This market predicts whether OpenAI will grant early access to their GPT-5 model to any alignment-focused organizations by December 31, 2024.
Resolves YES if:
OpenAI officially announces or confirms that they have granted early access to the GPT-5 model to any alignment-focused organization on or before December 31, 2024.
There is strong evidence that an alignment organization has been granted early access before December 31, 2024. Then the market will be extended by up to 6 months to June 1, 2025 pending confirmation.
Resolves NO if:
No alignment organization is publicly announced to have received early access to the GPT-5 model from OpenAI by December 31, 2024.
Resolves as NA if:
OpenAI ceases to exist, or the listed organizations merge, dissolve, or undergo significant restructuring, rendering the original intent of the market unclear or irrelevant.
"GPT-5" refers to the next major release of OpenAI's generative pre-trained transformer model following GPT-4, or any equivalent successor model that is designed for natural language understanding and generation. Must be externally designated GPT-5, so e.g. GPT-4.5 would not count even if employees refer to it as GPT-5 or internal code is leaked that refers to it as GPT-5. However, code on public-facing websites is strong evidence.
"Strong Evidence" means journalistic reporting alluding to credible sources, statements by employees or former employees, or similar.
"OpenAI" mainly means OpenAI, but additionally any executive or heavy investor is allowed to make the announcement. So if Microsoft makes the announcement, it will count. If OpenAI is acquired, the name refers to the acquirer. If OpenAI dissolves, this market resolves NA.
"Alignment-focused Organization" refers to Anthropic, Redwood Research, Alignment Research Center, Center for Human Compatible AI, Machine Intelligence Research Institute, or Conjecture. Additional examples may be added - if there is a dispute, a poll may be taken. A public release of early access counts as granting early access to these organizations. A leaked release also counts, as long as it is confirmed to be GPT-5. "Red-teaming" of capabilities counts, as long as testing is driven by the alignment organization and not by OpenAI.
"Early access" refers to providing any of the listed organizations with access to the GPT-5 model, its weights, or associated APIs before it becomes publicly available or accessible to the general user base.
Terms can be adjusted within one week after market creation. After that, terms can only be refined to have narrower meanings or to have additional examples added.
Only Microsoft (and perhaps a few researchers) got early access to GPT-4, and it was declared only a few weeks before release.
Given our predictions on this market, I think this is wildly overrated.