Andy supports the financial operations of CEA, having previously held a similar role at Effective Ventures Foundation, where he focused on accounting, reporting, and system improvement. Before that, he was a data analyst at Bloomberg L.P. and helped build the effective altruism community in Hong Kong. He holds a Bachelor of Science in Decision Analytics from the University of Hong Kong and a CIMA Advanced Diploma in Management Accounting. Andy has a keen interest in the effective operation of non-profits and improving transparency in the philanthropy sector.
Yufeng (Andy) Tao
Karma: 230
Hi Jesse, I think the information about this may not be available yet since the standard is still in its early stage. However, you may refer to the Country Champions, and Supporters page of the INPRF website for more information.
Hi Abraham, thanks for sharing your thoughts! I really appreciate you taking the time to read through the implementation guidance.
I’m especially curious about your point on the modified cash basis. I’m not familiar with this approach, and I can see the appeal of simplicity of this basis compared to the full accrual basis. My concern is whether this would create comparability problems:
Lack of standardisation: I don’t know if there is a uniform way to implement modified cash basis—each organisation could choose which items to treat as cash vs. accrual, making cross-organisation comparisons difficult.
Timing artifacts: Organisations with different payment cycles could show different expense patterns even with identical underlying activities. For example, an org that pays invoices immediately vs. one that batches payments quarterly could look artificially different.
Missing obligations: Without capturing accrued expenses or other provisions, we wouldn’t be able to see the org’s outstanding commitments, which seems important for understanding the true financial position.
You mentioned you’d be excited to see a standard that supports modified cash for audit purposes—which I think would be really interesting! Creating such a standard would presumably need to address these comparability challenges by defining exactly which modifications are acceptable and how to apply them consistently.
Hi Siobhan, thanks for your question!
I’m writing from my personal perspective here, so this doesn’t represent CEA’s official position. However, I think this is worth discussing internally, and I appreciate you raising it.
I’d expect that for any organisation considering adopting INPAS, there would need to be some time to properly understand the specific requirements—including how they relate to existing local regulatory reporting obligations and what the implementation process would look like.
This was a really great read. Thank you for sharing!
Data is here, in both human and machine-readable formats.
Thanks for sharing this! Flagging that the link can’t be opened: “Sorry, unable to open the file at present. Please check the address and try again.”
This is amazing to see, George! As a local who needs to try very hard to find authentic juicy tofu in Yunnan, I can attest to how delicious it is and feel excited about the possibility of popularising it elsewhere!
As someone who had never been to the UK and moved to Oxford for the first time a month ago, I can testify that Trajan House made me immediately feel at home. Jonathan warmly reached out to me offering help on personal life ops things such as bank accounts, SIM cards, and accommodations. Working in the same office room with Jonathan, I have no doubt that he is a lovely person to work with and a great friend to know. I’d be excited to see the office reaching its next level and lots of impacts to follow!
Thank you Aadit for sharing your journey and plan here as well!
For the estimated number of EAs within companies, HIP would be a good next step to reach out to. It is possible that some companies were accidentally omitted in the process and great to know your enthusiasm on starting a workplace group! Looking forward to your future updates if you feel comfortable sharing :)
Hey Cristina, thanks so much for pointing these out! Fully agreed here and hope others won’t go so far through a loss to realize these.
It was a life-changing experience working with you and Vaidehi. Both of you continue to inspire me more than you might think! Grateful to have known you :)
Thanks for the post, Jack!
I like the idea of more rigorously measuring the work done among AI governance organisations. Below are some example indicators people may consider, depending on the types of work.
[I used AI to brainstorm these based on orgs in https://aisafety.com/map. So apologies for any misclassifications.]
--
Policy research and analysis (GovAI, IAPS, CSET, LawAI, RAND, CLTR)
Uptake:
Cited in a named government consultation response, committee report, agency RFI or impact assessment
Invited to give formal evidence to a named legislature or agency
Count of unsolicited briefing requests from officials above a defined seniority
Adoption:
A definition, threshold or framing traceable to a named publication appears in draft or enacted text
Staff placed into government or standards roles, destination named
Direct advocacy (Encode, Americans for Responsible Innovation, Secure AI Project, ControlAI, Good Ancestors)
Uptake:
Bill introduced containing text the organisation drafted or substantially shaped
Co-sponsors added following engagement, or legislators publicly crediting the organisation
Coalition sign-ons, weighted by whether signatories were new to the issue
Outcome:
Bill progression through defined stages, each stage separately probability-weighted
Provision survives conference, amendment or reconciliation intact
Standards and technical governance (AI Standards Lab, SaferAI, Fathom, AVERI, METR, Apollo)
Uptake:
Submitted text accepted into a working draft at a named standards body
Adoption:
Evaluation protocol appears in a frontier lab’s published system card or safety framework
Count of labs voluntarily commissioning a third-party evaluation
A standard referencing the work is incorporated by reference into regulation
International convening and track II (SAIF, Simon Institute, Concordia AI, IASEAI)
Output:
Attendance by participant class rather than headcount: sitting regulators, frontier lab decision-makers, state-affiliated institutions
Uptake:
Bilateral follow-on engagements initiated within six months, reportable in aggregate
Outcome:
Joint statement or shared framework published by participants
Participants’ subsequent public positions move in the intended direction
Talent and field-building (MATS, BlueDot, ML4Good, Kairos, Talos, ERA)
Output:
Cohort size and completion rate
Outcome:
Placement into named target institutions at 12 and 24 months
Proportion reporting they would not otherwise have entered the field
Retention in the field at 24 months, and seniority progression
Watchdog and transparency (AI Lab Watch, The Midas Project, Palisade)
Uptake:
Documented response, correction or commitment from the named target
Outcome:
Verified policy change at a lab following publication
Finding cited in regulatory or litigation proceedings
Forecasting and epistemic infrastructure (Epoch, FRI, AI Futures Project, Metaculus)
Uptake:
Dataset or forecast cited in official documents or by named decision-makers
Outcome:
Published calibration score on the organisation’s own forecasts
--