DEVOURED
Trend extrapolation gets us to AGI in 1-4 years
Extrapolating current revenue and benchmark trends suggests that broadly superhuman AGI could arrive within 18 months, especially if algorithmic recursive self-improvement scales.
What: Benjamin Todd argues that AI progress is not slowing down but accelerating, with leading AI labs' annualized revenue growing at 6x annually and 'time-horizon' benchmarks for coding and agentic tasks doubling in speed. He posits that AI is increasingly leading its own R&D, potentially compressing years of advancement into months.
Why it matters: This highlights the 'accelerationist' viewpoint that assumes exponential growth in capability will continue until hitting physical, rather than purely software, limits.
Takeaway: If you are planning your career or research, consider the possibility that significant automation of knowledge work could reach a tipping point by 2028 or sooner.
Deep dive
- AI revenue grew to over $115 billion annualized by September 2026, showing a clear product-market fit.
- Time-horizon benchmarks (tasks a model can perform successfully) are accelerating, with coding agents moving from 3-hour to 100+ hour tasks.
- Anthropic reports AI was leading 26% of model R&D by August 2026, up from 0% in January.
- Algorithmic recursion allows models to design their own experimental processes, potentially creating a feedback loop independent of hardware constraints.
- Scaling laws suggest that more compute continues to yield capability gains despite diminishing returns at the margin.
Decoder
- Time horizon: A metric used by AI evaluators (like METR) to measure the duration of a task an AI can successfully complete, based on how long it would take an expert human to perform the same task.
Original article
Full article content is not available for inline reading.
Explore all 69 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi