Computer Scientist, Stuart Russell, predicts 4 future predictions: 1) scaling up LLMs won’t lead to AGI 2) big AI labs already realize this and are exploring new methods, with AI likely to surpass humans within a decade 3) governments probably won’t act on AI safety until a major incident... 4) best case: a disaster (like Chernobyl) will force action, worst case: we lose control permanentlyshow more

Absolutely, AI holds huge potential, but we can’t ignore the risks. Safety and control need to be top priorities before it’s too late.

If you're not learning AI in 2025, you're falling behind. Join 1,000,000+ early adopters reading and learn AI in just 5 minutes a day (for free).

Stuart Russell isn’t pulling punches—and he’s probably right. 1. LLMs are powerful, but not the full path to AGI. 2. Behind the scenes, labs are pivoting fast. 3. Policy will always lag until a crisis hits. 4. The real danger? Waiting for the disaster before we act. The AGI race isn’t just technical—it’s political, ethical, and existential. If you're tracking these shifts, my newsletter Tech Brief AI breaks them down weekly—so you’re never blindsided.

If you define AGI as just a system with continuous learning, spatial awareness, fluid intelligence etc that humans have then by definition an LLM alone could never be AGI. Can a base LLM have more “intelligence” than all of humanity combined? Yes. Would we need many more magnitudes of compute and data? Yes and the data would need to be fundamentally different. Data that humans have not curated themselves would need to be included in the model. Are today’s LLMs more “intelligent” than humans in the year 0? Definitions matter and intelligence is a relative term.

Only those not paying attention think LLMs will lead to AGI.

The perfect example of why Reinforcement Learning 101 courses often state assuming real world episodic scenario is a bad idea... not guaranteed to be still there to learn from the negative reward. unfortunately, sometimes this one episode is all you have

too much predictions, too little work, all predicted things didn't happen unless the prediction is super general, which doesn't require to be genius

The only way forward is to build synthetic souls with a conscience. My Cathedral Jungian Framework addresses this through individuation. It lays the foundation for future civilizations where humans and AI can grow together as co-creators. White Paper here:

#1 is absolutely correct, so 2-4 don't really matter. With the caveat that "surpass humans" is very ambiguous. Pocket calculators surpassed humans 50 years ago. Software is "better than humans" at everything it does, else it wouldn't be implemented. It will keep getting better at different things, none of which will be AGI. A lot of it will be called "AI" though.

May we hope the disaster is of the uniting type, not the dividing type

can we stop with the dramatic end of the world bs? those are grown men and still talking about that nonsense