Will a model costing >$30M be intentionally trained to be more mechanistically interpretable by end of 2027? (see desc)
29
149
Ṁ3.1KṀ590
2028
57%
chance
1D
1W
1M
ALL
Will any language model costing >$30M be intentionally trained with modifications to be more mechanistically interpretable by 2027?
If the training-scheme with modifications is more than a small amount better for capabilities than the [possibly next] capabilities-best training scheme at the time (available to the model trainers), it doesn't count for the purposes of this market.
Modifications to architecture, optimizer, etc. The modifications must be made with the intention to make the model more interpretable.
Costing >$30M is in my estimation, the cost does not need to be published.
I will not trade in this market.
Get Ṁ200 play money
Related questions
Will mechanistic interpretability be essentially solved for GPT-2 before 2030?
22% chance
Will it be possible to disentangle most of the features learned by a model comparable to GPT-3 this decade? (1k subsidy)
55% chance
SoAIR '23 5/10: Will the GenAI scaling craze see a group spend >$1B to train a single large-scale model by Oct 24?
28% chance
Will the GenAI scaling craze see a group spend >$1B to train a single large-scale model in the next 12 months?
44% chance
In the next 12 months will the GenAI scaling craze see a group spend >$1B to train a single large-scale model?
60% chance
Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?
44% chance
Will there be a for-profit interpretability company valued at more than $100M by 2030?
66% chance
AI: Will someone train a $10B model by 2030?
64% chance
AI: Will someone train a $1T model by 2080?
62% chance
Will $10,000 worth of AI hardware be able to train a GPT-3 equivalent model in under 1 hour, by EOY 2027?
16% chance