timelines-takeoff-ai-2027/timelines/params.yaml at main · uvafan/timelines-takeoff-ai-2027 · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
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
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
# Shared simulation parameters
simulation:
  n_sims: 5000     # Number of Monte Carlo simulation runs
  n_steps: 30000   # Maximum simulation steps before timeout
  t_0: 2025.25        # Starting year for simulation
  dt: 1.0          # Time step size in days

# Plotting style configuration
plotting_style:
  font:
    family: "monospace"   # Add this line to specify monospace font
    sizes:
      axis_labels: 18
      title: 18
      main_title: 32
      ticks: 16
      legend: 16

  colors:
    background: "#FFFEF8"  # Add this line for cream background (255, 250, 240)
    human:
      dark: "#484848"
    default: "#000000"
# Active forecasters configuration
forecasters:
  eli:
     name: "Eli (AI 2027 author)"           # Your display name in plots. SAME means same as Nikola.
     color: "#900000"      # Pick a unique color (hex code)
     distributions:
       h_sat_ci: [0.5, 15.0]     # 80% CI for horizon at RE-Bench saturation in hours
       h_SC_ci: [0.05, 6]  # 80% initial time horizon needed to achieve SC once it's increased via other gaps being crossed, in months
       T_t_ci: [0.5, 18.0]          # Base doubling time in months
       v_algorithmic_sat_ci: [0.05, 0.6]   # Progress multiplier at saturation - 1
       v_algorithmic_SC_ci: [1.5, 40.0]   # Progress multiplier at SC - 1
       t_sat_ci: ["2025-09-01", "2031-01-01"]  # 80% CI for date when capability saturates
       d_ci: [0.25, 6]                        # SAME: Announcement delay in months
       p_superexponential: 0.45             # Probability that horizon length grows superexponential
       p_subexponential: 0.1               # Probability that horizon length grows subexponentially
       se_doubling_decay_fraction: 0.1     # If superexponential, fraction by which each doubling gets easier
       sub_doubling_growth_fraction: 0.1   # If subexponential, fraction by which each doubling gets harder
     algorithmic_slowdowns:
       # Format: [P(no slowdown), Lower bound months, Upper bound months]
       complex_engineering: [0.0, 0.5, 18]    # Engineering complexity delays
       feedback_loops: [0.0, 0.8, 45]     # Working without feedback
       project_coordination: [0.0, 0.5, 4]    # Project coordination overhead
       specialization: [0.0, 0.5, 6]       # Specialized skill requirements
       cost_and_speed: [0.0, 1, 48]        # Computational constraint
       other: [0.0, 1, 30]                 # Other algorithmic challenges
  eli_patched_rd_speedup:
     name: "Eli_patched_rd_speedup"           # Your display name in plots. SAME means same as Nikola.
     color: "#900000"      # Pick a unique color (hex code)
     patch_rd_speedup: True
     distributions:
       h_sat_ci: [0.5, 15.0]     # 80% CI for horizon at RE-Bench saturation in hours
       h_SC_ci: [0.05, 6]  # 80% initial time horizon needed to achieve SC once it's increased via other gaps being crossed, in months
       T_t_ci: [0.5, 18.0]          # Base doubling time in months
       v_algorithmic_sat_ci: [0.05, 0.6]   # Progress multiplier at saturation - 1
       v_algorithmic_SC_ci: [1.5, 40.0]   # Progress multiplier at SC - 1
       t_sat_ci: ["2025-09-01", "2031-01-01"]  # 80% CI for date when capability saturates
       d_ci: [0.25, 6]                        # SAME: Announcement delay in months
       p_superexponential: 0.45             # Probability that horizon length grows superexponential
       p_subexponential: 0.1               # Probability that horizon length grows subexponentially
       se_doubling_decay_fraction: 0.1     # If superexponential, fraction by which each doubling gets easier
       sub_doubling_growth_fraction: 0.1   # If subexponential, fraction by which each doubling gets harder
     algorithmic_slowdowns:
       # Format: [P(no slowdown), Lower bound months, Upper bound months]
       complex_engineering: [0.0, 0.5, 18]    # Engineering complexity delays
       feedback_loops: [0.0, 0.8, 45]     # Working without feedback
       project_coordination: [0.0, 0.5, 4]    # Project coordination overhead
       specialization: [0.0, 0.5, 6]       # Specialized skill requirements
       cost_and_speed: [0.0, 1, 48]        # Computational constraint
       other: [0.0, 1, 30]                 # Other algorithmic challenges
  # nikola:
  #   name: "Nikola"         # Display name in plots and legends
  #   color: "#003000"       # Unique color for this forecaster's data
  #   distributions:
  #     h_sat_ci: [0.5, 12.0]     # 80% CI for horizon at RE-Bench saturation in hours
  #     h_SC_ci: [0.05, 6]  # 80% CI for initial time horizon needed to achieve SC once it's increased via other gaps being crossed, in months
  #     T_t_ci: [0.5, 18.0]        # 80% CI for base horizon doubling time in months
  #     v_algorithmic_sat_ci: [0.05, 0.6]   # 80% CI for algorithmic progress rate at saturation - 1
  #     v_algorithmic_SC_ci: [1.0, 20.0]   # 80% CI for algorithmic progress rate at SC capability - 1
  #     t_sat_ci: ["2025-08-01", "2026-11-01"]  # 80% CI for date when capability saturates
  #     d_ci: [0.25, 6]                        # 80% CI for announcement delay in months
  #     p_superexponential: 0.4             # Probability that horizon length grows superexponentially (vs exponential)
  #     p_subexponential: 0.1              # Probability that horizon length grows subexponentially (vs exponential)
  #     se_doubling_decay_fraction: 0.1     # If superexponential, fraction by which each doubling gets easier
  #     sub_doubling_growth_fraction: 0.1   # If subexponential, fraction by which each doubling gets harder
  #   algorithmic_slowdowns:
  #     # Format for each slowdown: [Probability of no slowdown, Lower CI in months, Upper CI in months]
  #     complex_engineering: [0.0, 0.5, 18]    # Delays from increased engineering complexity
  #     feedback_loops: [0.0, 0.5, 18]         # Delays from working without feedback
  #     project_coordination: [0.0, 0.5, 3]    # Delays from coordinating multiple projects
  #     specialization: [0.0, 0.1, 2]        # Delays from need for specialized AI skills
  #     cost_and_speed: [0.0, 1, 36]         # Delays from computational costs and speed
  #     other: [0.0, 0.5, 18]                 # Other potential algorithmic delays
  # futuresearch:
  #   name: "FutureSearch"        # Your display name in plots
  #   color: "#4287f5"        # Pick a unique color (hex code)
  #   distributions:
  #      h_sat_ci: [0.5, 15.0]     # 80% CI for horizon at RE-Bench saturation in hours
  #      h_SC_ci: [0.05, 6]  # 80% initial time horizon needed to achieve SC once it's increased via other gaps being crossed, in months
  #      T_t_ci: [0.5, 18.0]          # Base doubling time in months
  #      v_algorithmic_sat_ci: [0.05, 0.6]   # Progress multiplier at saturation - 1
  #      v_algorithmic_SC_ci: [1.5, 40.0]   # Progress multiplier at SC - 1
  #      t_sat_ci: ["2025-09-01", "2031-01-01"]  # 80% CI for date when capability saturates
  #      d_ci: [0.25, 6]                        # SAME: Announcement delay in months
  #      b_ci: [1, 24]                       # SAME: Non-algorithmic slowdowns in months
  #      p_superexponential: 0.45             # Probability that horizon length grows superexponential
  #      p_subexponential: 0.1               # Probability that horizon length grows subexponentially
  #      se_doubling_decay_fraction: 0.1     # If superexponential, fraction by which each doubling gets easier
  #      sub_doubling_growth_fraction: 0.1   # If subexponential, fraction by which each doubling gets harder
  #   algorithmic_slowdowns:
  #     # Format: [P(no slowdown), Lower bound months, Upper bound months]
  #     complex_engineering: [0.0, 2.4, 48]    # Engineering complexity delays
  #     cost_and_speed: [0.0, 4.5, 36]        # Computational constraints
  #     feedback_loops: [0.0, 1.7, 58]        # Working without feedback
  #     project_coordination: [0.0, 0.7, 5.3]    # Project coordination overhead
  #     specialization: [0.0, 0.5, 4.7]       # Specialized skill requirements
  #     other: [0.0, 5.0, 100]                 # Other algorithmic challenges
  # daniel:
  #   name: "Daniel"
  #   color: "#FF5733"       # Chosen unique color for Daniel
  #   initial_conditions:
  #     h_sat: 8.0          # Expected horizon length at RE-Bench saturation (hours)
  #   distributions:
  #     h_SC_ci: [32.0, 1024.0]            # 80% CI for horizon needed to achieve SC
  #     T_t_ci: [10, 200]                   # 80% CI for horizon doubling time (days)
  #     v_algorithmic_sat_ci: [1.2, 2]        # 80% CI for algorithmic progress rate at saturation
  #     v_algorithmic_SC_ci: [2.5, 10]       # 80% CI for algorithmic progress rate at SC milestone
  #     t_sat_ci: ["2025-06-01", "2026-11-01"]  # 80% CI for date when RE-Bench saturation is hit
  #     d_ci: [30, 180]                     # 80% CI for announcement delay (days) [~1 month, 6 months]
  #     b_ci: [2, 24]                       # 80% CI for non-algorithmic slowdowns (months)
  #     p_superexponential: 0.5             # Probability that horizon length grows superexponentially
  #     p_subexponential: 0.1               # Probability that horizon length grows subexponentially
  #     se_doubling_decay_fraction: 0.1     # Fraction by which each doubling gets easier (if superexponential)
  #     sub_doubling_growth_fraction: 0.1   # Fraction by which each doubling gets harder (if subexponential)
  #   algorithmic_slowdowns:
  #     # Format for each slowdown: [Probability of no slowdown, Lower CI in months, Upper CI in months]
  #     complex_engineering: [0.0, 1, 12]    # Delays from increased engineering complexity
  #     feedback_loops: [0.0, 3, 36]           # Delays from feedback loops
  #     project_coordination: [0.0, 1, 4]      # Delays from coordinating multiple projects
  #     specialization: [0.0, 0.1, 4]            # Delays from need for specialized AI skills
  #     cost_and_speed: [0.0, 4, 36]           # Delays from computational costs and speed
  #     other: [0.0, 0.01, 10]                 # Other potential algorithmic delays

# Template for adding new forecasters
# Copy this section, uncomment, and modify the values
#   your_name:
#     name: "Your Name"        # Your display name in plots
#     color: "#4287f5"        # Pick a unique color (hex code)
#     initial_conditions:
#       h_sat: 8.0            # Expected horizon at saturation (hours)
#     distributions:
#       # All confidence intervals are 80% (10th to 90th percentile)
#       h_SC_ci: [32.0, 2048.0]            # Horizon needed for SC (hours)
#       T_t_ci: [45, 180]                   # Base doubling time (days)
#       v_algorithmic_sat_ci: [1.05, 1.6]   # Progress rate at saturation
#       v_algorithmic_SC_ci: [2.0, 20.0]   # Progress rate at SC
#       t_sat_ci: [3, 40]                   # Months until saturation
#       d_ci: [30, 180]                     # Announcement delay (days)
#       b_ci: [2, 24]                       # Non-algorithmic slowdowns (months)
#     algorithmic_slowdowns:
#       # Format: [P(no slowdown), Lower bound months, Upper bound months]
#       complex_engineering: [0.0, 2, 16]    # Engineering complexity delays
#       cost_and_speed: [0.0, 6, 36]        # Computational constraints
#       feedback_loops: [0.0, 4, 20]        # Working without feedback
#       project_coordination: [0.0, 1, 3]    # Project coordination overhead
#       specialization: [0.0, 0.5, 2]       # Specialized skill requirements
#       other: [0.0, 3, 12]                 # Other algorithmic challenges