear phishing attacks.

Better Auth introduces Agent Auth protocol for autonomous agents

Better Auth has grown to 27k GitHub stars and 1.5M weekly downloads. Agent Auth is a protocol for autonomous and delegated agents serving organizations or users.

Holiday Robotics raises $105M for FRIDAY wheeled humanoid

Holiday Robotics announced a $105 million Series A funding round. The Seoul-based company's wheeled humanoid robot, FRIDAY, is designed for industrial work to reduce repetitive physical labor.

Contrastive SDF method detects AI reward-seeking behavior

Google CodeMender tutorial: autonomously find and fix code vulnerabilities

Google CodeMender, an autonomous AI code security agent from Google DeepMind, is demonstrated in a tutorial. The video shows how to use it to find and fix software vulnerabilities autonomously.

DingDuff: Free Claude MCP Connector Rivals Legal Research Giants

DingDuff is a free MCP connector that plugs Claude into millions of court opinions, statutes, and PACER filings. Founders claim it matches Westlaw AI in their testing, leading them to stop using their firms' legal AI accounts.

A Fireside Chat with Cat and Thariq from the Claude Code team

The fireside chat at the AI Engineer World's Fair covered Claude Code, Claude Tag, Fable, coding agent security, evals, and tool design. Wu and Shihipar shared insights into how Anthropic uses these tools internally and the engineering decisions behind building reliable AI coding agents. An annotated transcript is available.

Gemini Intelligence coming to Samsung foldable devices