MikeAnthony (@MikeAnthony) header
MikeAnthony (@MikeAnthony) profile picture

Horror scribe | AI tinkerer | Strange stories from the dark side | Talking hoops & haunts | London-based | Join the chaos 🏀🌑| The Siquijor Chronicles 📖

London, England
Joined December 2010
Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)Photo by MikeAnthony (@MikeAnthony)
People see a 60 on the Artificial Analysis Intelligence Index and think, “Wait, this AI only has an IQ of 60?” 😂 Mate, that's comparing apples to pineapples. 🍎🍍 The AA Index scores AI from 0–100 on ridiculously difficult PhD-level science, engineering and autonomous coding challenges. An average adult with an IQ of 100 would likely struggle to score even 15. So, for a rough perspective, here's how the scores might compare to human IQ levels. Not a scientific conversion, mind you. Just an illustration. • 5–15 AA ≈ IQ 100 — Average human intelligence. Your everyday problem-solving brain. • 30–45 AA ≈ IQ 125 — Richard Feynman territory, based on his reported school IQ score. Serious brainpower. • 50–55 AA ≈ IQ 135 — Garry Kasparov territory, based on his reported 1987 Der Spiegel testing. Now we're cooking! • 60–65 AA ≈ IQ 142–145 — Top 0.1% territory. Think Gemini 4 Neon, GPT-6.2 Astra and Fable 5.2. That's some serious intellectual horsepower! But here's where AI gets REALLY interesting. Imagine that level of problem-solving ability paired with a 1 MILLION-token context window. Even a human genius with an IQ of 145 gets tired, forgets details and occasionally loses the plot halfway through a 300-page manuscript. AI? No coffee breaks. No annual leave. No calling in sick on Monday. 😂 Give it a massive manuscript, ask it to analyse the plot, find inconsistencies, examine character development, check the logic and suggest improvements. Of course, AI isn't perfect. A million-token context window isn't flawless memory, and intelligence is far more complicated than one benchmark score. But think about what this means for ordinary creators like us. One person can now have something resembling an entire team of highly intelligent specialists at their fingertips. Writers, researchers, editors, programmers, analysts. All available on demand. That's the real revolution. AI doesn't replace human imagination, creativity or direction. It multiplies what one human being can accomplish. You provide the vision. AI provides the horsepower. Just remember who's driving. 😎 And please, stop comparing AA scores directly with human IQ. They're completely different measurements! The future isn't just about smarter AI. It's about what ordinary people can achieve with extraordinary tools.
5
Hey @elonmusk, serious question. 😂 If ChatGPT swallows its pride and Tito Sam @sama personally says “SORRY, ELON!”… would you lend OpenAI a little compute from your mighty Colossus cluster? Just a few hundred thousand GPUs. Nothing extravagant. 🤣 Come on, Elon! Even AI rivals deserve a little charity. Sharing is caring! 😂🔥
10
Funny how spoiled I’ve become with ChatGPT Work and Dots. I’m so used to handing over a workflow and letting AI do the heavy lifting. 😂 Now I’m experimenting with Google Antigravity for my video workflows, and it’s a different beast! Work = delegation. Dots = persistent follow-through. Antigravity = hands-on agentic building and control. Growing pains? Absolutely! But I’m starting to appreciate the strengths of each. Different toys, different superpowers. Why choose one when you can build an entire AI arsenal? 🚀 Competition is brilliant for us users!
1
27
I genuinely like Grok, but I’m beginning to wonder if Elon is getting bored with regular ChatGrok. Look at the signs. Reported cuts to the Grok Imagine team, increasing emphasis on coding, and Grok Bot getting serious investment. My suspicion? Grok Bot eventually becomes the flagship, gets the best models, compute and upgrades, while regular Grok gets slower releases, reduced features or is quietly nerfed. Imagine could suffer too. We've seen how quickly Elon changes direction at Tesla. He's not sentimental about product lines. It's his company, after all. His priorities, his decisions. I hope I'm wrong. Grok is a genuinely capable model, and competition needs it. But I wouldn't underestimate Elon's willingness to pivot overnight. Just my speculation, but the signs are worth watching.
24
Google is on fire.... 🥸
shipped a bunch of new things this week, here's a recap:
9
I’m rooting for Qwen, Kimi and Chinese labs. But copying ChatGPT, Gemini or Claude isn’t leadership. Invest in your researchers, labs, compute, training runs and safety testing. Distillation can create powerful models without matching safeguards. Build, don’t copy the homework.
16
After days of testing, my verdict: Astra 6.2 is seriously good! But AI isn't a loyalty contest. ChatGPT, Gemini, Claude, Grok, Kimi, Qwen, DeepSeek—plenty of choices. For me, ChatGPT and Gemini are golden. Claude? Excellent model, no use case for me. Use what works. Unsubscribe if it doesn't.
1
30
For writing stories, ChatGPT is king. Give Astra 6.2 a detailed draft and it’ll catch the line your villain stole from your heroine, fix it, and still leave the story sounding like you. Fable 5.2, Opus 5.5, Neon 4, Carbon 4? For that final edit, it’s not even close. Bank it.
18
While everyone is busy arguing over AI benchmarks, Google and DeepMind are playing a much bigger game. AlphaFold, AI-driven drug discovery, and now $5B for Isomorphic Labs. I genuinely believe the next decade could transform medicine and science beyond what we imagine today. Winning benchmarks is impressive. Changing humanity's future? That's the real prize.
Thrilled to announce $5B in funding for @IsomorphicLabs to accelerate our mission to Solve All Disease. We continue to make rapid progress reimagining drug design from first principles with AI to radically cut the time & cost to find new medicines. Gradually, then all at once...
1
31
Is Dots running on Astra 6.2 now?
9
It would be nice to have Dots in the UK officially... @thsottiaux 😎
1
1
22
I like Anthropic, and I’d love Dario to be right. Bristol Myers Squibb is already using Claude to help plan clinical trials. Generate Biomedicines used it to redesign immune cells that killed tumour cells more effectively in the lab. That deserves credit. On track record, though, I’d give Google DeepMind the edge. Here in the UK, they’ve worked with Moorfields Eye Hospital on eye scans, University College London Hospitals on planning radiotherapy for head and neck cancer, and Royal Surrey County Hospital on spotting breast cancer in mammograms. Just today, DeepMind spinout Isomorphic Labs announced a partnership with MSD to look for a new cancer drug. Both teams are doing work worth backing. A cure in patients is still to be shown. Do I buy the ten-year prediction? I’m sceptical. But that’s one argument I’d happily lose.
86
Nano Banana Pro turns one tomorrow. Is Google bringing cake or the rumoured Gemini 4-powered Nano Banana 3 Pro? 🍌🎂
2
83
MikeAnthony retweeted
🚨 GOOGLE JUST DROPPED A 1-HOUR COURSE THAT TEACHES YOU AGENTIC AI FROM SCRATCH. No fluff. No expensive bootcamp. In just 60 minutes, you’ll learn how to: → Build your first AI agent → Give agents persistent memory → Create long-running agentic loops → Build MCP and understand MCP vs APIs → Connect multiple AI agents into one system This could save you hours of jumping between paid courses. 🔖 Bookmark this. Watch it this weekend.
39
277
11
2,670
351,374
My AI stack is so much cleaner now... Thank you @GeminiApp @GoogleDeepMind
25
Final seven weeks of 2026. My Gemini bet? Ultralight Neon before New Year at 51–53 AA. Luna speed, more brains. Krypton for cyber defenders first, wider in Jan/Feb. Pure pattern-guessing. If Krypton hits that rumoured 68–70 AA, Gemini’s got OpenAI and Anthropic by the throat.
59
I'm sorry, but Google isn't just competing anymore. They're starting to toy with the competition. Argon. Carbon. Hydrogen. Now Neon. FOUR models in barely FIVE weeks. And if Neon's reported 61 AA score holds up, we're talking about a FAST model outperforming Opus 5.5, Sol 6.2, and even Google's own Carbon. At this point, Google's release cadence is starting to look suspiciously like recursive self-improvement (RSI). Not proof, but the pace is ridiculous. This is turning into an absolute bloodbath. My prediction? Google isn't finished. I suspect they're keeping something like Krypton in reserve, possibly for an early December surprise or a January 2027 release. And I wouldn't be shocked if that next model starts knocking on 70 AA. The competition barely has time to celebrate a launch before Google drops another model. It's like watching a cat play with its food. The scary part? I don't think Google has shown its full hand yet.
1
44
> People keep asking if Gemini 4 Argon is "the last of the old models" because it sits at 50 on Artificial Analysis. > Nope—it’s the opposite. Argon was the first Gemini 4 checkpoint, and Google’s RSI flywheel just lapped it in 6 weeks. > How the lineup connects: > • Argon (50 AA): The Gen-1 seed that proved the 1M architecture and kicked off the internal RSI loop. > • Carbon (59–60 AA): The RSI-boosted flagship on that same base—~10 points smarter. > • Hydrogen (51–54 AA): The fast Flash-class engine ($1.50/$7.50) that edged past Argon on the newer pipeline. > • Neon ($3/$15, out today): The new Pro sweet spot between Hydrogen and Carbon’s $4/$20 standard tier. > Argon wasn’t a leftover—it was Ground Zero for RSI squeezing a year of iteration into a month. > Full orchestrator workflow breakdown: youtube.com/watch?v=yE8gqufO…
98
🔥 GOOGLE'S RECURSIVE SELF-IMPROVEMENT (RSI): THE 2027 COMPETITIVE BLOODBATH Let’s be entirely real: The launch of the Gemini 4 family isn't just a standard upgrade cycle—it’s the beginning of a paradigm shift. Google co-founder Sergey Brin recently let the cat out of the bag: Google has officially stepped onto the Recursive Self-Improvement (RSI) treadmill. With toolsets like DeepMind's AlphaEvolve and Gemini 4 Argon actively working with researchers to write and refine its successor models, Google has cracked the ultimate flywheel: AI building better AI. By using their vast, unparalleled TPUs (like the Ironwood infrastructure) to let Gemini evaluate, debug, and optimize its own next-generation weights and agentic pipelines, the rate of acceleration is compounding. Google is no longer just running in the AI race; they are toying with the competition. They've built an autonomous evolutionary pipeline while other frontier AI companies are still manually tuning RLHF and wrestling with training data bottlenecks. If Google has truly unlocked a self-improving cascade, the gap between Gemini and everyone else won't just widen—it will become an impassable chasm. Watch out for 2027. It's going to be an absolute bloodbath. 💀🪐
🤖 Made with AI
68