The future is very hard to predict. We don't think this model, or any other model, should be trusted completely. The model takes into account what we think are the most important dynamics and factors, but it doesn't take into account everything. Also, only some of the parameter values in the model are grounded in empirical data; the rest are intuitive guesses. If you disagree with our guesses, you can change them above.

Nevertheless, we think that modeling work is important. Our overall view is the result of weighing many considerations, factors, arguments, etc.; a model is a way to do this transparently and explicitly, as opposed to implicitly and all in our head. By reading about our model, you can come to understand why we have the views we do, what arguments and trends seem most important to us, etc.

The future is uncertain, but we shouldn’t just wait for it to arrive. If we try to predict what will happen, if we pay attention to the trends and extrapolate them, if we build models of the underlying dynamics, then we'll have a better sense of what is likely, and we'll be less unprepared for what happens. We’ll also be able to better incorporate future empirical data into our forecasts.

In fact, the improvements we’ve made to this model as compared to our timelines model at the time of AI 2027 (Apr 2025), have resulted in a roughly 2-4 year shift in our median for full coding automation. The plurality of this has come from improving our modeling of AI R&D automation, but other factors such as revising parameter estimates have also contributed. These modeling improvements have resulted in a larger change in our views than the new empirical evidence that we’ve observed. You can read more about the shift in our blog post. 

In our blog post, we give a brief survey of other methods for forecasting AI timelines and takeoff speeds, including other formal models such as FTM and GATE but also simpler methods like revenue extrapolation and intuition: you can read that here.

You can find our forecasts here. These are based on simulating the model lots of times, with each simulation resampling from our parameter distributions. We also display our overall views after adjusting for out-of-model considerations.

With Eli's parameter estimates, the model gives a median of late 2030 for full automation of coding (which we refer to as the Automated Coder milestone or AC), with 19% probability by the end of 2027. With Daniel’s parameter estimates, the median for AC is late 2029.

In our results analysis, we investigate which parameters have the greatest impact on the model’s behavior. We also discuss the correlation between shorter timelines and faster takeoff speeds.

Our blog post discusses how these predictions compare to the predictions of our previous model from AI 2027.

In terms of how how much they affect Eli’s all-things-considered views (i.e. Eli’s overall forecast of what will happen, taking into account factors outside of the model), the top known limitations are:

Not explicitly modeling data as an input to AI progress. Our model implicitly assumes now that any data progress is proportional to algorithmic progress. But data in practice could be either more or less bottlenecking. Along with accounting for unknown unknowns, this limitation pushes Eli’s median timelines back by 2 years.

Not modeling automation of hardware R&D, hardware production, and general economic automation. We aren’t modeling these, and while they have longer lead times than software R&D, a year might be enough for them to make a substantial difference. This limitation makes Eli place more probability on fast takeoffs from AC to ASI than the model does, especially increasing the probability of <3 year takeoffs (from ~43% to ~60%).

Meanwhile, Daniel increases his uncertainty somewhat in both directions in an attempt to account for model limitations and unknown unknowns and makes his takeoff somewhat faster for reasons similar to Eli’s.

A more thorough discussion of this model’s limitations can be found here, and discussion of how they affect our all-things-considered views is here.

The future is very hard to predict. We don't think this model, or any other model, should be trusted completely. The model takes into account what we think are the most important dynamics and factors, but it doesn't take into account everything. Also, only some of the parameter values in the model are grounded in empirical data; the rest are intuitive guesses. If you disagree with our guesses, you can change them above.

Nevertheless, we think that modeling work is important. Our overall view is the result of weighing many considerations, factors, arguments, etc.; a model is a way to do this transparently and explicitly, as opposed to implicitly and all in our head. By reading about our model, you can come to understand why we have the views we do, what arguments and trends seem most important to us, etc.

The future is uncertain, but we shouldn’t just wait for it to arrive. If we try to predict what will happen, if we pay attention to the trends and extrapolate them, if we build models of the underlying dynamics, then we'll have a better sense of what is likely, and we'll be less unprepared for what happens. We’ll also be able to better incorporate future empirical data into our forecasts.

In fact, the improvements we’ve made to this model as compared to our timelines model at the time of AI 2027 (Apr 2025), have resulted in a roughly 2-4 year shift in our median for full coding automation. The plurality of this has come from improving our modeling of AI R&D automation, but other factors such as revising parameter estimates have also contributed. These modeling improvements have resulted in a larger change in our views than the new empirical evidence that we’ve observed. You can read more about the shift in our blog post. 

In our blog post, we give a brief survey of other methods for forecasting AI timelines and takeoff speeds, including other formal models such as FTM and GATE but also simpler methods like revenue extrapolation and intuition: you can read that here.

You can find our forecasts here. These are based on simulating the model lots of times, with each simulation resampling from our parameter distributions. We also display our overall views after adjusting for out-of-model considerations.

With Eli's parameter estimates, the model gives a median of late 2030 for full automation of coding (which we refer to as the Automated Coder milestone or AC), with 19% probability by the end of 2027. With Daniel’s parameter estimates, the median for AC is late 2029.

In our results analysis, we investigate which parameters have the greatest impact on the model’s behavior. We also discuss the correlation between shorter timelines and faster takeoff speeds.

Our blog post discusses how these predictions compare to the predictions of our previous model from AI 2027.

In terms of how how much they affect Eli’s all-things-considered views (i.e. Eli’s overall forecast of what will happen, taking into account factors outside of the model), the top known limitations are:

Not explicitly modeling data as an input to AI progress. Our model implicitly assumes now that any data progress is proportional to algorithmic progress. But data in practice could be either more or less bottlenecking. Along with accounting for unknown unknowns, this limitation pushes Eli’s median timelines back by 2 years.

Not modeling automation of hardware R&D, hardware production, and general economic automation. We aren’t modeling these, and while they have longer lead times than software R&D, a year might be enough for them to make a substantial difference. This limitation makes Eli place more probability on fast takeoffs from AC to ASI than the model does, especially increasing the probability of <3 year takeoffs (from ~43% to ~60%).

Meanwhile, Daniel increases his uncertainty somewhat in both directions in an attempt to account for model limitations and unknown unknowns and makes his takeoff somewhat faster for reasons similar to Eli’s.

A more thorough discussion of this model’s limitations can be found here, and discussion of how they affect our all-things-considered views is here.