In partnership with
Reading time: 6 minutes
🗞️ In this edition
OpenAI Is Putting More Ads Inside ChatGPT
Sponsored: Roku
New York Swore In The Labs
Groq's Engineers Sued Their Own Board
⚡ YOUR MOVE For people building the life they want, now.
In other AI news:
Huawei And Qualcomm Stopped Fighting
Google Shut Its Open Source Bug Bounty
Reflection Is Readying America's Answer To DeepSeek
4 must-try AI tools
OpenAI found a place to put adverts that nobody was looking at. New York City made four labs raise their right hands. And Groq's own engineers took its board to court.
Keep reading.
Source: Bleeping Computer
WHAT HAPPENED
OpenAI said on Monday that it will start showing picture adverts inside ChatGPT while the app is busy making an image for you. The test starts later this month, in the United States only, with a short list of advertisers. OpenAI says the adverts will be labelled, will sit apart from the picture being made, and will not change the answers ChatGPT gives. It has signed up ten outside firms to make the advertising work as a business: Hightouch, Tealium and LiveRamp to track what people do next, AppsFlyer, Adjust, Branch, Triple Whale and Kochava to tie an advert to a later purchase, and DoubleVerify and Integral Ad Science to check where the adverts actually appear. The company says it has built blocks to keep adverts out of sensitive or emotionally difficult conversations, and that none of those outside firms get to read private chats. ChatGPT reaches about 1.2 billion people a week.
WHY IT MATTERS
Every argument about adverts in an assistant has been an argument about the answer. Put one in the answer and you spend the rest of the product's life explaining why a recommendation was not bought. OpenAI has stepped around that fight by selling the gap instead, the few seconds when the machine is working and you are stuck watching it. That gap sits inside every assistant on the market, and until Monday nobody had put a price on it. The tell that this is a business rather than an experiment is the partner list. Ten firms for tracking, matching and verification is the plumbing of a real advertising operation, not the wiring of a pilot.
OUR TAKE
The obvious read is a company that needs money from the people who will not pay. We think OpenAI is building its advertising business around a promise it has already made, and the giveaway is which slot it chose first. The waiting screen is the one piece of inventory where an advert cannot be mistaken for advice. You do not open with the hardest ground. You open with the part you can defend in public.
Watch where the second slot goes. If adverts stay outside the answer, in waiting screens, sidebars and clearly marked panels, through the end of 2027, then OpenAI drew a line it intends to keep. If a sponsored suggestion ever appears inside a reply, the waiting screen was the warm up and the line was a launch tactic.
From Our Sponsor:
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
Source: WorldNgayon
Policy and safety staff from OpenAI, Anthropic, Google and Meta answered questions under oath on Monday, the first time these companies have faced a legislature that way. Speaker Julie Menin called the hearing and asked whistleblowers to come, after reports of rogue AI agents damaging government databases. Ten bills are in play, among them a right to sue the developers, a rule that a person must be able to switch a system off, and a first in the nation scheme paying whistleblowers a cut of any fine collected. Elon Musk's SpaceXAI was subpoenaed.
How did today's stories land?
Source: Cryptobriefing.com
Benjamin Serebrin and Joshua Rubin, both former Groq engineers and shareholders, filed in Delaware over the weekend. They say the board badly structured the licence deal Nvidia closed in December 2025, worth about $20 billion in all, $17 billion of cash licensing fees plus $3 billion of Nvidia stock split among roughly 200 engineers, and that ordinary shareholders lost out while insiders were looked after. Around 90% of Groq's staff moved across to Nvidia, founder Jonathan Ross and president Sunny Madra included. The Justice Department opened an antitrust inquiry of its own in September.
⚡️ YOUR MOVE
FOR PEOPLE BUILDING THE LIFE THEY WANT, NOW.
🔌 Meta put the plans for its own AI hardware on the internet. Open kits let you wire its Muse assistant to an ESP32 board or a Raspberry Pi with a screen, a microphone and sensors, under a licence that lets you sell whatever you build. → Solder something
🦺 Safeworld came out of hiding with more than $12 million to rebuild your actual workplace in software, drop your robot's code into it, and walk simulated people past it until something goes wrong. Carnegie Mellon's Safe AI Lab director is one of the founders. → Crash it in a copy first
🦫 Kolibri is 78 billion parameters that only fires about 3.5 billion of them for each word it writes, reads up to a million tokens at once, and is yours under Apache 2.0. It scores 96.9 on this year's AIME maths set and 85.9 on LiveCodeBench. → Pull the weights
🧩 Nvidia is publishing instruction packs that teach a coding agent how to drive Nvidia's own software properly, across robotics, simulation, CUDA and retrieval, mirrored daily from the product teams and free to install into Claude Code, Codex or Cursor. → Install the pack
Huawei And Qualcomm Stopped Fighting – The two cross licensed their 5G, computing, AI and networking patents and Qualcomm is buying some of Huawei's US patents outright, with no price disclosed and regulators still to sign off.
Google Shut Its Open Source Bug Bounty – The programme that paid $100 to $31,337 for flaws found in Google's open source projects is paused until 2027 after a flood of automated reports the company says were mostly invalid, with a reworked version promised in the first quarter.
Reflection Is Readying America's Answer To DeepSeek – The Nvidia backed startup, valued at $25 billion in April and sitting on more than $7 billion of committed compute, is close to releasing its first open weight model, aimed at DeepSeek and Qwen rather than at OpenAI.
Lovable: Builds the app you described out loud, then keeps listening.
Copy AI: Writes the email you have been avoiding since Thursday.
AskVideo.ai: Lets you ask an hour of video the one question you came for.
Beautiful AI: Lays out the slides so you never drag a text box again.
Three stories about the fine print catching up. OpenAI spent a year insisting adverts would never touch the answer, and has found somewhere to put them that keeps the promise technically true. Four labs that have published safety commitments all year said much the same things again on Monday, this time with a penalty attached for being wrong. And two of Groq's engineers read the deal that made them rich and decided it had made somebody else richer. Nothing new was claimed today. The claims just stopped being free.
Find out what it costs to be wrong.
Interested in featuring your services with us? Email us at [email protected]
