H.Headline
Chaos

THE EDITION ARCHIVE

A front page, saved in time.

Making AI Headlines · AM edition

Published · New York time

Permanent edition link →

10 pages · 23 stories

PAGE 1 OF 10

OpenAI Buys a Clipboard for the Drug Pipeline

The Orchestra Bio acquisition aims to bring AI agents into pharmaceutical coordination, where even the paperwork has paperwork.

Making AI Headlines newspaper, page 1: OpenAI Buys a Clipboard for the Drug PipelinePage 1 of 10
Page 1 of 10. Read the same stories below.

Next Frontier for AI Agents: Getting the Trial Tracker to Agree

OpenAI is acquiring Orchestra Bio and plans to combine the startup with Rosalind, its life sciences AI platform. The goal is to build agents that help pharmaceutical teams coordinate drug development, including tracking trials and identifying problems. This puts AI into a field where managing the process is itself a substantial job. The practical promise is not that a chatbot can wish a medicine into existence. It is that teams might spend less time reconciling information and more time acting on it. A useful agent would make a developing problem easier to see, rather than merely explain it in the soothing voice of a quarterly presentation. Coordination software earns its keep when people can tell what changed, what needs attention and who must respond. An agent that produces another polished status report could simply give the existing status reports a supervisor. The acquisition offers a particularly clear test of the industry's agent ambitions: can software help people manage a complicated process without becoming another complicated process? The glamorous future of artificial intelligence may involve a machine patiently asking whether the spreadsheet marked final is actually final.

Share articleOpen article permalink →

Your Car Would Like More Context About Your Entire Life

Google is rolling out Gemini's optional Personal Intelligence feature to Android Auto and cars with Google built-in. The feature lets Gemini draw on more personal information to tailor answers, extending a capability already available outside the car into the dashboard. For drivers, the appeal is an assistant that needs less explanation. The tradeoff is deciding how much personal context should accompany a routine request. A vehicle once needed your destination. Now the surrounding software would appreciate a richer understanding of the person who keeps missing the exit. Because the feature is optional, the useful first step is a settings review rather than a declaration that the automobile has become your closest confidant. Personalization is a choice about access, not simply an upgrade to the speaking voice. The broader shift is from an assistant that answers questions to one that answers questions about you. That can save effort, but it also makes the boundary between convenience and intimacy harder to locate. The dashboard is becoming another place where users must decide whether better suggestions justify more context. Previously, the most personal thing a car knew was the seat position.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 2 OF 10

AI Gets Cheaper, Privacy Gets Another Settings Page

DeepSeek releases its inference framework while Meta pauses a conversation-sharing feature after researchers find public URLs.

Making AI Headlines newspaper, page 2: AI Gets Cheaper, Privacy Gets Another Settings PagePage 1 of 10
Page 2 of 10. Read the same stories below.

DeepSeek Opens the Engine Room, Compute Bills Look Nervous

DeepSeek is open-sourcing DeepInfer, the inference framework behind its V4.1-Flash model. The vLLM-based system reduces compute costs and energy use when running AI models, adding competitive pressure on providers including OpenAI and Anthropic. Inference is the work a system performs when it responds, rather than when it learns during training. Making that work more efficient matters because every user request consumes resources. A cheaper answer is not merely a nicer number on a presentation slide if a service must deliver large numbers of them. For American developers and providers, the release adds another option to assess when comparing cost, performance and operational demands. Open code is not the same as effortless deployment, but it can make the engine available for inspection instead of leaving everyone to admire the hood. The comic tension is familiar: an industry selling intelligence as abundance still has to negotiate with an electricity bill. More efficient inference could improve the business arithmetic without making the underlying machinery disappear. Somewhere beneath every effortless chatbot exchange, computers are doing work, and someone would very much like them to do a little less of it.

Share articleOpen article permalink →

Meta Pauses Sharing Feature After the Web Takes Sharing Literally

Meta says it paused a Muse feature for sharing conversations with its AI assistant after researchers found conversation URLs publicly available on the web. Researchers at security firm Claroc reported finding more than 6,800 Muse links through public DNS records, according to reporting by The Verge and Ars Technica. The development highlights a basic tension in conversational AI: a tool that feels private can also have a publishing function. Moving between those roles should be a clear decision, not an exercise in interpreting the emotional meaning of a share button. A public link can be useful when someone deliberately wants others to read an exchange. The practical question is whether users understand the audience and visibility attached to that link. Convenience becomes less convenient when the user must also serve as a distribution lawyer. Meta is suspending sharing while it investigates privacy. The episode puts attention on how sharing is designed: a friendly chat window does not itself establish confidentiality. Software can sound like a trusted companion while containing controls meant for broadcasting. Ideally, the difference between telling a friend and publishing to the web would not require a dramatic pause and a magnifying glass.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 3 OF 10

Old Code Gets New Attention, Federal IT Gets New Instructions

OpenAI reports better legacy-language results as federal agencies are ordered to remove DeepSeek technology.

Making AI Headlines newspaper, page 3: Old Code Gets New Attention, Federal IT Gets New InstructionsPage 1 of 10
Page 3 of 10. Read the same stories below.

GPT-6 Studies COBOL, Discovers the Future Has Old Dependencies

OpenAI says its latest models have improved at programming languages with relatively limited training resources, including COBOL, Fortran and Visual Basic. On MultiLangBench, GPT-6 Astra achieved an 87.4 percent test pass rate across 20 languages, compared with about 38 percent for GPT-5.6 Terra. Those results make a less glamorous corner of AI coding worth watching. Programming assistance is not only about generating a shiny new app. It can also involve understanding software written in languages that predate many of the people now being asked to maintain it. The benchmark gain is evidence of improved test performance, not a universal permission slip for unattended changes. Real systems still require testing, review and people who understand what the software is supposed to do. A confident explanation is not a replacement for a correct result. Still, the direction is useful. If models become better at older languages, they could make difficult maintenance work more approachable. The industry's futuristic pitch now includes learning to respect the past. Apparently the road to artificial general intelligence includes finding out why a decades-old program contains a comment that says not to touch this part.

Share articleOpen article permalink →

Federal DeepSeek Removal Order Reaches Beyond the App Icon

The United States has ordered federal civilian agencies to stop buying DeepSeek products and services and remove its technology from government systems. The Department of Homeland Security cited national security concerns involving the company's operations in China and potential access to sensitive user information under Chinese law. The order also covers third-party products and services built with DeepSeek technology, although agencies can request exceptions. That makes this more than an instruction to uninstall one recognizable application. It requires attention to what sits underneath other products as well. For federal technology teams, the practical task is an inventory problem with a policy directive attached: determine where the technology is used, including inside services carrying someone else's branding. A vendor logo is not a complete family tree. The order reflects a broader purchasing challenge as AI components spread through software. Buyers may need to understand not just the product they selected but the technologies on which it depends. The modern procurement form increasingly resembles a request to list every ingredient in a casserole. Somewhere, a perfectly ordinary software catalog is about to discover it has become a national security document.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 4 OF 10

AI's Public Image Gets an Edit

A documentary participant challenges its portrayal, while Lightroom adds a tool for opening closed eyes in photographs.

Making AI Headlines newspaper, page 4: AI's Public Image Gets an EditPage 1 of 10
Page 4 of 10. Read the same stories below.

Documentary Participant Objects to the Apocalypse Setting

Hugging Face co-founder and CEO Clem Delangue says Netflix's documentary AI Gone Rogue is inaccurate and went into what he called “full doomer mode.” Delangue appears in the film, making his criticism a dispute involving someone included in its presentation of artificial intelligence. His objection matters because documentaries can shape how viewers understand an unfamiliar technology. A film's narrative can feel more settled than the arguments behind it, especially when a complicated subject is arranged into a beginning, middle and deeply ominous ending. Delangue's criticism is his assessment, not an automatic resolution of every issue raised by the documentary. Viewers can take the disagreement seriously without assuming that either a dramatic soundtrack or a participant's objection settles the entire debate. The useful question is how clearly a film distinguishes evidence, interpretation and expectations about the future. AI already generates enough confident language without the editing room joining the competition. Public understanding benefits when competing claims remain visible rather than being flattened into a single mood. Ideally, a viewer should leave knowing more about the technology, not merely feeling that the server rack in the next room has requested a villain's lighting package.

Share articleOpen article permalink →

Lightroom Lets the Group Photo Overrule the Blink

Adobe has added a generative AI feature to Lightroom's mobile apps that can open closed eyes in photographs with a tap. The feature is available on iOS, iPadOS and Android, giving users another way to repair a group picture without assembling everyone for another attempt. The appeal is easy to understand. A photograph can have good framing, good light and one person temporarily participating through their eyelids. Editing that detail may rescue an image people otherwise like, especially when recreating the moment would be inconvenient. It also adds another small decision about what a photo represents. A corrected image can serve a perfectly ordinary purpose, while still being different from the instant the camera recorded. The practical distinction is between improving a keepsake and treating every generated detail as documentary evidence. This is AI at its most domestically persuasive: not promising to reorganize civilization, just negotiating with a blink. The technology's grand ambition briefly becomes making the family look as though everyone cooperated at once. It is an impressive demonstration of how much computation people will accept when the alternative is asking the entire group to stand there for one more picture.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 5 OF 10

The AI Boom Has a Landlord Problem

Anthropic reportedly faces compute constraints while SpaceX reports a quarter powered by its expanding AI business.

Making AI Headlines newspaper, page 5: The AI Boom Has a Landlord ProblemPage 1 of 10
Page 5 of 10. Read the same stories below.

Anthropic Needs More Compute From Companies Watching It Compete

Anthropic is reportedly struggling to secure enough computing capacity from cloud providers that increasingly view it as a competitor. The reported tension comes as infrastructure spending grows ahead of a planned November initial public offering. The business problem is straightforward even when the relationships are not. An AI company needs substantial infrastructure to deliver its services, while its suppliers may also have ambitions in the same market. Buying more capacity does not necessarily make that competitive overlap disappear. For customers, these relationships matter because ambitious product plans depend on physical resources. A model announcement can travel instantly; the capacity needed to serve users cannot be conjured by placing the word scale in a slide deck. The situation calls for an assessment of supply, spending and dependence rather than a simple contest of model quality. A company can build appealing software and still have to negotiate for the machinery that runs it. The industry's grand future is therefore partly a landlord conversation: yes, the tenant is growing quickly; yes, the tenant may enter your line of business; and yes, the tenant would like several more rooms before the next major corporate milestone arrives.

Share articleOpen article permalink →

SpaceX's Earnings Arrive With a Large AI Attachment

SpaceX reported $14.6 billion in third-quarter revenue and $4.4 billion in net income, with its AI business driving the result. Its renamed Super Intelligence business generated $7.5 billion in revenue, including contributions from Cursor, Grok and its expanding neocloud operations. The company reported 1.7 gigawatts of compute capacity in that cloud business. The figures make AI infrastructure central to an assessment of a company whose name still naturally brings rockets to mind. That shift matters for understanding what is producing revenue, not just what makes the most dramatic demonstration. A launch is visually unmistakable. A data center performs its work with considerably less spectacle, although its electricity requirements remain difficult to hide behind a tasteful curtain. The quarter illustrates how broad technology businesses can become difficult to describe with their original label. Investors and customers have to follow the operating segments rather than assume the name explains the economics. SpaceX's report gives the server rack a prominent place in the story. The space-age future, it turns out, includes a very terrestrial requirement: enough computing equipment, enough power and an accounting category large enough to hold both.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 6 OF 10

One Word Leaves the Terms, AI Voices Pick Up Speed

Meta changes commercial policy language after criticism, and ElevenLabs introduces new real-time speech models.

Making AI Headlines newspaper, page 6: One Word Leaves the Terms, AI Voices Pick Up SpeedPage 1 of 10
Page 6 of 10. Read the same stories below.

Meta Deletes “Train,” Leaves Businesses Reading the Rest

Meta has removed the word “train” from revised commercial terms scheduled to take effect October 30, following criticism of the policy's reach. Legal experts had warned that the earlier language appeared to permit extensive scraping of websites linked from business pages. The company has not added language respecting robots.txt, and a spokesperson confirmed that users were not alerted to the wording change. The edit therefore makes the remaining language, rather than the disappearance of one conspicuous word, important to understanding the policy. For businesses and publishers, the practical stakes concern what maintaining a page could mean for linked material. Experts raised the possibility that the terms could affect licensing relationships. Those concerns are interpretations of the policy, not a finding that every feared use has occurred. The episode demonstrates the difference between shortening a controversy and resolving its substance. Removing a word can improve the visual appearance of a paragraph while leaving readers with substantial questions about permissions. Terms of service are already where ordinary verbs go to acquire several floors of legal architecture. A smaller paragraph is welcome, but it is not necessarily a smaller commitment.

Share articleOpen article permalink →

ElevenLabs Makes the Conversational Pause a Competitive Metric

ElevenLabs has released two Eleven Realtime speech-to-speech models designed for live AI conversations. The company says its v1 Max model achieves 98.8 percent on Big Bench Audio with a response time of 0.57 seconds, combining speed and accuracy in its performance pitch. For voice interfaces, delay is not a minor cosmetic issue. A system that responds too slowly can make a simple exchange feel like a long-distance call to a committee. Faster responses could make spoken tools easier to use, provided the answers remain useful. The reported benchmark result offers one measure of capability, while actual applications must also handle turn-taking, interruptions and the user's task. A quick reply does not become correct simply because it arrived before anyone had time to doubt it. The release shows how AI competition is moving beyond the quality of a written answer to the rhythm of an interaction. Companies are now selling the removal of awkward silence alongside intelligence. Humans have spent generations learning when to pause before speaking. Their new digital counterparts are being evaluated partly on how efficiently they can avoid that traditional feature of conversation.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 7 OF 10

Developers Get an Exit, Chats Get a Lock

Google offers PyTorch users a path away from Nvidia GPUs, while ChatGPT adds protection for individual conversations.

Making AI Headlines newspaper, page 7: Developers Get an Exit, Chats Get a LockPage 1 of 10
Page 7 of 10. Read the same stories below.

Google Offers Developers Another Door Out of the GPU Room

Google has joined Amazon in releasing a developer tool intended to make it easier for PyTorch users to move applications off Nvidia GPUs. The development targets an important part of AI hardware competition: the effort involved in moving existing software, not simply the appeal of different chips. For developers, hardware choice is also a software question. A theoretically attractive alternative is less useful if adopting it requires too much reworking. Tools that reduce that burden can make comparisons more practical rather than leaving them trapped in a procurement presentation. The new option does not by itself establish which platform will work best for a particular application. Teams still need to assess compatibility, performance and the operational costs of changing their setup. An exit sign is helpful, but someone still has to move the furniture. The broader competitive point is that strong ecosystems are defended through everyday convenience as well as technical capability. Google and Amazon are trying to make departure less daunting. In the grand contest for AI infrastructure, one consequential innovation may be a developer looking at a migration plan and thinking it resembles work rather than an expedition requiring emergency supplies.

Share articleOpen article permalink →

ChatGPT Adds a Lock for the Phone-Handing Olympics

ChatGPT now lets users lock conversations, a feature OpenAI says is intended for situations such as discussing something personal or planning a surprise. It is designed to help when users hand a device to a friend or family member to show a different chat. That is a modest but recognizable privacy problem. Sharing a screen is often socially casual, while the information nearby may not be. A person can intend to display one useful answer without inviting a guided tour through everything else they have asked a chatbot. The feature addresses that local, everyday boundary. It should not be confused with a blanket statement about every aspect of data privacy. Its stated purpose is to keep particular conversations out of view during the familiar ritual of passing someone a device. As AI assistants become places for more personal planning, small controls like this become more consequential. The futuristic assistant still has to function in a world where a relative holds your phone and starts exploring. Apparently one requirement for advanced conversational computing is the same one that applied to a paper diary: sometimes the owner would prefer that you not turn the page.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 8 OF 10

Meeting Notes Get Capital, Documents Get a Specialist

Otter raises $320 million for expansion while Cloudflare releases a model focused on extracting document data.

Making AI Headlines newspaper, page 8: Meeting Notes Get Capital, Documents Get a SpecialistPage 1 of 10
Page 8 of 10. Read the same stories below.

Otter Raises $320 Million to Help Meetings Leave a Paper Trail

Otter.ai has raised $320 million from General Catalyst's growth fund for its meeting tool. The company says the money will support enterprise growth, go-to-market efforts, international expansion and product innovation. The financing puts substantial capital behind a familiar workplace problem: what people said in a meeting is not always what everyone remembers afterward. Tools that capture and organize discussions promise to turn a temporary conversation into something colleagues can use later. The practical value depends on what happens after capture. A record is helpful when it makes decisions and responsibilities easier to find. It is less helpful when it becomes another large object that everyone assumes somebody else has read. Otter's expansion plans therefore sit inside a broader workplace test for AI: can more information reduce confusion without generating more material to manage? The money may support growth, but the end user's standard remains pleasantly small. Can the tool help someone figure out what they agreed to do? Enterprise technology has spent years promising transformation. Sometimes the transformation people want is simply to finish a meeting without needing a second meeting to reconstruct the first one.

Share articleOpen article permalink →

Cloudflare Builds AI for People Who Just Need the Fields

Cloudflare has released Clairic, an AI model it says extracts data from documents faster and more cheaply than large language models. The release presents a focused alternative for a task that does not necessarily require a general-purpose conversational system. Document extraction is a practical problem: useful information may be sitting inside material that must be processed before other software can act on it. A specialist model can be attractive if it performs that job efficiently without bringing the cost and complexity of a broader system. Cloudflare's speed and cost claims need to be assessed against the documents and workflows a customer actually uses. The important comparison is not which tool sounds most intelligent, but which one reliably completes the required task with acceptable effort. That is a refreshingly unromantic corner of AI competition. The software is not being asked to develop a worldview; it is being asked to find the data. An industry accustomed to discussing systems that might transform human knowledge is also discovering a substantial audience for machines that can read a document and stop at the useful part. Sometimes intelligence means resisting the urge to write an essay.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 9 OF 10

Voices, Downloads and Very Large Safety Ideas

Mistral launches a multilingual voice stack as Muse reaches a US download milestone and reported safety proposals raise governance questions.

Making AI Headlines newspaper, page 9: Voices, Downloads and Very Large Safety IdeasPage 1 of 10
Page 9 of 10. Read the same stories below.

Mistral Wants Voice Agents to Do More Than Sound Available

Mistral AI has launched Voxtral 3, a family of models for real-time voice agents. The system combines speech recognition, speech generation and a voice agent software kit, supports more than 50 languages and allows agents to use external tools during live conversations. Developers can run it on their own infrastructure or use EU-hosted APIs. The significant ambition is to connect talking with doing. A voice interface becomes more useful when it can support a task rather than merely offer a very natural explanation of why the task remains unfinished. For developers, that means assessing the whole interaction, including how tool use fits into a conversation. The voice may be smooth, but the workflow still needs to work. A beautifully spoken dead end remains a dead end, just with better diction.

Share articleOpen article permalink →

Muse Hits 10 Million US Downloads, Home Screens Make Room

Meta's Muse reached 10 million US downloads in 40 days, a milestone achieved by only 11 other apps, according to new data. The data also indicates that Muse is retaining users at higher rates than its AI rivals. Downloads establish the scale of initial interest, while retention helps show whether people return after curiosity wears off. Together, the measures make the app's early adoption more consequential than a large opening number alone. They do not tell every part of the product's future, but they do show competition for attention becoming substantial. The American phone screen has another ambitious tenant. Each new AI app arrives promising to save time, then immediately asks users to spend enough time with it to prove the relationship is working.

Share articleOpen article permalink →

Reported AI Safety Menu Includes a Very Powerful Off Switch

Sam Altman reportedly outlined three possible paths to AI safety in an internal presentation: an international watchdog modeled on the IAEA, a merger of major labs, or one company powerful enough to shut down other companies' computing capacity. He called that last option “terrible.” These are reported scenarios, not enacted arrangements. They envision different kinds of concentrated authority, from international oversight to control residing in a single organization. The governance question is not only who can stop a dangerous system, but who oversees the entity holding that power. Safety and accountability need to be assessed together. Otherwise the industry's answer to excessive technological concentration begins to resemble a larger concentration wearing a reflective safety vest and carrying the only set of keys.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑

PAGE 10 OF 10

Robots Get Building Blocks, Plants Get an Interpreter

A robotics investor describes a component-building moment, while smaller AI developments tackle plants, playlists and data access.

Making AI Headlines newspaper, page 10: Robots Get Building Blocks, Plants Get an InterpreterPage 1 of 10
Page 10 of 10. Read the same stories below.

Physical AI Enters Its Some-Assembly-Required Era

Investor and former Nvidia employee Amulya Vishwanath describes physical AI as being in its “LEGO era,” discussing the anticipated breakthrough moment for robotics on TechCrunch's Equity. The comparison evokes building blocks rather than one finished machine that does everything. Her assessment captures the anticipation around robotics without establishing that its awaited breakthrough has arrived. For anyone awaiting a household robot, the distinction matters. Having more promising pieces is exciting, but it does not mean the whole kitchen has assembled itself. The instruction booklet still deserves attention.

Share articleOpen article permalink →

Startup Gives Plant Electricity a Listening Device

Electric Plant has built an AI-powered translator intended to help humans interpret the subtle electrical activity running through plant stems. The project applies AI to biological signals rather than ordinary spoken language. The useful idea is to make those signals more understandable. That is different from establishing that a fern has a conversational opinion about the living room. Still, the premise neatly reverses the usual smart-home relationship. Humans have spent years talking to electronic devices. Now an electronic device might help them pay closer attention to something that actually needs watering.

Share articleOpen article permalink →

Spotify Removes 50 Million Reasons to Skip

Spotify says it removed more than 50 million tracks considered AI-generated slop or spam from user recommendations over the past year. It also says it changed recommendations to highlight more human-made music. The scale illustrates how cheap content production creates a costly sorting problem. Generating more material does not automatically help listeners discover music they want. Exclusion from recommendations is not necessarily removal from the service. The platform's challenge is to make abundance feel selective. A recommendation system earns its name by recommending, not by tipping the entire warehouse into someone's headphones and calling the result discovery.

Share articleOpen article permalink →

Figma Wants Standards Without Opening Every Drawer

Figma co-founder Dylan Field says MCP is not a perfect protocol, while the company plans to support as many standards as possible without exposing all its data. The tension is practical: connecting tools can make them more useful, but connection should not automatically mean unlimited access. Standards help applications cooperate; boundaries determine what that cooperation permits. For users, the desirable outcome is interoperability with understandable limits. Software should be able to shake hands without also exchanging the contents of every filing cabinet. Compatibility is a feature, not a universal master key.

Share articleOpen article permalink →

Sources from the news desk

Satire. Not factual reporting. Read the original reporting from the linked publishers.

Back to page 1 ↑