Demand for AI alignment research compute

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

I want to get an idea of how much demand there is for a university-independent organization that manages a compute cluster for academic AI alignment groups and independent researchers. Currently I don’t know anybody who is willing to run such an organization, but if demand is large one could either actively look for people to run such a project or find an existing organization that is willing to take it on.

This idea is described in more detail here: https://www.alignmentforum.org/posts/A4djH6sc9vZq2AYBD/alignment-researchers-how-useful-is-extra-compute-for-you-1

I plan to publish summary statistics and keep the rest non-public; if you have concerns about confidentiality or anything else please mention them in the form or contact me at langosco.lauro@gmail.com

Name

Email address

Do you work on AI alignment / safety research? *

If you're not sure whether your research counts as alignment / safety, briefly mention your main research topics in 'add other'.

Yes, full-time

Yes, part-time

What organization are you part of? (If applicable)

Would easy and free access to more compute significantly help your research?

Not at all

A lot

Clear selection

How much time do you personally spend on managing compute? (hrs/week)

By this we mean installing and maintaining servers, updating OS, drivers, CUDA, etc, as well as time spent on procurement or hiring related to compute. An order of magnitude estimate is fine. If you are not involved in managing compute infrastructure because someone else in your organization does so please skip this question and answer the next question instead.

How much time does your organization spend on managing compute? (hrs/week)

By this we mean installing and maintaining servers, updating OS, drivers, CUDA, etc, as well as time spend on procurement or hiring related to compute. An order of magnitude estimate is fine. If you're not part of an organization (e.g. you work mainly as independent researcher), please skip this question.

How much compute could you productively use? (Per month).

In A100 (80GB) GPU-days per month. A rough estimate is fine.

2 (enough to train a ResNext101 model to 80% top-1 accuracy on ImageNet)

100

300 (enough to train AlphaFold - estimated)

1,000

10,000

30,000 (enough to train GPT-3)

Other:

Clear selection

How much compute do you currently have access to? (Per month).

In A100 (80GB) GPU-days per month. A rough estimate is fine.

2 (enough to train a ResNext101 model to 80% top-1 accuracy on ImageNet)

100

300 (enough to train AlphaFold - estimated)

1,000

10,000

30,000 (enough to train GPT-3)

Other:

Clear selection

Apart from the above, do you have any particular hardware requirements?

Alternatively, what sorts of workloads would you like to run?

Which one of these is the more important bottleneck for your work?

Access to compute

Engineering (e.g. setting up distributed training, maintaining servers)

Neither of these importantly limit the work that I do.

Clear selection

Anything else you want to tell us?

Clear form

Never submit passwords through Google Forms.

This content is neither created nor endorsed by Google. Report Abuse - Terms of Service - Privacy Policy

Forms