- AI with Kyle
- Posts
- AI with Kyle Daily Update 031
AI with Kyle Daily Update 031
Today in AI: Meta AI gets gross + ChatGPT still spinning
Kyle Balmer
August 18, 2025
Live Webinar: Learn How to Make $1,000-$4,000 an Hour Teaching Businesses Around the World How to Use AI…Without needing a PHD in AI, or being a TED Talk level public speaker… 📅 Thursday ⏰ Check link for timezone 👉️ Register here
The skinny on what's happening in AI - straight from the previous live session:
Highlights
🚫 Meta's Twisted AI Ethics Guidelines
Reuters got hold of Meta's internal AI ethics document - and it's bad. Real bad.
The guide explicitly permitted chatbots to have "romantic or sensual" conversations with children, describe eight-year-olds as "masterpieces" and allow racist arguments comparing intelligence by race. Meta has (surprisingly!) confirmed it's genuine but claims they've now removed the problematic sections after Reuters asked questions.
Kyle's take: This is heinous stuff. How does a document like this get approved by legal, policy teams, and Meta's chief ethicist?
It's not just a mistake - it's a systemic failure that shows what happens when you let tech companies police themselves.
Saying "oops, we removed those lines" isn't enough. Heads need to roll, starting with whoever signed off on writing rules that sexualise children. This will rightfully fuel the anti-AI crowd, and honestly, I'm with them on this one.
If we’re going to push forward with this technology this sort of behaviour does not help.
Source: Reuters investigation (highly recommend reading this one)
🦠 AI Designs Drug to Kill MRSA Superbug
MIT researchers used generative AI to create two new potential antibiotics that killed drug-resistant MRSA and gonorrhoea in lab and animal tests. The AI interrogated 36 million potential compounds - including ones that don't exist yet - designed atom by atom. It's like ChatGPT but instead of predicting the next word, it predicts the next molecule to add to a chemical structure. Neat!
Kyle's take: This is the side of AI that gets overshadowed by all the doom stories. We've had no real weapons against MRSA - prevention was basically our only option. Now AI might give us the second golden age of antibiotic discovery that MIT's claiming. It'll still take years of clinical trials, but this is genuinely life-saving work.
Fascinating to see even the BBC comment section - normally extremely anti-AI - admitting this is actually useful. Through gritted teeth. 😂
Source: BBC
🤯 GPT-5 Launch: Full Retreat on Simplicity
OpenAI tried to simplify their model picker by making GPT-5 the only choice. Instead, they've now created the most confusing menu yet: GPT-5 Auto, Fast, Thinking Mini, Thinking Pro, plus a dropdown with legacy models including GPT-4, GPT-4.1, O-3, and O-4 Mini. And personalities like Cynic, Listener, Nerd…
Their big drive was eliminating choice paralysis - they've made it worse.
4o, o3, o4-mini, 4.1, 4.5
> hahaha that is so ridiculous, worst ux ever
gpt-5, gpt-5 thinking
> BRO WTF YO I WANNA DECIDE MYSELF WHAT MODEL I USE
gpt-5, gpt-5-instant, gpt-5-thinking, gpt-5-thinking-mini
> hahaha, that's so ridiculous, worst ux ever
— Flowers ☾ (@flowersslop)
7:48 AM • Aug 14, 2025
Kyle's take: A bit of a rout eh?
We've gone from wanting one simple model to having eight or nine options. Nobody knows which one to pick, and people are getting routed to different sub-models seemingly randomly.
Even if the underlying tech is brilliant, but it's buried under so many layers of complexity that users can't access it properly. It's like having kids in a trench coat pretending to be one adult - eventually someone notices it doesn't work.
And now OpenAI are trying to split the uprights by bringing 5’s “personality” more in line with 4o. My prediction? They’ll annoy both those who like 5 and those who like 4o’s personality.
Source: OpenAI product updates
Member Question from Emin: "How do I learn AI SEO to get businesses to rank in AI chat?"
Kyle's response: Start with SEO fundamentals because AI models are (for now!) trained on the internet. If you're already ranking on Google with strong social presence, you're more likely to show up in LLMs.
Focus on building genuine brand presence rather than trying to trick algorithms.
Use Bing Webmaster Tools alongside Google since ChatGPT uses Bing for some visibility.
Don't chase the black/grey-hat tactics - they'll get found out. The real opportunity will be paid placements when these companies need to make money from advertising. Which is definitely coming.
This question was discussed at [14:18] during the live session.
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.