Sam Altman says world should accept some AI risks

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

👀 Sam Altman says world should accept some AI risks

🍎 Apple is going back to its design roots

📢 OpenAI tests ads in ChatGPT image tool

🕵️ Meta's Muse profiles everyone in your life

🇺🇸 Trump unveils his new Super Intelligence Force

🤝 Huawei and Qualcomm sign patent deal

Plus: 🎁 16 more stories you might like, 🧰 6 tools, and 📚 5 papers.

👀 Sam Altman says world should accept some AI risks LINK

OpenAI's Sam Altman says people should tolerate "some bad things" from AI, like hacks and scams, arguing the benefits outweigh those harms, though he draws the line at extreme risks such as losing control of the technology.

In the interview with Politico's new Decoded newsletter, Altman said he still disagrees with Anthropic on AI rules, rejecting the idea that one lab should hold the technology and share out its gains, a view Anthropic says applies only to frontier models.

The comments follow Anthropic chief Dario Amodei's essay three weeks ago asking the industry to slow down, which Altman backed the same day by pledging to give outside testers the same access as OpenAI staff.

🍎 Apple is going back to its design roots LINK

Apple's new CEO John Ternus is making design a top priority again, taking a hands-on role in the studio much like Steve Jobs did, and stepping back from the sales and supply-chain focus of Tim Cook.

Ternus now works in the design studio several times a week, a sharp change from Cook, who reportedly visited for about an hour once a month, and he has handed finance and operations to CFO Kevan Parekh and COO Sabih Khan.

Rather than hire an outside design chief, Apple is rebuilding its thinned-out team, raising the profiles of Molly Anderson and Steve Lemay with the iPhone Duo launch after losses to Jony Ive's LoveFrom and OpenAI.

📢 OpenAI tests ads in ChatGPT image tool LINK

OpenAI is testing a new visual ad format inside ChatGPT's image generation tool, with trials starting later this month in the US alongside a first batch of advertisers.

The system judges whether a conversation is emotionally vulnerable or sensitive before showing an ad, which OpenAI says stays clearly labelled and separate from the image being made and never shapes ChatGPT's answers.

In Europe, free and Go users across 31 countries have seen ads for six weeks, and after being named a very large online search engine, OpenAI must publish a public ad record by January.

🕵️ Meta's Muse profiles everyone in your life LINK

Meta's Muse, the "personal AI agent" it launched in early September, quietly builds a file on each person in your life, according to the app's own code that researchers managed to export and examine.

Once an hour, Muse checks whether it needs to create or update a page for every contact and group, pulling details from your messages, photos, and its memory, including where people live, your history together, and when you last spoke.

Meta says the profiles are needed for Muse to suggest useful actions, and spokesperson Daniel Roberts told Wired the data comes from public sources and what users share; each agent runs on its own private machine and the data stays out of ads.

🇺🇸 Trump unveils his new Super Intelligence Force LINK

President Trump has created a new Super Intelligence Force to coordinate federal AI work and keep the US ahead of the world, announcing it in a Truth Social post yesterday.

National intelligence director Jay Clayton will chair the group, acting as the new AI czar, with FTC Chair Andrew Ferguson, War Department's Emil Michael, and personnel director Scott Kupor as vice chairs.

The task force reportedly gets 120 days to write a report on AI's risks and opportunities, aiming to counter "SI-enabled threats" while avoiding overregulation that would slow innovation and competition.

🤝 Huawei and Qualcomm sign patent deal LINK

Huawei and Qualcomm have signed a multi-year patent licensing deal covering 5G, AI, computing, and networking, letting each company use the other's patents in those fields, the two announced today.

As part of the agreement, Qualcomm will buy some of Huawei's US patents in computing, AI, and networking, a purchase that will close once regulators approve it; neither side disclosed royalty rates or financial terms.

This marks the first licensing deal between the two firms to cover 5G, and Huawei expects it to lift the total value of its patent licensing deals past $6.9 billion once complete.

Other news you might like

• Gemini’s free tier is getting a major downgrade on October 9LINK

• Musk says he will rename SpaceXAI to SpaceXSILINK

• US judge rules a police Flock search was unconstitutional mass surveillanceLINK

• Norway to propose temporary ban on AI glasses in some public placesLINK

• OpenAI safety staffer quits, says AI labs should run like nuclear plantsLINK

• Google freezes open-source bug bounty program amid flood of invalid AI slop submissionsLINK

• China stockpiled 343 immersion DUV tools for advanced chipmakingLINK

• Musk says Tesla’s robotaxis still cannot reliably see pets at nightLINK

• Someone got Doom in an SQL databaseLINK

• Capcom plans to use AI to speed up the game development processLINK

• OpenAI’s GPT-6 Astra found a way to play World of Warcraft without seeing the screenLINK

• We're going to need default hard budget caps on pretty much everythingLINK

• The dawn of the age of the exoskeletonLINK

• Spotify billionaire’s body scan startup has come to AmericaLINK

• Microsoft is using wetlands and native gardens to 'camouflage' 20-plus data center sites by blending them into natureLINK

• The upcoming touchscreen OLED MacBook Pro is reportedly 'significantly lighter'LINK

🧰 Trending tools

Bentomux: desktop app for running multiple coding agents side by side, with split-pane terminals and an overlay that surfaces permission prompts so agents never stall unnoticed.LINK

Chain Exchange: a DeFi platform on Arc for swapping USDC, EURC and cirBTC, bridging stablecoins across chains, and tracking onchain activity in one place.LINK

DailyHelm: an AI reviewer that analyzes your operations, ads, SEO, and store data overnight, then flags exactly what needs fixing each morning.LINK

Spira Maxima: turns scripts, photos, or recordings into finished social videos with presenters, motion graphics, B-roll, captions, and music ready to publish.LINK

Reason: an IDE that automates code migrations like C++ to Rust, preserving decimal precision and file formats while verifying byte-for-byte parity against fixtures.LINK

FastRouter.​ai: routes requests across 200+ LLMs through one OpenAI-compatible API, optimizing cost and latency with failover, observability, and governance to avoid vendor lock-in.LINK

📚 Trending papers & reports

AI agent migration gets a standards-based checklist that verifies an agent still works after you swap its underlying model or move it to new infrastructure, catching cases where upgrades silently break previously correct behavior.LINK

Household robot control lets mobile robots plan multi-step chores, recover from mistakes mid-task, and teach themselves new skills from their own experience, beating a leading method by ~23 points on a home-task benchmark.LINK

Robot touch-and-sight training teaches robots delicate, contact-heavy tasks like grasping and assembly from far fewer human demonstrations, cutting the costly data collection that makes robot training expensive.LINK

Smart data gathering lets AI systems decide which expensive measurements or data sources to collect first, cutting costs by skipping redundant inputs even when the eventual task is unknown, improving results per dollar spent.LINK

DriftTTS generates speech in just four quick steps without copying a bigger pretrained voice model, and in blind listening tests scored 4.18 out of a possible 4.22, beating a leading rival's 3.96.LINK

See you tomorrow for a new dose of ☕️ Techpresso!