ABCDEFGJKLMNOPQRSTUVWXYZ
1
Start year2025
2
End year2100
3
FLOP per dollar at the start of period (2025)4.00E+17
4
Compute price halving time in this period, in years1.5
5
Maximum FLOP per dollar in this period1.00E+26
6
Compute cost for the most expensive training run at the start of period (2025), in 2020 USD1.00E+10
7
Doubling time of spending on compute for the most expensive training run at start of period (2025), in years.1
8
Willingness to spend on computation, as a fraction of frontier GDP in 2020 USD2.00%
9
Annual growth rate (%) of real frontier GDP in this period (2025 to 2100)5%
10
Probability that the FLOP to train a transformative model is larger than all paths at the beginning of the period (2025)5%
11
Probability that the FLOP to train a transformative model is larger than all paths at the end of the period (2100)1%
13
Lifetime anchorShort horizon neural networkGenome AnchorMedium horizon neural networkLong horizon neural networkEvolution anchor
14
At the start of this period (2025), how many OOMs higher (+) or lower (-) are the training FLOP required under this hypothesis compared to the imported distribution (shown in another tab)?000000
15
What weights would you assign to each path, given that at least one path works? (No need to sum to 1)0.10.40.20.10.10.05
16
Probabilities assigned to each hypothesis at the beginning of the period (2025)10%40%20%10%10%5%
17
Cumulative probabilities10%50%70%80%90%95%
18
Halving time of compute requirements per path over this period (2025 to 2100), in years2.522111
20
What is the maximum OOMs of improvement for this hypothesis by the end of the period (2100)?334456
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103