Posts

Thinking About Propensity Evaluations 2024-08-19T09:23:55.091Z
A Taxonomy Of AI System Evaluations 2024-08-19T09:07:45.224Z
Review of METR’s public evaluation protocol 2024-06-30T22:03:08.945Z
List of projects that seem impactful for AI Governance 2024-01-14T16:53:07.854Z

Comments

Comment by JaimeRV (jaime-raldua-veuthey) on ARENA 4.0 Impact Report · 2024-12-12T08:38:21.626Z · LW · GW

Thanks for sharing this! Great to see the impact of ARENA!

According to the OpenPhil public grant[1] this iteration of Arena got £245,895, and with this you were able to achieve the points mentioned in this post right?

Also it is great to hear that there are 4 new people working in AIS thanks to the program! It would be nice to know how did you manage it (and what was the counterfactual). Getting 4 people through full hiring processes within 4 weeks seems impresive, did you manage because they got jobs at orgs who were also at LISA? or there were other networking effects or other factors that made this possible?

[1] https://www.openphilanthropy.org/grants/alignment-research-engineer-accelerator-ai-safety-technical-program-2024/

Comment by JaimeRV (jaime-raldua-veuthey) on Is cybercrime really costing trillions per year? · 2024-09-27T09:14:53.868Z · LW · GW

I found this useful https://impact.economist.com/perspectives/technology-innovation/measuring-cost-cybercrime/article/what’s-number-estimating-cost-cybercrime

Comment by JaimeRV (jaime-raldua-veuthey) on jacquesthibs's Shortform · 2024-08-14T10:22:17.192Z · LW · GW

I have been using sider for a few weeks and found it pretty helpful:

Setup:

  • use gpt4o-mini which is basically free and faster than doing anything in Claude or ChatGPT
  • mostly for papers and LW/EAF articles
  • I have a shortcut to add "https://r.jina.ai/" to the url before to convert to markdown and then I just ctrl+A the entire page and chat
  • For privacy reasons I have only allowed the extension in https://r.jina.ai/* and https://www.youtube.com/*
  • I use similar prompts than Jacques. Some additional ones: -- Justify your previous answers citing the from original text -- Challenge my knowledge (here I have a longer promt where it asks me to du stuff like draw a mindmap, answer questions,...)
  • I also have it with (external) whisper cause often I think better outloud

Pros:

  • Fast
  • Basically free
  • Way easier to digest and interact with dry papers/articles
  • Customazible prompts for the conversation which make workflow faster cause you only have to click
  • For youtube as a first filter

Cons:

  • gpt40-mini (at least) hallucinates a bunch so you often have to ask to justify the answers
  • (as with all the chatbots) you shall take the responses with a grain of salt, be very specific with your questions and reread the original relevant sections to double check.

Other:

  • IMO if you end up integrating something like this in LW I think it would be net positive. Specially if you can link it to @stampy or similar to ask for clarification questions about concepts, ...
Comment by JaimeRV (jaime-raldua-veuthey) on jacquesthibs's Shortform · 2024-08-11T17:07:25.485Z · LW · GW

I used to use that one but I moved to Sider: https://sider.ai/pricing?trigger=ext_chrome_btm_upgrd it works in all the pages, including youtube. For Papers and articles I have shortcut to automatically modify the url (adding the prefix "https://r.jina.ai/") so you get the markdown and then do Sider on that. With gpt4o-mini it is almost free. Also nice is Sider is that you can write your own prompt templates

Comment by JaimeRV (jaime-raldua-veuthey) on Announcing the Double Crux Bot · 2024-07-10T11:23:08.555Z · LW · GW

Cool idea! thanks for making this! Do you happen to have also a Telegram bot for it?

Comment by JaimeRV (jaime-raldua-veuthey) on AI Safety Chatbot · 2024-01-03T16:23:10.165Z · LW · GW

I think at thumbs up/down with a field to enter feedback would be very helpful, but there is an open issue already for that https://github.com/StampyAI/stampy-chat/issues/35

Comment by JaimeRV (jaime-raldua-veuthey) on AI Safety Chatbot · 2024-01-03T16:19:21.560Z · LW · GW
  1. https://chat.openai.com/g/g-O6KK4ERZz-qaisi is a customer GPT that uses the Q&A from aisafety.info. https://chat.aisafety.info/ shows the sources more accurately