grantmaking.ai Launch Round
CareerMap is an interactive career discovery tool, designed to make non-obvious career paths visible. The aim is to offer talent from outside the standard Western EA-adjacent circles a visibility to the field, career options and pathways to these through tooling and interaction with others already in the field.
In that sense, I view CareerMap as an early infrastructure for addressing the pipeline problem of AI Safety space in particular.
High level, rough breakdown:
-
$500: LLM API credits and data processing for role extraction.
-
$2,000: LLM costs for for continued development of CareerMap (Path-mapping logic, UI and personalised support features for the community)
-
$2,000: Unrestricted usage of LLM features in the tool for the community for up to 6 months (assuming less than 300 calls per day or about 30 visits per day)
-
$7,000: Coordinator time + access stipends for underserved talent).
__________
Fund use further details:
The grant money will be used to scale the tool, its feature sets as well as adjacent tool-led experiments to bring talent into high-impact AI safety and alignment roles.
Below are some illustrative deliverables I’m working towards (these may evolve with time; captures high level view of current plan)
-
Path-mapping engine development (dev time)
- Working routes + milestones + time-estimates between a starting role and a selected destination role. Currently, my processing flow is to break the career-paths of each profile into atomic-hops, normalise these hops across the profile dataset and then stitch the path user asks for, from these atomic-hops.
But the data messiness is challenging. I'm hoping to see improvement as I have more data or better processing.
- Working routes + milestones + time-estimates between a starting role and a selected destination role. Currently, my processing flow is to break the career-paths of each profile into atomic-hops, normalise these hops across the profile dataset and then stitch the path user asks for, from these atomic-hops.
-
Role-identity extraction & normalization
- More AI-safety-specific roles correctly represented in the dataset, pulled from the existing 604k-record corpus that was more focussed on non-AI roles. Also, some normalization needs to be improved for roles such as “Founder” - which have wildly different requirements across domains. And there are so many more such data quality challenges.
-
LLM API costs ("Look harder" calls)
- Route generation beyond the static dataset
-
New features and programs
- Building and running the "curated intros" experiment beyond initial cohorts to facilitate low-stakes exposure between professionals and existing AI safety networks.