𝗔𝗜 𝘄𝗶𝗹𝗹 𝘁𝗮𝗸𝗲 𝗼𝘃𝗲𝗿 𝗺𝗼𝘃𝗶𝗲 𝗺𝗮𝗸𝗶𝗻𝗴 𝗯𝘆 𝟮𝟬𝟮𝟵 - 𝘀𝘁𝗶𝗹𝗹 𝗻𝗼 𝗵𝗲𝗹𝗽 𝗶𝗻 𝘀𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗺𝘆 𝗵𝗮𝗶𝗿 𝘀𝘁𝘆𝗹

𝗔𝗜 𝘄𝗶𝗹𝗹 𝘁𝗮𝗸𝗲 𝗼𝘃𝗲𝗿 𝗺𝗼𝘃𝗶𝗲 𝗺𝗮𝗸𝗶𝗻𝗴 𝗯𝘆 𝟮𝟬𝟮𝟵 - 𝘀𝘁𝗶𝗹𝗹 𝗻𝗼 𝗵𝗲𝗹𝗽 𝗶𝗻 𝘀𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗺𝘆 𝗵𝗮𝗶𝗿 𝘀𝘁𝘆𝗹𝗲 …

Really loved this episode[1] of Standford AI Club with Jason Wei, where he talks the audience through three AI Ideas and forecasts. I think my teaser is only fully understandable if you watch the video entirely - and this is intended as it is a very good session.

Besides this, those are the three key ideas:

🔥 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝗮𝘀 𝗮 𝗖𝗼𝗺𝗺𝗼𝗱𝗶𝘁𝘆 AI intelligence is becoming a resource anyone can access cheaply and instantly. With adaptive compute, the cost of gaining knowledge or reasoning power is dropping rapidly, enabling democratized access to skills like coding and healthcare insights. Public information is nearly free, but private insider data gains premium value. Soon, personalized AI-driven access to information will be seamless and frictionless.

🔎 𝗩𝗲𝗿𝗶𝗳𝗶𝗲𝗿’𝘀 𝗟𝗮𝘄: 𝗣𝗿𝗼𝗴𝗿𝗲𝘀𝘀 𝗧𝗵𝗿𝗼𝘂𝗴𝗵 𝗩𝗲𝗿𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 AI’s capability to solve tasks is closely tied to how easily the correctness of solutions can be verified. Tasks like Sudoku or code validation are easier to check, so AI masters these faster. Conversely, tasks like crafting factual essays present tougher verification challenges, slowing automation. Techniques like DeepMind’s AlphaEvolve leverage repeated verification and sampling for iterative improvements. This sets the stage for automating jobs with clear, measurable success criteria.

🎢 𝗧𝗵𝗲 𝗝𝗮𝗴𝗴𝗲𝗱 𝗘𝗱𝗴𝗲 𝗼𝗳 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲: 𝗨𝗻𝗲𝘃𝗲𝗻 𝗔𝗜 𝗣𝗿𝗼𝗴𝗿𝗲𝘀𝘀 AI development won’t happen with a sudden superhuman leap but will vary widely across tasks. Digital, data-rich tasks that humans find easier, such as translation or competitive math, progress rapidly. Physical or niche skills like hairdressing or rare language translation improve slowly or remain human-led. Each task’s rate depends on data availability, verifiability, and its digital nature, producing a jagged capability landscape.

💡 𝗦𝘂𝗺𝗺𝗮𝗿𝘆 AI will make intelligence fast and cheap, progress will be driven by how well tasks can be verified, and AI’s impact will differ strongly by domain. The world of software and digital knowledge work will accelerate tremendously, but human-centric, physical tasks will remain much harder for AI to replace—at least for some time.

𝗪𝗵𝗮𝘁 𝗶𝘀 𝘆𝗼𝘂𝗿 𝘃𝗶𝗲𝘄 𝗼𝗻 𝘁𝗵𝗶𝘀?  𝗔𝗻𝗱 - 𝘄𝗵𝗮𝘁 𝗮𝗯𝗼𝘂𝘁 𝗺𝗲 𝗴𝗲𝘁𝘁𝗶𝗻𝗴 𝗮 𝗽𝗿𝗼𝗽𝗲𝗿 𝗵𝗮𝗶𝗿 𝗰𝘂𝘁?!

[1] Stanford AI Club: Jason Wei on 3 Key Ideas in AI in 2025 on YouTube - https://www.youtube.com/watch?v=b6Doq2fz81U