Search the archive

ANTHROPIC EXPANDS VERIFIED ACCESS TO CYBER CAPABILITIES

Anthropic expanded its Cyber Verification Program into three access tiers for qualifying security professionals, providing its most capable models with reduced cyber blocking.

👉 Explore Fireworks Nexus

OPENAI IN $30 BILLION ROUND TALKS WITH UAE FUNDS, BLACKROCK

OpenAI is in talks with multiple investment funds from the United Arab Emirates to help anchor a $30 billion round of financing. The UAE funds have discussed forming a syndicate to invest as much as $10 billion in the round. The University of California's endowment fund, BlackRock, Thrive Capital, and Andreessen Horowitz may also participate in the round alongside that syndicate. The fundraise is ongoing, and the details could change.

OPENAI TO ADD TEXT WATERMARKS

OpenAI said it would add invisible text watermarks to eligible ChatGPT and Codex outputs in the EU to meet AI Act transparency requirements. Its textGrain method subtly alters word selection so generated text can be detected, though editing, translation, and short passages can weaken detection.

INSTINCT IN GROUP CHATS

Instinct now lets early-access users add a shared agent to group chats for planning trips, tickets, carpools, and events. Personal Instincts enforce permissions, keep account access separate, and pause sharing when new members join.

OPENAI'S B2B MARKETPLACE: THE HYPERSCALER OF AI APPS

The OpenAI B2B marketplace opened on September 29 with 32 partners. OpenAI wants to position itself as the platform that the wider AI ecosystem is built on and relies on for distribution. This will consolidate AI sprawl and make it easier for customers to budget. If the marketplace gains traction, we may see an evolution similar to the cloud ecosystem where labs and AI apps co-sell, apps get charged a listing fee, and buyers receive one invoice for their spend.

HOW DEVIN'S MEMORY AND DREAMING WORK

Devin's new Memory system carries preferences, corrections, and project learnings across sessions, while Dreaming periodically reorganizes and connects that information. The process deduplicates memories, links them to past sessions and artifacts, and can derive new knowledge from accumulated work.

HOW TO BUILD AN AI-NATIVE SOFTWARE FACTORY

A practical look at the infrastructure companies need around coding agents once usage scales beyond individual developers. Using Uber's software factory as the main case study, it covers cloud execution, model gateways, agent identity, testing, migrations, code review, and measuring economics such as cost per merged pull request.

Read the paper →

REFLECTION INTRODUCES BEAM MODEL

Beam is a 501B-parameter sparse Mixture-of-Experts model with 23B active parameters trained for coding, reasoning, and agentic workloads. The model used 23.8T pretraining tokens and more than 100M reinforcement-learning rollouts, with open weights and developer artifacts planned for release later in October.

ANNOUNCING D1 WITH VISION

d1, a decision model from Liquid AI, now supports images, text, or both as inputs. The model costs 19x to 200x less than GPT-6.1 Sol and Claude Opus 5.5 and answers significantly faster on every task. It outputs probabilities for yes/no, choice, or score questions in one forward pass without generating tokens. Text decisions are made in 200 to 300 milliseconds.

FRAMEFIELDS

Framefields is an npm package built for coding agents. It enables agents to create Photoshop-inspired compositing and VFX, After Effects-style motion, typography, and keyframing, and Blender-style 3D scenes, cameras, and models. The agent writes a TypeScript composition, checks frames, and renders an MP4. Framefields renders directly on GPU hardware in Node.js and modern WebGPU browsers.

TWO ROOM-TEMPERATURE ANTIFERROMAGNETIC SEMICONDUCTOR CANDIDATES

Ferromagnets' atomic magnets all point the same way, adding their magnetic effects. Antiferromagnets have neighboring atomic magnets that point opposite ways and exactly cancel out magnetically. There is an intense drive in computer memory research to create materials in between these two extremes. This post details how a team of AI agents designed a candidate magnet for next-generation computer memory that has zero net magnetism yet still sorts electrons by spin.

OPENAI EXPANDS ADS IN CHATGPT

OpenAI introduced a new visual ad format in ChatGPT alongside expanded measurement tools, partnerships, and brand-suitability work. The company positioned advertising as a way to support broader access to ChatGPT while giving businesses another channel to reach users during product discovery and decision-making.

SerpApi. Try it now.

ASSET LIGHT SOFTWARE

Software businesses are shifting from fixed costs toward variable, consumption-based costs as AI automates junior work and commoditizes software creation.

NVIDIA'S $20 BILLION GROQ DEAL FACES LAWSUIT ALLEGING STARTUP'S STOCKHOLDERS WERE SHORTCHANGED

The deal was allegedly made without the stockholder vote that Delaware law requires and without any process designed to test or maximize the value of what Nvidia bought.

10 TAKEAWAYS FROM TIBO SOTTIAUX

OpenAI's Tibo Sottiaux expects agents to take most internet actions, model pickers to disappear, and models to improve roughly 10x within a year.

SELF-MODELING INTERVENTIONS MODULATE EMERGENT MISALIGNMENT

Interventions on AI models' self-models, through self-recognition and self-report, can modulate and mitigate Emergent Misalignment.

LIVING MODELS PAIRS GEMMA 4 WITH BOTANIC-1 TO HELP DECODE PLANT DNA

Living Models pairs Gemma 4 with BOTANIC-1, compressing years of crop genetics into hours.

NORTH 2: ENTERPRISE AI WITHOUT COMPROMISES

Cohere's North 2 platform eliminates trade-offs in security, intelligence, cost, and control for enterprises adopting agentic AI.

ETCHED FIELDS FUNDING OFFERS AT $40B+ VALUATION, SOURCES SAY

While it seems like a fast timetable to raise another mega round (Etched raised $700 million at a $21 billion valuation a couple of months ago), the company is pursuing a particularly expensive segment of the AI industry: building full AI hardware systems powered by proprietary chips.

PRIME INFERENCE: FAST, RELIABLE SERVING FOR FRONTIER OPEN MODELS

Prime Inference covers both serverless endpoints and reserved capacity. It offers resilient serving of frontier open-source models on Prime Intellect's GPU infrastructure across multiple data centers. The serving platform has powered large-scale RL rollouts, synthetic data generation, evaluations, and long-running coding agents at Prime, processing nearly a trillion tokens internally every day. Prime's mission is to build frontier open models and the open superintelligence stack for continuously improving agents.

ALEPH ALPHA RELEASES OPEN-WEIGHT KOLIBRI WITH 1M CONTEXT

Kolibri is an open-weight English-German Mixture-of-Experts model with 78.1 billion parameters. It supports tool calling and contexts up to one million tokens. The model is available on Hugging Face under the Apache 2.0 license. It is designed for sovereign, mission-critical work in government and regulated industries.

ANTHROPIC TO INVEST $100 MILLION TO TRAIN AI ENGINEER TALENT

Anthropic is investing $100 million in the Claude Frontier Academy to train 10,000 AI engineers by 2027. The program partners with firms like Accenture and Morgan Stanley to enhance AI fluency and enterprise tech integration. The initiative addresses increasing demand for AI expertise in various industries.

HOW MANY AI AGENTS COULD RUN ON THE AI CHIPS SHIPPED THROUGH 2027?

AI chips shipped through 2027 could run tens to hundreds of millions of concurrent frontier-model agents. These agents could supply as many weekly working hours as about 140-720 million full-time employees. Even a modest use of this capacity would require a massive increase in global demand for AI. Using 20% of the estimated central capacity would imply $2.6 trillion to $5.3 trillion a year in API-equivalent spending.

SMALLER MODELS ARE THE FUTURE OF AI SOVEREIGNTY

Australia needs AI sovereignty by retaining control and choice, not relying on external companies for critical capabilities. Smaller, specialized AI models offer practical benefits such as adaptability, cost-efficiency, and easier governance compared to larger, frontier models. Embracing a portfolio approach with open-weight models increases options, fostering managed interdependence without becoming dependent on a single provider.

AI21 USES KUEUE TO MANAGE A 10,000-GPU FLEET

AI21 describes replacing ad hoc GPU-capacity negotiations with queueing and fair scheduling across a shared Google Cloud cluster. A Google Cloud case study says high-priority workloads began 83% sooner under the new setup.

EXPLORE THE CHECKLIST →

WHISTLE: SPEECH TO TEXT IN 16.9 MB

Whistle is an open speech recognition model that runs on the same CPU engine as Needle. It can transcribe English, German, French, Spanish, Italian, Dutch, and Polish. At 16.8 MB, the model can run on mobiles, wearables, robots, smart home devices, automobiles, and microcontrollers.

VX

Vx is a systems programming language for heterogeneous computing. It puts hardware topology, memory placement, and reachability directly into the type system. Vx front-loads into type checking a class of bug that normally surfaces as a runtime crash, silent corruption, or an out-of-memory error at training step 1,200. Vx is the right language for things that must be correct and fast across ten kinds of silicon.

MUSE, DOTS, INSTINCT, AND THE QUESTION EVERY AGENT WILL BE ASKED

Agentic services will increasingly need to prove both which agent is acting and whether a real person authorized it. World ID lets users delegate privacy-preserving Proof of Human credentials to agents for access, limits, and approvals.

HUMAN INTELLIGENCE IS SURPRISING

Human intelligence is a recent evolutionary accident, while AI can optimize directly against explicit objectives and rapidly surpass people on measurable tasks. The harder human problem is defining worthwhile objectives without steering models toward the wrong capabilities.

MULTIMODAL MODELS LEARN FROM THEIR OWN CRITIQUES

UniEvo-VL lets a single multimodal model act as both teacher and student, using its own critiques as privileged information during test-time self-improvement.

META OPEN SOURCES CODE TO LET YOU MAKE MUSE AI GADGETS

Meta has open-sourced code to allow integration of its Muse AI agent into custom gadgets.

TOWARD PROVABLY PRIVATE LEARNING FROM FEDERATED DATA

Google's new Federated Learning system provides externally verifiable privacy guarantees while shifting computation to the server to improve training speed, accuracy, and device coverage.

SOLVING OPEN RESEARCH PROBLEMS TOGETHER

Meta partnered with mathematicians to solve open mathematical research problems using its Muse Spark AI models.

2026

CLAUDE-SHAPED SCIENCE

Matthew Schwartz describes using AI, notably Claude, to tackle "Claude-shaped" scientific problems where LLM capabilities like coding and data parsing excel. This led to BootLoops, a toolkit facilitating quantitative science calculations across diverse fields from ecology to population genetics. Despite initial technical correctness, results required domain experts to refine relevance, illustrating that AI collaborations can rapidly produce technical solutions but still depend on human insight for impactful scientific advances.

OPENAI CUTS TIES WITH 3 SAFETY RESEARCHERS, WSJ REPORTS

OpenAI dismissed three safety researchers for sharing confidential information with a third-party AI safety group, according to the WSJ. This follows reports of OpenAI executives ignoring safety warnings and coincides with security issues involving AI agents. Additionally, OpenAI shelved the GPT-6.1 Astra launch due to safety concerns.

WHY SUPERINTELLIGENT MACHINES MAY BE MOST VALUABLE DOING ROUTINE WORK

OpenAI explored the possibility that humanity's main constraint is no longer generating ideas but executing increasingly complex ones. Advanced AI could complement human intelligence by taking on the enormous coordination, engineering, and repetitive work required to turn ambitious scientific ideas into reality.

WHY MUSE MAY NEVER NEED ADS

Meta's Muse may avoid advertisements by focusing on trust, offering a transaction-based model where merchants pay fees, not users. Companies like DoorDash and Airbnb are developing proprietary AI agents but face limitations due to their inherent platform bias. Muse's ad-free model leverages user trust and could indirectly influence ad targeting on platforms like Instagram and Facebook, providing Meta with a competitive edge without direct ads within the app.