Time
-0.24999
Relative Brier Score
150
Forecasts
17
Upvotes
Forecasting Activity
Forecasting Calendar
Past Week | Past Month | Past Year | This Season | All Time | |
---|---|---|---|---|---|
Forecasts | 3 | 12 | 125 | 109 | 290 |
Comments | 0 | 3 | 16 | 15 | 62 |
Questions Forecasted | 3 | 9 | 31 | 22 | 52 |
Upvotes on Comments By This User | 0 | 1 | 8 | 8 | 35 |
Definitions |
New Prediction
New Prediction
Probability
Answer
0%
(-6%)
Yes
100%
(+6%)
No
New Prediction
Probability
Answer
40%
Yes
60%
No
SMEE already has working 28nm process, although there has been an argument that there are a quite a few problems & it is not yet commercially viable.
Files
New Prediction
Probability
Answer
14%
Yes
86%
No
Why do you think you're right?
- Huawei is on the US entity list, and is almost fully focused on the markets within China. There are little incentives for China to pivot to Western-led, non-proprietary solutions, especially when their current solutions can be integrated in intelligence-gathering operations more easily.
- Proprietary technology appears to ahead of O-RAN.
Files
Why might you be wrong?
- O-RAN could catch up and become a global standard.
Files
New Prediction
Probability
Answer
21%
(-2%)
Yes
79%
(+2%)
No
Why do you think you're right?
Updating slightly downwards and expanding on my previous forecast. Also expanded reasons I might be wrong with possible US actions.
- ASML has virtual monopoly on EUV machines, and have had it for a while. Incentives for creating similar technology are extremely high, and doing so seems extremely difficult.
- Even though EUV is difficult, China has made significant progress towards 7nm and 14-20nm process with their current equipment. 28nm process is already in use with SMEE. EUV is not strictly necessary, although having the tech would be beneficial.
- The drive towards frontier chip-making is the fact that older chips use a lot of energy, which makes energy a bottleneck in scaling up AI efforts. Even if older chips would use 4x more Watts per FLOPS it might be more feasible for China to just expand their grid much faster than what the US could hope to do, for cheaper than making progress in EUV would be.
- EUV litography machines are also dependant on multiple pieces of specialized technology, such as optics and mirrors produced by Carl Zeiss. Carl Zeiss is also close to a monopoly and is a virtual single point of failure for semiconductor manufacturing chains. Carl Zeiss is unlikely to end up cooperating with Chinese efforts.
- I don't expect EUV production capabilities to increase significantly beyond ASML ecosystem unless some very advanced AI system enables extremely significant levels of technological progress (and even then it is not obvious that whatever progress is made would be using EUV instead of some other currently unknown optimal solution).
Files
Why might you be wrong?
- I might underestimate the possibility of AI-accelerated technological progress and whether or not EUV capabilities are a constraint that would be solved as a part of that progress.
- Production of the tech that EUV machines need might not be as difficult to replicate as I think. Is Carl Zeiss a central actor in the space because they do something no-one else is able to do, or are they acentral actor because they do niche stuff and just don't have much competition by coincidence?
- It might be more difficult to garner up the political will to significantly scale Chinese electric grid and energy production to stay competitive in AI than I think.
- Eric Grunewald has expanded on possible ways to mitigate semiconductor scaling in China.
- US could identify specific researchers and companies that produce key components and put them on an entity list (like they did with Huawei; if you want to sell them things, especially if it's from the US, you need an export licence from the dept of commerce, which is a significant pain in the ass, and often-times you will not get the licence (although it is said that too often you will)).
- US could also pull funding from any academic collaboration that could exacerbate the issue.
Files
New Prediction
Probability
Answer
70%
(+45%)
Yes
30%
(-45%)
No
Nvidia skyrocketing.
Files
New Prediction
Probability
Answer
6%
(-7%)
Yes
94%
(+7%)
No
Time+Trump
Files
New Badge
Active Forecaster
New Prediction
Probability
Answer
Forecast Window
37%
(-20%)
Yes
Nov 7, 2024 to Nov 7, 2025
63%
(+20%)
No
Nov 7, 2024 to Nov 7, 2025
Updating closer to consensus
Files
New Badge
Top Forecaster - Oct 2024
Earned for being in the Top 10% of the leaderboard in a month.
Time