The skinny on what's happening in AI - straight from the previous live session:
Highlights:
Opus 4.5 spotted in APIs, release this week expected. My guess is that it’ll be the best coding model ever made. Also completely unusable due to rate limits.
Kyle's take: Opus 4.5 is coming this week it seems - it’s been found already hidden in Poe and API endpoints found last week. It'll be better than Sonnet 4.5 at everything, especially coding.
One problem: nobody uses Opus because Anthropic's limits are insulting. Hit your daily limit in a few requests on $20/month plan. Weekly limits mean mid-project it says "see you next Monday." That's not a product, that's hostage-taking.
And I say this as someone who loves Claude and uses it as my main model for most tasks.
With Codex 5.1 Max and Gemini 3 catching up on coding, Anthropic can't afford to keep their best model locked behind prison bars. Unless they announce "it's cheap and limits are gone" (they won't), Opus 4.5 will be the greatest model on earth that nobody uses…Again.
Last minute update: it looks like I’m wrong which is super exciting! AAAND it’s released. More tomorrow.
📝 Nano Banana Pro Fills Out Exam Papers in Handwriting
Google's image model can now complete physics/chemistry exams with full workings, diagrams, even sketches - indistinguishable from human student.
Kyle's take: Forget memes and celebrity photos - this is the terrifying bit. Feed Nano Banana Pro an exam paper, it fills it out in handwriting. Full workings, margin notes, crossed-out attempts - exactly like a nervous student.
Andrej Karpathy (OpenAI co-founder) is frantically posting about it - when top AI researchers are excited, pay attention.
We’re seeing image understanding and manipulation moving scarily close to AGI levels here.
🕵️ AI Detection Expert Admits Defeat: "We Can't Tell Anymore"
Jeremy Carrasco (@showtoolsai), professional AI image detector with 370k followers, says Nano Banana Pro is basically undetectable. Society needs new plan.
Kyle's take: This guy's whole thing was spotting AI images - pointing out tells, artefacts, impossibilities. He’s great at it and has massive following doing frame-by-frame breakdowns.
His video above is well worth watching. We're entering world where visual evidence means nothing. Courts, journalism, relationships - everything built on "pics or it didn't happen" needs rethinking. Government and regulation move at snail pace while tech moves at light speed. The gap between those speeds is where chaos lives - and we’re very much in that space now.
💷 UK Throws Billions at AI (Still Won't Beat US or China)
Government announces AI investment in the billions, a set of new initiatives, AI ambassadors and even a vibe coding MP!
Nice to see some movement in the sector. Especially an MP sitting down Cursor to vibe code a site. He even has a Github repo! The first MP to have a repo?? Could be!
The government has also appointed AI ambassadors: an economist, startup founder/YC leader and a DeepMind VP - decent choices, not politicians at least!
But here's reality: UK is brilliant at creating researchers then watching America pay them 10x more... DeepMind was British founded - Google bought it for £400M in 2014, now looks like history's best acquisition. UK top talent gets offered millions in San Francisco vs £100k in expensive London. Do the maths. Or math in this case.
We're in a two-horse race between US and China. UK could repurpose entire economy for AI and still lose. Nice to see activity but we're playing different game entirely.
🚫 GPT-4o Gets Death Date: February 16, 2026
OpenAI officially deprecating 4o. The model people rioted to keep alive gets execution date. "We encourage you to explore 5.1" Exactly what proponents of 4o don’t want to do!
Kyle's take: Remember the GPT-5 launch disaster? They killed 4o overnight, people revolted, OpenAI panic-reversed saying "fine, legacy model, temporary." Now they've set execution date: February 16, 2026.
The phrasing "we encourage you to explore 5.1" is deliberately provocative - they KNOW people are attached to 4o's personality vs 5's robotic efficiency. This is the companion AI crowd's favourite model, the one that agreed with them, made them feel heard.
The backlash will be biblical when people realise what's happening.
Kyle's Planning Advice for Vibe Coders:
When asked about #1 tip for building apps with AI, Kyle shared his pre-production process:
Don't just vibe - Yes he made an app in the bath last night, but serious projects need planning
Interview technique - Tell Claude/ChatGPT basic idea, then say "interview me for details you need"
Voice responses - Answer planning questions with voice, keeps it conversational
Iterate questions - "Ask me more clarifying questions" until fully specified
Create spec - "Capture all this into a project specification for my vibe coding tool"
Then vibe - Take spec to Lovable/Cursor/Claude Code with structure already figured out
Special UK Charity Drive:
Google giving £11 to Movember for every Nano Banana moustache selfie posted with #GenAIMo. Up to £1 million available. One week left. Free money for charity, just upload selfie, add moustache, post.
Horrific! But it’s for charity!
👉 Jump onto Gemini prompt via your phone or laptop
👉 Copy and paste the following prompt into the prompt box: Pick a moustache that would suit my style, and then generate an instant camera portrait of me, with the moustache style you selected, in full colour. I should be centred in the instant photo against a white backdrop, facing the camera. It should look like the photo was taken with a bright, direct flash, making the photo high contrast with shadows. Place the instant camera photo on a matte white surface.
👉 Magic! Your new moustached face photo is created.
👉 Upload your photo onto socials, and don't forget to tag @googleUK and @movember
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.
Streaming on YouTube (with screen share) and TikTok (follow and turn on notifications for Live Notification).
