"We need AI" is NOT a strategy. "We need to make inventory forecasting 40% more accurate" is. UK businesses succeeding with AI start with specific problems, not technologies. The question that unlocks real value isn't "Which AI should we buy?" but "Which decision, if improved,

5/5: The implications for business: don't confuse task-specific AI excellence with general intelligence. Your AI might be brilliant at one thing and useless at the next. Plan accordingly. True AGI, when it arrives, will be fundamentally different from what we have today.

4/5: Ilya's example: current models pass PhD-level exams but fail at basic debugging. They lack the "reliability" humans have. A 15-year-old won't alternate between two bugs infinitely. But AI does. That's not intelligence that's jagged pattern matching.

3/5: The documentary actually reveals this limitation. After StarCraft, DeepMind still needed years to crack AlphaFold. Why? Because each domain required new architectures, new training approaches, new everything. A human scientist switches domains far more easily.

2/5: But Ilya would push back hard on this framing. He argues that beating humans at specific tasks (Go, chess, protein folding) is "narrow AI" impressive but fundamentally limited. Real AGI means general learning ability, not specialised superhuman performance.

🧵𝗪𝗵𝗮𝘁 𝗖𝗼𝘂𝗻𝘁𝘀 𝗮𝘀 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲? 1/5: DeepMind's documentary celebrates "Move 37" the moment AlphaGo played something that seemed wrong but proved brilliant. Demis Hassabis calls it "machine creativity." It felt like a watershed moment: AI had transcended

DeepMind's AlphaGo made "Move 37" - a move no human would play. Commentators called it a mistake. It won the game. Ilya argues this isn't real intelligence. It's pattern matching without understanding. The bar for AGI keeps rising.

𝗪𝗵𝘆 𝗜𝗹𝘆𝗮 𝗪𝗮𝗹𝗸𝗲𝗱 𝗔𝘄𝗮𝘆 𝗙𝗿𝗼𝗺 $𝟭𝟬𝟬 𝗕𝗶𝗹𝗹𝗶𝗼𝗻 When Ilya Sutskever left OpenAI to found Safe Superintelligence, he walked away from potentially the most valuable startup in history. The documentary reveals his reasoning: he believes the commercial race

5/5: For UK businesses: this tension between shipping and researching is your tension too. Do you deploy today's AI (imperfect but available) or wait for tomorrow's breakthroughs (better but uncertain)? There's no universal answer. But understanding the trade-off is critical.

4/5: Safe Superintelligence's bet: stay in pure R&D for years. No products. No revenue. Just research. Only emerge when they have genuine superintelligence, not another chatbot. It's either visionary or delusional. Time will tell.

3/5: Ilya learned a different lesson from OpenAI. He watched a research lab become a product company. ChatGPT brought billions in attention and pressure. His conclusion? Commercial products force "the rat race" - shipping incremental updates instead of fundamental research.

2/5: Demis Hassabis navigated this by staying in London, maintaining research culture, and resisting product pressure. The result? AlphaFold took years with no commercial path. But they solved a 50-year problem and gave it to the world for free.

🧵𝗧𝗵𝗲 𝗖𝗼𝗺𝗺𝗲𝗿𝗰𝗶𝗮𝗹 𝗣𝗿𝗲𝘀𝘀𝘂𝗿𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 1/5: DeepMind documentary shows something revealing: after AlphaGo beat Lee Sedol, they immediately faced pressure. "When do we make money from this?" "How does this become a product?" Google's acquisition brought

DeepMind's approach: Spend 2 years in the "valley of death" perfecting AlphaFold architecture. Ilya's approach: Avoid commercial products entirely. "Straight shot to superintelligence." One company ships. The other researches. Both claim this is the path to AGI.

The Thinking Game - our latest article on Substack https://t.co/hEPVf2sMVS

𝗧𝗵𝗲 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 𝗗𝗶𝘃𝗶𝗱𝗲: 𝗜𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝘃𝘀 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 DeepMind's "The Thinking Game" is a masterclass in what we might call "brute force elegance." Watch them solve protein folding and you see a simple formula: assemble brilliant

5/5 For UK SMEs, this matters. Do you bet on the established giants with massive infrastructure? Or on the research labs promising smarter, more efficient AI? The answer shapes your AI strategy for the next decade.

4/5 Think about it: DeepMind proves AI works by throwing resources at hard problems until they crack. Ilya says the next leap requires AI that can "think" for 30 seconds and get reliable answers, not AI that reads the entire internet and still makes basic mistakes.

3/5 Here's the disconnect: DeepMind's AlphaFold used continent-sized compute to solve a 50-year problem. Ilya argues future breakthroughs won't come from bigger clusters, but from "unlocking" capabilities already inside models through better research.

2/5 But Ilya Sutskever - co-founder of OpenAI, now at Safe Superintelligence - says this model is broken. The "Age of Scaling" (2020-2025) worked by adding more GPUs and data. That era is over. Diminishing returns have arrived.

🧵The Two Paths Diverging DeepMind's documentary shows a 10-year journey: Atari → Go → StarCraft → AlphaFold → AGI. Each step methodical. Each breakthrough built on massive compute scaling. The message: patience, resources, and incremental progress win.👇👇