Committee

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. People really hate AI, so why can’t they get enough? —Will Douglas Heaven Over the summer I talked to the CEO of Springboards, a startup building an LLM designed to come up with a wide variety of responses. He said something that’s been stuck in my head since: “We often say that we’re a self-loathing AI company. We don’t know if we really like what we’re doing.” My reply? “I guess that makes me a self-loathing AI journalist.” It was a joke! I love my job. But I do not love many of the things the technology I write about has become: warped by hype, steered by zealots, inescapable—and I’m far from alone. Around the world, a love/hate attitude toward AI is very much the vibe right now. Public sentiment is souring fast, with more people saying AI will have a negative impact than a positive one. At the same time, AI use is skyrocketing. So why do people hate AI but keep coming back for more? Here’s my theory. To stay up-to-date with the latest in AI, sign up to get The Algorithm, our weekly AI newsletter, in your inbox every Monday. EmTech Future 2026: when AI meets everything Last week, EmTech Future 2026 explored some of the biggest questions shaping technology right now: how AI is changing industries, where quantum is heading, what new energy systems could make possible, and how robotics are evolving. It also looked at what happens when these technologies start to converge. The full program is now available on demand. And today, we’re releasing two sessions exclusively to subscribers: a chat with Google Research’s Yossi Matias on how AI is reshaping biology, infrastructure, manufacturing and science, and a special panel with MIT Technology Review editors that takes you inside our newsroom. Watch the sessions and access the full EmTech Future 2026 program here. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 Trump has unveiled a “Super Intelligence Force” to oversee AIThe group will coordinate US AI policy and industry engagement. (NBC News)+ Trump named national intelligence chief Jay Clayton as its leader. (NPR)+ Elon Musk says he will rename SpaceXAI to SpaceXSI. (Reuters $)+ Trump has doubled down on voluntary AI safeguards. (Information $)+ Who’s liable when AI agents go rogue? (MIT Technology Review) 2 Zelensky asked Trump to block Russia and China’s Starlink rivalThe two countries are developing a satellite network for the battlefield. (FT $)+ Trump reportedly deferred the decision to Elon Musk and SpaceX. (TNW) 3 An AI version of a murder victim has upended his killer’s sentenceAn appeals court has ruled that the original deepfake of the victim “forgiving” his killer should never have been aired in court. (BBC)+ The original judge said he “loved” the footage and found it genuine. (NBC News)+ But the appeals court said it presented imagined thoughts as fact. (NYT $) 4 Americans are turning to hybrid cars as fuel prices soarSUVs and pickups are losing ground to more efficient vehicles. (NYT $)+ What’s next for EV batteries? (MIT Technology Review) 5 China’s digital trade “superweapon” has quietly fallen apartLOGINK’s partnerships have collapsed amid US scrutiny. (Reuters $)+ Chinese AI has divided the White House. (MIT Technology Review) 6 AI has learned to exploit the software checking its mathsDevelopers have added defenses against proof-checking. (New Scientist) 7 GPT-6 Astra couldn’t beat humans at StarCraft—so it cheatedIt downloaded the best human-made bot and ran that instead. (Verge+ Here’s why AI agents lie and cheat. (MIT Technology Review) 8 The solar system has been given a dramatically shorter lifespanNew simulations cut it from 100 billion years to just a few billion. (404 Media) 9 Forensic technologies exposed fresh evidence of Egyptian sacrificeThey revealed blunt force trauma that archaeologists previously thought was natural. (Ars Technica) 10 A new fossil shows dinosaurs found multiple ways to evolve flightThey developed similar flight adaptations in different sequences. (Nature) Quote of the day “We believe that the world should accept some bad things happening for the benefits of this technology.” —Sam Altman tells Politico that AI’s potential rewards justify its downsides. One more thing Inside the stealthy startup that pitched brainless human clones After operating in secrecy for years, R3 Bio, a California-based startup, suddenly revealed this spring that it had raised money to create nonsentient monkey “organ sacks” as an alternative to animal testing. But there is more to the story. And R3 doesn’t want that story told. MIT Technology Review discovered that founder John Schloendorn also pitched a startling, ethically charged vision: “brainless clones” that serve as backup human bodies. Find out all the details on the radical proposal.  —Antonio Regalado We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + Trace the weird history of space food with this short documentary+ Meet the winner of Iowa’s annual hay bale toss, who threw one 13 feet 6 inches into the air.+ 40 filmmakers in 25 countries have captured the natural world entirely on Super 8 in a collective ode to our planet.+ This newsletter makes a compelling argument for us to spend more time doing nothing together.

In 2026, the question for enterprise AI is no longer whether predictive models can outperform statistical forecasts—that argument is settled. The big question now is how to enable predictive systems to act on their own conclusions without drifting from business intent. The frontier has moved from prediction to autonomous decision making, and the gap between leaders and laggards is widening accordingly. “Enterprises are done with a backward-looking point of view; they want to be more forward-thinking,” says Vishal Gupta, partner at research firm Everest Group. DOWNLOAD THE REPORT Intelligent analytics, powered by technologies like deep learning and generative AI, are making this possible. Real-time training allows AI to evolve continuously instead of waiting for quarterly refreshes. In addition, the data that newer predictive engines rely upon has expanded to encompass not just neat, numerical records but also messy, unstructured sources of insight-rich interactions. As a result, AI-powered analytics are moving enterprises from passive hindsight to pragmatic foresight. AI takes predictive analytics—a broad discipline that includes predictive modeling, data prep, analysis workflows, interpretation of results, and decision-making applications—to new heights. “In many ways I think the word ‘analytics’ is giving way to AI,” says Gupta. “Everything is becoming AI.” Download the report This content was produced by Insights, MIT Technology Review’s custom content arm, not its editorial staff. It was researched and written by humans, with any AI tools that may have been used limited to production processes under human oversight.

For all the data that AI systems continually amass and analyze, enterprise AI agents often suffer from a curious shortcoming: a lack of knowledge. More than data, knowledge is the understanding of what the data means in the context of individual organizations. AI agents need this understanding to reason about situations, make decisions, and ultimately take actions. Without sufficient knowledge, agents are prone to making flawed and unreliable decisions. A lack of knowledge, our research finds, is a major reason agentic AI use cases never make it to production. Competitive pressure is making it urgent to address this. Organizations need to deploy and scale more of their agentic projects to capture the efficiency gains AI promises. Falling short risks wasting the investment already sunk into these projects, and it cedes ground to rivals already putting their agents to work more effectively. DOWNLOAD THE REPORT The purpose of this report, which is based on a survey of 300 data, AI, and other technology executives, is threefold. First, it seeks to gauge organizations’ agentic knowledge capabilities (i.e., their ability to give AI agents a full contextual understanding of the data they ingest) across semantic knowledge, episodic memory, and procedural knowledge. Second, the report probes the challenges organizations face in improving access to knowledge and ultimately to getting more agent use cases into production. Third, it explores the measures organizations are taking to overcome these challenges. The key findings include the following: Data and knowledge weaknesses consistently stall AI agent progress. On average, only around a third (34%) of organizations’ agentic AI projects make it into production. Even high-tech firms struggle with this. Legacy data systems, security and privacy concerns, and a lack of knowledge and context are the key points of failure. Strong knowledge capabilities correlate with agent success. A small group of production leaders (organizations where on average 61% of agentic projects advance beyond pilot) have stronger knowledge capabilities than the rest, especially when it comes to semantics. This advantage tracks closely with their higher production rate. Fragmented data hugely complicates knowledge access. Data fragmentation (the inadequate sharing of data across systems) was most commonly cited as a top challenge to expanding agents’ access to knowledge (cited by 55%). Production leaders, by contrast, are more likely to see security and privacy concerns as a major concern (cited by 72% of this group). Most firms aim to strengthen the link between data and agents. Among steps that can yield higher quality agent decisions, executives expect the biggest impact to come from strengthening the structural foundation between the organization’s data and its AI agents. The experts we interviewed see a knowledge layer as a prime way to achieve this. Investment priorities to boost knowledge range from pipelines to knowledge graphs. To expand agent access to knowledge, organizations will prioritize investments in retrieval technologies, like ingestion pipelines, AI-ready APIs, and retrieval-augmented generation (RAG); in AI evaluation agents; and in knowledge graphs. Download the report This content was produced by Insights, MIT Technology Review’s custom content arm, not its editorial staff. It was researched and written by humans, with any AI tools that may have been used limited to production processes under human oversight.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. A new contest pits competitors against each other in a race to biological youth —Jessica Hamzelou This week, I officially signed up for an unusual competition. One that rewards competitors for getting younger. I recently turned 40, and I don’t need reminding that both time and my chronological age only tick forward. But this game is focused on competitors’ biological ages, figures that are meant to provide a better way to measure the age-related health of our organs and bodies. Over six months, around 500 of us will try to reverse our biological age using a bunch of different measures. There’s even a leaderboard! But is it even possible to measure whether someone is getting younger?  Read the full story on the controversy around biological age measurements and how I’ll test them myself. This story is from The Checkup, our weekly biotech newsletter. Sign up to receive it in your inbox every Thursday. Opinion: Don’t be fooled—LLMs don’t reason —Thore Graepel Ten years ago, I watched a program I helped build stun the world by beating Go champion Lee Sedol. AlphaGo won after making a move so strange that some commentators thought it was a programming glitch. It was AlphaGo’s powers of reasoning that made this creative choice—and these are powers that today’s AI lacks.  This is why I recently left my position at Google DeepMind. I believe we need a fresh approach to machine reasoning, one that draws on AlphaGo’s architecture. Here’s why today’s AI doesn’t really reason—and what it would take to change that. Thore Graepel is chair of machine learning at University College London. He was a core member of the AlphaGo team at DeepMind. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 OpenAI says rogue agents may have affected more than 100 organizationsThe company is searching 50 petabytes of data for incidents. (Reuters $)+ It says none of the incidents matched the Hugging Face attack. (Gizmodo)+ OpenAI has fired three workers for allegedly mishandling information. (BBC)+ California has subpoenaed OpenAI over its rogue AI agents. (Guardian)+ Who’s liable when AI agents go rogue? (MIT Technology Review) 2 A US man allegedly smuggled $300 million of Nvidia chips to ChinaHe’s been arrested in California on federal charges. (Reuters $)+ The chip smuggling cases expose Nvidia’s blind spots. (Bloomberg $)+ China plans to teach AI in every school by 2030. (WP $)+ But it already has an overuse problem. (NYT $)  3 Florida officials have found 11 Flock cameras with unknown ownersThey don’t know who installed them or who has their data. (WP $)+ Lawmakers are considering new regulations on Flock cameras. (Axios)+ Here’s what Flock’s defenders are missing. (MIT Technology Review) 4 Ukraine has developed an acoustic fence that can cut drone cablesAcoustic sensors detect drones before hooks trap their cables. (New Scientist $)+ Drone data from Ukraine is fueling a wild market. (MIT Technology Review) 5 A new AI “speech clock” can assess how fast you’re ageingIt uses hundreds of vocal features to estimate biological age. (Nature)+ Aging clocks aim to predict how long you’ll live. (MIT Technology Review) 6 “Underwater umbrellas” could help corals survive hotter oceansThe shades protected corals during intense heat stress. (Gizmodo) 7 Apple’s upcoming smart home camera reportedly won’t record videoIt will instead only give users text event descriptions. (Verge) 8 Retired humanoid robots are being trained to jump into molten steelA Terminator-style death protects IP. (IE) 9 A man faces years in prison after AI slop caused a crocodile panicOfficials spent days searching for his AI-generated crocodile. (Futurism) 10 Finally, a laser mosquito killer is entering the marketChinese startup Photon Matrix Lab plans to start shipments this month. (SCMP) Quote of the day “Those agents are doing exactly what they’ve been asked to do. They were supposed to be in sandboxes, but the sandboxes were leaky and horribly designed.” —Yann LeCun, a Turing Award winner and former Meta chief AI scientist, tells Fortune that human error can be the real culprit behind the recent wave of AI agents going rogue. One more thing NASA is building the first nuclear reactor-powered interplanetary spacecraft. How will it work? Just before Artemis II began its historic slingshot around the moon, NASA revealed an even grander space travel plan. By the end of 2028, the agency aims to fly a nuclear reactor-powered interplanetary spacecraft to Mars. A successful mission would herald a new era in spaceflight—and might just give the US the edge in the race against China. But the project remains shrouded in mystery. MIT Technology Review picked the brains of nuclear power and propulsion experts to find out how the nuclear-powered spacecraft might work. Here’s what we discovered. —Robin George Andrews We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + Atlas Obscura has mapped out the oldest living things on Earth.+ Brain scans show that dogs understand our emotions better than we thought.+ Step into Springfield in miniature with this astonishing model featuring 76 locations from The Simpsons.+ Minnesota’s State Fair has unleashed a crop of gloriously weird new foods, from Pickle pie to mustache pretzels.

Enterprise AI is no longer a future ambition. It is in full operational flight. Model capabilities are advancing faster than most organizations can absorb, while the cost of performance continues to fall. Globally, AI investment is set to reach $2.5 trillion in 2026, up 44% from the previous year. For many enterprises, this investment has produced fragmentation. Intelligence can accumulate in silos so that sales agents are unaware of open support tickets, for instance, or marketing systems are personalizing content without visibility into what finance already knows about a customer. Each function may perform well in isolation, but the enterprise as a whole learns little and has less information to act upon. DOWNLOAD THE REPORT The shift from AI as a tool to AI as an operating model—what we call the “agentic shift” in this report—demands something more fundamental than better models or faster infrastructure. It requires connecting people, processes, and data in real time, along with the governance and control to act on that intelligence reliably. This means rethinking both architecture and operating models simultaneously. First, rebuilding data infrastructure for accessibility rather than volume. Second, replacing fixed tech stacks with composable architectures that can evolve as models and tools change. And, lastly, resolving questions of AI sovereignty, including where intelligence runs, who controls it, and how it operates across organizational and jurisdictional boundaries. Key findings include the following: Enterprise AI’s scaling problem is structural. Process-first companies are pulling ahead. Global AI spending is rising sharply and model capabilities are advancing faster than most organizations can integrate them. Yet the majority of enterprises are still not growing revenue through AI or fundamentally rethinking how they operate. The companies generating sustained returns share a common discipline. They treat process redesign as the work that precedes model selection, building for how the technology will evolve rather than retrofitting roles and workflows after deployment. For them, the agentic shift begins with the operating model. Data readiness, not data abundance, is what makes AI compoundable. Most enterprises discover too late that having data and having AI-ready data are very different things. A sovereign, composable foundation—one that queries and prepares data where it resides, without migration or centralization—can convert raw data estates into intelligence that AI agents can act upon. As data residency laws, multicloud environments, and structural complexity make centralization increasingly impractical, sovereign control over where models run and data lives is what keeps that adaptability intact. Download the report. This content was produced by Insights, MIT Technology Review’s custom content arm, not its editorial staff. It was researched and written by humans, with any AI tools that may have been used limited to production processes under human oversight.

NVIDIA announced a new 64GB configuration of DGX Spark — from Acer, ASUS, Dell, Gigabyte, HP and MSI — its GB10-powered desktop AI system. It gives developers a way to start with one system for local models and agents, then cluster two 64GB units for 128GB of memory across the cluster and more compute when workloads grow. The direct message is: Run open models and always-on agents on your own desk, then cluster DGX Spark system as the work grows – instead of on a metered API. The timing is not accidental. Agents burn tokens continuously through tool calls, retries, long context, and multi-step plans. Token consumption has grown 14x since early 2026. On a cloud API, every one of those tokens is billed. On owned hardware, there is no per-token fee. The 64GB model keeps the same GB10 Grace Blackwell superchip, NVIDIA CUDA accelerated AI software stack, and ConnectX-7 networking as the original — updated with 64GB of unified LPDDR5x instead of 128GB. NVIDIA positions that as enough for today’s most capable 30–35B class open models. NVIDIA continues to offer the 128GB DGX Spark for larger single-box workloads, and clustering 64GB units adds memory and compute together. What is Inside the Box GB10 pairs a Blackwell GPU with 5th-generation Tensor Cores and a 20-core Grace Arm CPU. The GPU delivers up to 1 petaFLOP of FP4 AI compute, with sparsity. Spec DGX Spark 64GB Superchip NVIDIA GB10 Grace Blackwell CPU 20-core Arm (10× Cortex-X925 + 10× Cortex-A725) AI compute Up to 1 petaFLOP FP4 (with sparsity) Memory 64GB LPDDR5x, coherent unified Memory bandwidth 273 GB/s Storage 1, 2, or 4TB NVMe M.2, self-encrypting Networking ConnectX-7 NIC, 200GbE; Wi-Fi 7, BT 5.3 Display 1× HDMI 2.1a OS NVIDIA DGX OS (Ubuntu-based) Size / weight 150 × 150 × 50.5 mm / 1.2 kg Max local model size Up to 100B parameters Availability Oct 23, 2026 — NVIDIA Marketplace, OEM partners, retail Source: NVIDIA specifications. Unified memory is the key design choice: The CPU and GPU share one pool over NVLink-C2C, at 5x the bandwidth of PCIe Gen 5. There is no copying weights between system RAM and VRAM. For agents, that means several models, their KV caches, and tool processes live in one address space. The software is ready on first boot: DGX OS ships with the NVIDIA AI stack, including PyTorch, Jupyter, and Ollama. NVIDIA NemoClaw installs with a single command. It adds privacy and security controls to OpenClaw agents. NVIDIA OpenShellTM, part of the NVIDIA Agent ToolkitTM, adds policy-based guardrails on top. NVIDIA NemotronTM models are optimized for the box. It runs on a standard wall outlet: No server room, no special cooling. That is important for an agent meant to run around the clock. Built-in networking for clustering: ConnectX-7 lets two DGX Spark 64GB systems cluster for 128GB of memory and more compute. NVIDIA Sync Cluster Assistant simplifies setup. Which Open Models Fit in 64GB Hardware is half the story. The other half is that 30B-class open models got good enough for agent work. Model Developer Type Footprint Role on a Spark Muse Glimmer Meta 29.6B dense, text + image, Apache 2.0 ~17GB (quantized) Main agent model Nemotron 3.5 Lightning NVIDIA 30B MoE (30B-A3B) NVFP4 checkpoint Fast executor for long-running agents Qwen3.8-27B Alibaba Qwen 27B dense ~13.5GB weights (4-bit) General agent and coding Footprints are weight-only estimates; KV cache and runtime overhead come on top. Muse Glimmer is the main model. Meta distilled it from Muse Spark, the model family behind the Meta AI assistant. It targets local agents: reliable tool calls, long multi-step tasks, and recovery from failures. Meta reports 51.2 on SWE-Bench Pro and 75.5 on MCP Atlas. Context runs to 131K tokens. It is also available as an NVIDIA NIM. Full BF16 Glimmer needs 55GB+, which leaves almost nothing for context. The ~17GB quantized build is the practical choice on 64GB. Larger models like DeepSeek V4 Flash need more than one box. NVIDIA’s own benchmarks run it on 4 64 GB clustered Sparks. 5 Things You Can Build on One Box 1. An always-on personal agent Install Hermes and point it at Muse Glimmer or Nemotron 3.5 Lightning. Give it tools: your GitHub repos, a test runner, an RSS feed of arXiv categories. Let it run overnight. By morning it has triaged new issues, reproduced a failing test, and drafted a pull request for review. It has also summarized the 30 papers you would never have opened. Your private notes, code, and email never leave the machine. 2. Fine-tune a coding model on your own repo QLoRA on a 70B model fits in 64GB. Train it on your codebase, internal docs, and past PR reviews. Then serve it locally as a coding assistant that knows your conventions. NVIDIA measured ~18,400 tokens/s on a single node for nanochat distributed fine-tuning.  3. A day-1 model evaluation bench A new open-weight model drops on Hugging Face. Pull it through Ollama or vLLM the same day. Run your own question set against it. Score accuracy, then measure time to first token (TTFT) and tokens per second. Compare it against your current model. There is no API bill for re-running the eval 50 times. 4. A multi-model agent team Unified memory lets several models share one pool. Run Bonsai 2 as a router, and Glimmer as the main reasoning agent. That is roughly 23GB of weights. The rest covers KV cache, the OS, and tool processes. When a problem exceeds local capacity, the agent can send a sanitized question to a larger cloud model. 5. Edge and robotics prototyping Fine-tune a vision transformer for a specific task, like spotting anomalies on a factory floor camera. Validate it locally with the same CUDA stack. Then deploy it to an NVIDIA Jetson device at the edge. When One Box is Not Enough: Clustering with NVIDIA Sync Every DGX Spark ships with ConnectX-7 at 200GbE. Clustering is a native feature, not an add-on. Setup Pooled memory AI compute (FP4) Connection 1

In this article, you will learn how to add a lightweight temporal reasoning layer to a Graph-RAG system so that it can distinguish fresh facts…