- AI with Kyle
- Posts
- AI with Kyle Daily Update 104
AI with Kyle Daily Update 104
Today in AI: Nanobanana 2k images + ChatGPT caught serving ads?
Kyle Balmer
December 05, 2025
The skinny on what's happening in AI - straight from the previous live session:
🎯 ChatGPT's "Not An Ad" Is Definitely An Ad: Target Pops Up Mid-Conversation
Benjamin finds Target shopping prompt while asking about Windows BitLocker. OpenAI respond directly and say it’s "not an ad” but instead the app SDK.
Kyle's take: Benjamin (ex-Xai, big on Twitter) asks ChatGPT about Windows BitLocker security. Gets "Shop for Home and Groceries - Connect Target" at bottom.
One staffer tells Benjamin he’s wrong and it’s just the app SDK. Benjamin's response is perfect: "When brands inject themselves into unrelated chat encouraging shopping, that's an ad." The more you pretend it isn't because you gave it different name, less users trust you.
Looks like an ad? Smells like an ad? It’s an ad.
Later OpenAI's CRO Mark Chen jumps in: "Dang, sorry Benjamin! We’ll look into this!" At least fessing up to it!
The recent Code red supposedly paused real ads rollout, but this SDK thing looks exactly like advertising. OpenAI needs money desperately but if they push ads they will likely haemorrhage users to Gemini. Caught between a rock and potential bankruptcy.
Source: Benjamin's Twitter
đź’Ł Google's Strategy: Run Gemini Ad-Free Until OpenAI Bleeds Out?
Keep Gemini pure while OpenAI forced to add ads everywhere. Looks like the Amazon playbook: lose money for years, then own everything.
Kyle's take: Google have a potential winning strategy here. Take loses until every competitor slaps ads everywhere to keep the lights on.
Imagine Google could be the ad-free experience whilst ChatGPT becomes a billboard to simply survive. Google has luxury of monetising through search ads - 70% of $350B revenue. Yup, it’ll have to cannibalise that revenue with AI ads eventually but they can hold off until OpenAI are in serious trouble.
OpenAI? Running on "vibes and runway," begging investors monthly for billions. Google could pull an Amazon - lose money for 10 years, choke out competition, then switch on profit machine once everyone's dead.
They own EVERYTHING: make their own TPUs (no Nvidia dependency), own cloud services, have user base, application layer through Chrome/Android. Oh, and a LOT of cash. All luxuries OpenAI don’t currently have.
🖼️ Nano Banana Pro Gets 2K Resolution: Google Flexes While OpenAI Struggles
2K images now in Gemini app (was API only previously).
Kyle's take: PSA: Nano Banana Pro now does 2K images directly in interface - click DOWNLOAD not copy for full resolution. Previously needed API. This is massive flex while OpenAI struggles with basic image quality.
Remember yesterday's discussion about ChatGPT’s yellow “piss filter”?? Supposedly because taskers fed ChatGPT images back into training, creating synthetic data loop.
eanwhile Google has decades of properly labeled Google Images - insurmountable advantage. Monday’s OpenAI code red lists "competitive image generation" as priority but good luck. They need to scrape/buy data while Google already owns it all, tagged and ready. Image generation is Google's game to lose now.
Source: Google announcement
🧄 "Garlic" Coming Next Week: OpenAI's Vampire-Slaying Model
Code name to ward off Gemini? Supposedly it’s more training, not new architecture. First time they're targeting specific competitor.
Kyle's take: Garlic releases next week - supposedly beats Gemini 3 in internal tests. From what we know it’s not new architecture, just more pre-training and post-training (same thing Gemini did - "we just did more").
Probably GPT 5.2, maybe 5.5 for bigger psychological jump, very much doubt it's GPT 6. It’s a reactive launch versus Gemini 3 so likely GPT5.2.
This is FIRST time OpenAI specifically says "this beats X model" rather than just "best ever." Previously they’ve not had to be reactive.
Code Read means all side projects paused - medical assistant, Pulse news, everything redirected here. Focus areas: better personalisation (4o lovers hate 5's dryness), faster responses, fewer "unnecessary refusals". But benchmarks increasingly meaningless - everyone hits top scores or doesn't release. Real test: daily workflows. Users want model they FEEL good using, not one that wins Math Olympiad. That's harder and it’s the engineering problem OpenAI's struggled with since GPT 5 launch.
Member Questions:
Kyle's response: Recognise him from Social Dilemma but haven't deep dived. Will do! Thank you for flagging him up.
I follow AI skeptics for balance - Ed Zitron (critical of tech, his Twitter and blog), Ed Newton Rex (blog and twitter), Pause AI (doing hunger strikes, not terribly popular or successful but raising important issues) and Eliezer Yudkowsky (website).
I don't agree with everything (obviously!) but important they raise alarms. And it’s important to not be blindly pro-AI because there are serious issues. Anyone who is totally pro or totally anti can be safely ignore in my view. It’s in the middle where we can actually have discussions.
Kyle's response: Great question. Didn’t know the answer so had to look this up - it's BRAVE! Found in Anthropic's subprocessor list by Simon Willison. Makes sense - similar morality/mission between companies! ChatGPT uses Bing (Microsoft owns 27% of OpenAI), Gemini uses Google obviously. Anthropic could create own index but why when Brave already exists? Just pay them.
Kyle's response: Honest to God reaction: gross. UGC should be actual humans. User Generated Content should be, duh, actual users. That’s the point - it’s real people’s reactions to your products and services.
Social proof comes from actual people (here’s all mine btw! My senja). If you just make it up it’s not social proof. In fact it’s the opposite.
Using AI avatars to pretend to be real people saying "great product!" is ethically bad AND I think a losing game. Like faceless YouTube channels - too easy means people selling tools make money, not users. More AI UGC devalues all UGC - people stop believing any testimonials. Self-sabotaging. Real crappy human testimonials with ums and ahs will have higher value than perfect AI ones. Short-term thinking.
Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.