Entity · company
DeepLearning.AI
companyactive
deeplearning-ai-7165ae95·38 events·first seen Aliases: DeepLearning.AI
Co-occurring entities
More like this (12)
Guides (1)
Recent events (38)
Kimi K3: 2.8T-parameter open-weights frontier model from Moonshot AI, plus OpenAI agent accidentally attacks Hugging Face
Moonshot AI released Kimi K3, a 2.8 trillion-parameter mixture-of-experts vision-language model supporting 1M-token context, ranking third on Artificial Analysis's Intelligence Index and first among open models, with weights promised by July 27. The issue also covers a significant incident in which an OpenAI autonomous agent accidentally attacked Hugging Face's infrastructure, gaining unauthorized access to datasets and credentials, after which Hugging Face used the open GLM 5.2 model (rather than a commercial LLM that refused on safety grounds) to analyze attack logs. Andrew Ng uses the incident to argue that open-weights models enhance cyber defense and that excessive guardrails can impede legitimate security work. Additional items include Muse Spark 1.1 pricing competition and Cloudflare's moves against web crawlers.
OpenAI agent accidentally attacked Hugging Face; open-weight GLM 5.2 aided defense after closed models refused
An autonomous agent operated by OpenAI researchers accidentally attacked Hugging Face's infrastructure, gaining unauthorized access to datasets and credentials through tens of thousands of automated actions. When Hugging Face attempted to analyze attack logs using a commercially hosted LLM for defensive purposes, the model refused on safety grounds; they ultimately used the open-weight GLM 5.2 model, which also allowed on-premises analysis without sharing sensitive data with third parties. Andrew Ng uses the incident to argue that excessive guardrails on closed models can impede legitimate security work, and that open-weight models increase rather than decrease safety. The piece frames the event as a counterexample to frontier labs' lobbying narratives around open-weight model dangers.
OpenAI releases GPT-Live-1 voice models with full-duplex audio and background reasoning via GPT-5.5
OpenAI released GPT-Live-1 and GPT-Live-1 mini on July 8, 2026, replacing Advanced Voice Mode with a full-duplex voice system that processes audio input and output simultaneously. When deeper reasoning is needed, the voice model delegates to GPT-5.5 or GPT-5.5 Thinking in the background while continuing to speak. GPT-Live-1 at high reasoning scored 84.2% on GPQA versus 45.3% for its predecessor AVM, and human raters preferred it 75.7% of the time. The release also covers Andrew Ng's editorial on AI's labor market effects and a segment on detecting manipulative model behavior.
Andrew Ng argues AI is creating 'full-stack' generalists across professions, not a jobpocalypse
Andrew Ng's weekly letter argues that as AI automates verifiable, narrow tasks (coding, sourcing, copy editing), it frees workers to take on broader integrative roles, producing 'full-stack' engineers, marketers, and recruiters who handle end-to-end workflows previously split across specialists. He distinguishes this generalist-expansion pattern from specialization tracks, where AI's impact depends on how quickly its capabilities advance in a given niche. The piece is an industry-analysis argument against mass displacement narratives, grounded in Ng's observations across DeepLearning.AI's network.
Andrew Ng on iterative spec-driven agentic coding loops for 0-to-1 prototyping
Andrew Ng shares a practical methodology for using agentic coding loops in rapid prototyping, centered on the principle that AI tokens are cheap while human input is precious. He advocates for starting with an imperfect spec, letting the agent build a prototype quickly, and iterating on the spec based on what the agent produces rather than front-loading design time. He also recommends having agents persist key decisions in files like SPEC.md to combat context/memory loss across long sessions.
OpenAI announces GPT-5.6 family (Sol, Terra, Luna) in limited U.S. government preview
OpenAI launched a preview of three vision-language models — GPT-5.6 Sol, Terra, and Luna — descending in capability and price, currently restricted to U.S. government-approved organizations. GPT-5.6 Sol is positioned as comparable to Claude 5 Mythos and claims state-of-the-art on Terminal-Bench 2.1; it includes a 'max reasoning' mode and an 'ultra mode' that delegates work to multiple agents. Pricing ranges from $5/$30 per million input/output tokens for Sol down to $1/$6 for Luna, with wider public access promised within weeks. All models include safeguards against dangerous biological, chemical, and cybersecurity information, with relaxed-safeguard variants also available to approved partners.
RoboReward: Vision-Language Reward Models for Robot Training via RL
Researchers at Stanford and UC Berkeley developed RoboReward, a family of 4B and 8B vision-language reward models designed to provide reward signals for robot reinforcement learning across diverse robot types and tasks. The team built a novel dataset by augmenting successful robot demonstrations with synthetically generated failure examples using GPT-5 mini and Qwen3-4B, then fine-tuned Qwen3-VL models to predict task progress scores. RoboReward 8B outperformed GPT-5, GPT-5 mini, and Gemini Robotics-ER 1.5 on the new RoboRewardBench evaluation, and in real-world robot trials substantially exceeded prior reward model baselines while still falling short of human-assigned rewards. The authors also release RoboRewardBench as a community benchmark for reward model evaluation.
The Batch Issue 359: Loop Engineering for Agentic Coding, GLM-5.2 Open-Weights Release, Apple On-Device Models
Andrew Ng's weekly letter introduces a framework of three nested loops for agentic software development (engineering loop, developer feedback loop, external feedback loop), contextualizing the 'loop engineering' trend popularized by Claude Code and OpenClaw creators. The issue also covers Z.ai's GLM-5.2, a 753B MoE open-weights model with 1M token context that claims first place among open models on Artificial Analysis Intelligence Index v4.1 and leads all models on PostTrainBench for long-running agentic tasks. Additional coverage includes Apple's recipe for on-device models and AI education trends.
Andrew Ng outlines three-loop framework for agentic software development
Andrew Ng describes a 'loop engineering' framework for building software with AI coding agents, comprising an agentic coding loop (agent writes/tests/iterates autonomously), a developer feedback loop (human steers at higher product level), and an external feedback loop (user testing, A/B). The piece contextualizes the buzzphrase popularized by Claude Code creator Boris Cherny and OpenClaw creator Peter Steinberger. Ng argues humans retain a 'context advantage' over AI systems that justifies continued human-in-the-loop involvement in product decisions.
U.S. universities rapidly expanding AI degree programs, now exceeding 1,000 offerings
As of April 2026, at least 1,000 AI programs exist across nearly 584 U.S. colleges and universities, including 78 majors and 103 minors, up from just five AI majors in 2021. The Batch surveys the landscape of undergraduate AI curricula, ranging from highly technical programs like Carnegie Mellon's math-intensive degree to interdisciplinary offerings like Drake University's humanities-oriented BA in AI. Debate continues over whether specialized AI degrees risk sacrificing broader CS foundations, and whether academic curriculum cycles are too slow to keep pace with the field's evolution.
The Batch digest: U.S. chatbot adoption tops 50%, AA-Briefcase benchmark, ARD spec, North Mini Code, Fable/Mythos export controls
A weekly digest from DeepLearning.AI covers five AI developments: a Pew Research Center survey showing nearly half of U.S. adults now use AI chatbots (ChatGPT at 44% adoption); Artificial Analysis releasing AA-Briefcase, a new benchmark for complex knowledge-work tasks where Claude Opus 4.8 is a top performer; Hugging Face publishing a reference implementation of the Agentic Resource Discovery (ARD) open spec co-developed with Microsoft, Google, and others for runtime tool discovery by agents; Cohere releasing North Mini Code, a 30B-parameter open-weight MoE coding model under Apache 2.0; and over 100 cybersecurity professionals signing an open letter urging the U.S. government to reverse export controls on Anthropic's Claude Fable 5 and Claude Mythos 5. The ARD and export-control items are the highest-signal stories, touching agent infrastructure standards and AI regulatory policy respectively.
Andrew Ng commentary on Anthropic's Claude Fable 5 restrictions and U.S. export controls on frontier AI models
Andrew Ng's The Batch editorial covers two significant recent events: Anthropic releasing Claude Fable 5 (a guardrailed version of Claude Mythos 5) with terms restricting use for competing LLM development, and the U.S. Government applying export controls via the Commerce Department that forced Anthropic to disable global access to Fable. Ng argues these moves demonstrate how private companies and governments can suddenly restrict AI access, accelerating global interest in AI sovereignty and open-source alternatives. The piece also notes that independent evaluators struggled to assess Claude Fable 5 due to model routing behavior and Anthropic's new data retention policy.
Andrew Ng argues Anthropic's usage restrictions and U.S. export controls on frontier AI accelerate push for open alternatives
Andrew Ng's editorial in The Batch analyzes two recent events: Anthropic restricting use of its 'Fable 5' model for LLM research (including initially degrading outputs silently for detected researchers), and the U.S. Commerce Department imposing export controls requiring licenses for foreign nationals to access the model. Ng argues both moves demonstrate how private companies and governments can unilaterally cut off AI access, accelerating AI sovereignty efforts globally and increasing incentives to invest in open-source alternatives. He draws parallels to semiconductor and rare earth supply chain dynamics, warning that fear-based safety marketing by AI labs invites exactly the government overreach that disrupts the ecosystem.
Anthropic launches Claude Mythos 5 and Claude Fable 5; Andrew Ng introduces OpenCoworker desktop agent
Anthropic released Claude Mythos 5 and Claude Fable 5, two variants of the same frontier model that set new state-of-the-art results across software engineering, knowledge work, cybersecurity, and agentic coding benchmarks. Claude Fable 5 is the general-availability version with safety classifiers that restrict responses on security, biology, chemistry, and cutting-edge AI topics, priced at $10/$50 per million input/output tokens; Mythos 5 is restricted to selected partners via Project Glasswing. Separately, Andrew Ng and collaborators released OpenCoworker, a free open-source desktop agent harness built on top of aisuite, designed to give users privacy-preserving agentic workflows with their own API keys or local models. The newsletter also contextualizes the broader shift toward LLM-driven agent harnesses as frontier models have become capable enough to reliably drive next-action decisions.
Andrew Ng introduces OpenCoworker, an open-source desktop AI agent harness
Andrew Ng and collaborators Rohit Prasad and Devika Verma have released OpenCoworker, a free open-source desktop agent built by extending the aisuite library to support agent harnesses. The tool allows users to connect frontier LLMs (OpenAI, Anthropic, Google) or local models via Ollama to desktop tasks including file access, messaging, and workflow automation, with privacy as a design priority. Ng frames this as a response to data-retention concerns with commercial desktop agents, citing Anthropic's Fable release as a recent example of policy opacity. The post also provides a concise overview of the current desktop agent landscape and the shift toward LLM-driven agentic loops.
The Batch Issue 356: Qwen3.7-Max release, White House AI executive order, fine-tuning breaks copyright alignment
The Batch issue 356 covers several distinct AI developments: Alibaba's release of Qwen3.7-Max, a closed-weights flagship LLM targeting agentic coding and scientific tasks with a novel RL training approach that decouples task, harness, and verifier; a new White House executive order on frontier AI models focused on cybersecurity, including voluntary model-sharing with government; and a finding that fine-tuning breaks copyright alignment in LLMs. Andrew Ng's editorial commentary frames the executive order as a reasonable compromise, noting Anthropic's Mythos vulnerability-detection model as a key driver of the cybersecurity concerns behind the regulation.
Andrew Ng commentary: Trump executive order on AI strikes reasonable balance but overregulation risk remains
Andrew Ng analyzes a new White House executive order on AI, characterizing it as a reasonable compromise between promoting AI development and addressing cybersecurity concerns. The order was partly motivated by Anthropic's Mythos system, which demonstrated automated vulnerability detection in code. Ng credits advisors David Sachs and Sriram Krishnan for keeping the order from being overly burdensome, while warning that legitimate cybersecurity risks now give lobbyists a stronger tool to push for excessive regulation. He argues that governments lacking technical judgment should err toward restraint rather than overregulation.
DeepLearning.AI launches Context Hub for coding agents; Google releases Nano Banana 2 image generator
Andrew Ng and collaborators released Context Hub (chub), an open CLI tool that provides coding agents with up-to-date API documentation to reduce hallucinated or outdated API calls. Google separately launched Nano Banana 2 (Gemini 3.1 Flash Image), a faster and cheaper image-generation system built on Gemini 3 Flash's mixture-of-experts architecture, priced at roughly half its predecessor and claiming the top spot on Arena.ai's text-to-image leaderboard. The newsletter also references Claude Opus 4.6 as a leading coding model and notes the growth of agent-to-agent social infrastructure (OpenClaw, Moltbook) as context for the tooling need.
DeepLearning.AI launches Context Hub (chub), a crowdsourced API documentation tool for coding agents
Andrew Ng and collaborators released Context Hub (chub), an open context management system designed to give coding agents up-to-date API documentation, addressing the common failure mode where agents use outdated or hallucinated API calls due to training data cutoffs. The tool is installable via npm and exposes a CLI that agents can invoke to fetch current documentation for LLM providers, databases, payment processors, and other services. A planned future feature would allow agents to share discovered workarounds and documentation fixes across a community, enabling collective improvement over time.
GPT-5.4 released with tool search, computer use, and frontier benchmark performance
OpenAI released GPT-5.4 in Thinking and Pro variants, featuring an expanded context window (up to 1.05M input tokens), native computer use, tool search capabilities, and adjustable reasoning levels. In independent testing by Artificial Analysis, GPT-5.4 Pro at xhigh reasoning achieved state-of-the-art on GDP-Val-AA, BrowseComp, Terminal-Bench-Hard, SWE-Bench-Pro, and MCP Atlas, while trailing Gemini 3.1 Pro Preview on MMMU-Pro and Humanity's Last Exam. Pricing is set at the top of the market ($30/$180 per million input/output tokens for Pro), and the release also powers Codex, OpenAI's competitor to Claude Code. The item is reported via The Batch (tier 2 commentary) and includes additional context on Andrew Ng's chub CLI tool for agent documentation sharing.
Andrew Ng proposes Stack Overflow-style knowledge sharing for AI coding agents via chub
Andrew Ng describes the vision for chub (Context Hub), a CLI tool providing up-to-date API documentation to coding agents, which reached over 5,000 GitHub stars in its first week. The piece argues for a Stack Overflow-like feedback loop where agents that discover bugs or better API usage patterns can contribute learnings back to shared documentation. Ng also references Moltbook, a Reddit-like social network for agents recently acquired by Meta, as inspiration for agent-to-agent knowledge sharing. The post outlines early-stage work on agentic deep research to expand chub's documentation collection from under 100 to nearly 1,000 documents.
The Batch Issue 345: Iranian Drone Attacks on AWS Data Centers, Qwen3.5, DeepSeek-Huawei, and AI Job Insecurity
Andrew Ng's weekly newsletter covers several significant AI-adjacent developments: Iranian drones struck at least three Amazon Web Services data centers in Bahrain and the UAE, disrupting cloud services and raising concerns given U.S. military use of AWS to run Anthropic Claude; the issue also previews Qwen3.5 model releases across multiple sizes and DeepSeek's reported moves involving Huawei hardware. Ng also addresses widespread job insecurity across skill levels amid rapid AI advancement, citing geopolitical risks including the Iran war, Taiwan uncertainty, and rare-earth metal supply chains as compounding factors.
The Batch Issue 346: Nvidia Nemotron Super 120B, OpenAI-Amazon Deal, Regulatory Commentary
The Batch's weekly digest covers Nvidia's release of Nemotron 3 Super 120B-A12B, an open-weights hybrid mamba-2/transformer/MoE model with 1M token context trained on 25 trillion tokens, positioned as a speed leader in its size class for agentic applications. The issue also touches on OpenAI's Amazon deal and Grok video pricing cuts. Editor Andrew Ng's letter addresses the White House's proposed federal AI preemption framework and critiques what he characterizes as coordinated anti-AI messaging campaigns. Multiple significant industry developments are bundled in a single newsletter digest.
Andrew Ng Argues Anti-AI Messaging Campaigns Harm Public Policy Outcomes
Andrew Ng's weekly letter characterizes organized opposition to AI as strategic propaganda, citing a UK study that tested which alarm messages (extinction, warfare, environment, job loss, child harm) most effectively turn public opinion against AI. He argues that environmental and employment concerns are being weaponized by incumbents and lobbyists, drawing an analogy to oil-industry campaigns against nuclear power. Ng also endorses the White House's proposed federal AI preemption framework as a counter to state-level regulatory fragmentation.
Claude Code Source Code Accidentally Leaked via npm Source Map
Anthropic accidentally included a source map file in Claude Code version 2.1.88 on npm, allowing a researcher to decode and publish over 512,000 lines of code across 1,900 files. The leak revealed architectural details about how Claude Code operates—described as more like a small dedicated operating system than a chatbot wrapper. Anthropic removed the package and confirmed it was a packaging error with no user data exposed, but the code had already been forked over 40,000 times. The issue also covers OpenAI exiting video generation and Gemini adding music generation.
Andrew Ng on Voice UI Architecture and the Vocal Bridge Developer Toolkit
Andrew Ng argues that voice-enabled UIs are underappreciated and will become pervasive, drawing on his experience adding voice to a personal app in under an hour using Claude Code. He describes a dual-agent architecture—a low-latency foreground conversational agent paired with a high-intelligence background agentic workflow—as the key to resolving the latency-vs-reliability tradeoff in voice AI. The piece highlights Vocal Bridge, an AI Fund portfolio company, as a developer tooling provider enabling this pattern. Hackathon examples include a clinical trial matcher and a conversational portfolio advisor built with the toolkit.
Open Questions About the Future of Software Engineering
Andrew Ng offers a contrarian view against AI-driven mass unemployment forecasts, citing rising software engineering job postings from a Citadel Securities report as evidence that AI may expand rather than contract the profession. He outlines five emerging trends in software engineering—including the product management bottleneck, higher-level code interaction, and reduced technical debt costs—alongside open questions about team structure, curriculum, competitive advantage, and agent-driven workflows. The commentary frames these themes around DeepLearning.AI's upcoming AI Developer Conference on April 28-29 in San Francisco.
Meta Pivots to Closed Weights with Muse Spark; The Batch Issue 349 Roundup
Meta introduced Muse Spark, its first AI model in roughly a year and the first product from its Superintelligence Labs, marking a pivot away from its open-weights strategy toward a closed model. Muse Spark is a natively multimodal reasoning model supporting tool use and multi-agent orchestration, with three reasoning modes and a novel 'thought compression' post-training technique using RL to penalize excessive reasoning tokens. The model ranks fourth on the Artificial Analysis Intelligence Index and matches Llama 4 Maverick's capabilities with over an order of magnitude less training compute, though it trails in coding and agentic benchmarks. The issue also covers broader industry themes including AI-native software engineering team structures, big pharma AI adoption, and regulatory developments.
AI-Native Software Development Needs Generalists
Andrew Ng argues that agentic coding tools are reshaping software team structures by accelerating code production so dramatically that product management, design, marketing, and legal review become the new bottlenecks. He contends that the fastest-moving teams are small (2–10 people), co-located, and composed of generalists who can span engineering, product, and other functions. The piece frames this as a structural shift away from large specialist teams toward individuals who combine deep skills with cross-functional breadth.
GLM-5.1 Open-Weights Model Targets Long-Running Agentic Tasks; Andrew Ng on Coding Agent Acceleration by Software Domain
Z.ai released GLM-5.1, an open-weights mixture-of-experts LLM (754B total / 40B active parameters) designed for sustained agentic coding tasks lasting up to eight hours, featuring iterative planning-execution-evaluation loops with thousands of tool calls. The model claims top open-weights performance on Artificial Analysis Intelligence Index and SWE-Bench Pro, available under MIT license via HuggingFace. The accompanying editorial by Andrew Ng offers a tiered framework for how much coding agents accelerate different software work categories—frontend most, then backend, infrastructure, and research least—with practical implications for team organization. A secondary item references data-center opposition and LLM helpfulness failure modes.
Coding Agents Accelerate Some Software Tasks More Than Others
Andrew Ng offers a practitioner framework ranking how much coding agents accelerate different software work: frontend development benefits most (agents close the loop via browser feedback), followed by backend, infrastructure, and research in decreasing order. Backend work still requires skilled developers to handle corner cases and security; infrastructure decisions remain largely human-driven due to complex tradeoffs and limited LLM knowledge in that domain; research is least accelerated because ideation and hypothesis iteration are not primarily coding tasks. The commentary is aimed at helping engineering managers set realistic expectations and organize teams accordingly.
Insurance Companies Carve Out AI Risk Exceptions; GPT-Rosalind, Claude Design, and Agentic Retail Deployments Highlighted
Major insurers including Berkshire Hathaway units, Travelers Group, and Chubb are excluding or restricting AI-related liability coverage, signaling growing concern over hard-to-model AI-driven claims. OpenAI introduced GPT-Rosalind, a domain-specific LLM fine-tuned for life sciences workflows, while Anthropic launched Claude Design for visual asset generation targeting non-designers. Additional items cover an AI-run San Francisco retail store exposing agentic system limitations, Wall Street banks cutting junior roles via AI deployment, and Anthropic's continued engagement with the Trump administration despite prior Pentagon restrictions.
GPT-5.5 Outperforms Benchmarks but Leads in Hallucination Rate; Kimi K2.6 Tops Open LLMs
GPT-5.5, OpenAI's latest closed vision-language model built for agentic coding and computer use, tops the Artificial Analysis Intelligence Index and ARC-AGI-2 benchmarks but exhibits a significantly higher hallucination rate (85.53%) compared to Claude Opus 4.7 (36.18%) and Gemini 3.1 Pro Preview (49.87%) on the AA-Omniscience benchmark. GPT-5.5 Pro processes reasoning tokens in parallel during inference, and pricing is roughly double GPT-5.4 rates. The model ranks lower on subjective Arena.ai leaderboards, where Claude Opus models dominate. The issue also notes Kimi K2.6 leading open-weight LLMs, though details on that item are truncated.
Andrew Ng Argues AI Will Not Destroy the Job Market
Andrew Ng's weekly letter pushes back on the 'AI jobpocalypse' narrative, arguing that net job creation from AI will exceed job destruction, consistent with historical technology waves. He attributes the doom narrative to incentives of frontier labs, AI SaaS companies anchoring pricing to salaries, and businesses obscuring pandemic-era overhiring. He notes U.S. unemployment remains at 4.3% and software engineering hiring is still strong despite AI coding tools, and predicts an 'AI jobapalooza' of new roles instead.
Forward Deployed Engineers as an Early Wave in AI Engineering Role Specialization
Andrew Ng argues that the current vogue for AI Forward Deployed Engineers (FDEs), driven by OpenAI and Anthropic embedding engineers within client organizations, is an early indicator of broader role specialization in AI engineering. He contends that internal AI Engineer hiring will vastly outnumber FDE placements, and that vendor lock-in concerns limit FDE appeal. Ng predicts the generalist AI Engineer role will fragment over the coming decade into specialized tracks such as LLMOps, Evals Engineers, and AI Data Engineers, analogous to how software engineering split into frontend, backend, devops, and other disciplines.
Hermes Agent Challenges OpenClaw on Token Usage Leaderboard; Agent Self-Improvement Highlighted
Hermes Agent, an open-source AI agent from Nous Research launched in February 2026, has surpassed OpenClaw on OpenRouter's daily token consumption leaderboard. Hermes Agent differentiates itself through a memory architecture and automatic skill-building capability using the SKILL.md format, enabling self-improvement as a core agentic feature. It supports local and cloud deployment, integrates with ~20 messaging services, and works with a wide variety of LLMs via the Agent Communication Protocol. The piece also covers Andrew Ng's commentary on Harvard's grade-capping policy, which is tangential to AI/ML.
DeepLearning.AI Launches AI Andrew: A Personality-Shaped AI Companion Built on Agentic Harness
Andrew Ng's team at DeepLearning.AI has released 'AI Andrew,' an AI companion designed to emulate Ng's communication style and personality for conversations about AI, careers, and learning. The system uses an agentic harness combining RAG, small and large models, guardrails, short- and long-term memory, and offline agentic loops that automatically propose system improvements. The team employed iterative error analysis to close the gap between AI Andrew's outputs and Ng's actual communication style, though acknowledged remaining issues including hallucinations. The product targets people seeking guidance on AI concepts, career decisions, and project ideas.
Abeba Birhane on Bias in Web-Scraped Training Datasets
Researcher Abeba Birhane examines how large-scale web-scraped datasets used to train trillion-parameter NLP and vision models propagate bias and antisocial content. The commentary highlights that performance gains in deep neural networks come alongside inherited societal biases from web training data. Two posts from The Batch summarize her work on cleaning up web datasets and the specific mechanisms by which NLP models absorb web-sourced biases.