Deadline: 18 February 2024
Registration for the OpenAI Superalignment Fast Grants 2024 are now open. These grants aim to support technical research into the alignment and safety of superhuman AI systems, including weak to strong generalisation, interpretability, scalable supervision and much more.
With these grants, they are particularly interested in funding the following areas of research:
https://googleads.g.doubleclick.net/pagead/ads?client=ca-pub-7238845118856992&output=html&h=280&slotname=3344323800&adk=1370513923&adf=4293638591&pi=t.ma~as.3344323800&w=742&fwrn=4&fwrnh=100&lmt=1706832729&rafmt=1&format=742×280&url=https%3A%2F%2Fopportunitydesk.org%2F2023%2F12%2F27%2Fopenai-superalignment-fast-grants-2024%2F&fwr=0&fwrattr=true&rpe=1&resp_fmts=3&wgl=1&uach=WyJXaW5kb3dzIiwiMTAuMC4wIiwieDg2IiwiIiwiMTA2LjAuNDk5OC42MSIsbnVsbCwwLG51bGwsIjY0IixbWyJOb3RfQSBCcmFuZCIsIjguMC4wLjAiXSxbIkNocm9taXVtIiwiMTIwLjAuNjA5OS4yMTciXSxbIk9wZXJhIEdYIiwiMTA2LjAuNDk5OC42MSJdXSwwXQ..&dt=1706869811993&bpp=1&bdt=746&idt=257&shv=r20240131&mjsv=m202401300101&ptt=9&saldr=aa&abxe=1&cookie=ID%3D982bccc1650bbfbd%3AT%3D1702979975%3ART%3D1706869577%3AS%3DALNI_MYUpDHIHjCvdAd9qpjwkIAoHR_Dmw&gpic=UID%3D00000ce6a34beb5d%3AT%3D1702979975%3ART%3D1706869577%3AS%3DALNI_Mbd8_6PMZZnaNxdD61mpTbhPuKlmg&eo_id_str=ID%3D41a60e3297e4b646%3AT%3D1706710309%3ART%3D1706869577%3AS%3DAA-AfjbY6WJEKbhZwY_O3PaRUgmV&prev_fmts=0x0%2C728x90&nras=1&correlator=5190292105312&frm=20&pv=1&ga_vid=669982640.1702979975&ga_sid=1706869812&ga_hid=1251832202&ga_fc=1&u_tz=-60&u_his=1&u_h=768&u_w=1366&u_ah=728&u_aw=1366&u_cd=24&u_sd=0.9&dmc=8&adx=169&ady=1110&biw=1455&bih=696&scr_x=0&scr_y=0&eid=44759876%2C44759927%2C44808398%2C31080779%2C31080818%2C95322183%2C95323006&oid=2&pvsid=3972707384887977&tmod=95914282&uas=0&nvt=1&ref=https%3A%2F%2Fwww.google.com%2F&fc=1920&brdim=0%2C0%2C0%2C0%2C1366%2C0%2C1366%2C728%2C1472%2C696&vis=1&rsz=%7C%7CEebr%7C&abl=CS&pfx=0&cms=2&fu=1152&bc=31&bz=0.93&psd=W251bGwsbnVsbCxudWxsLDNd&ifi=3&uci=a!3&btvi=1&fsb=1&dtd=261
- Weak to strong generalisation: Humans will be weak supervisors in relation to superhuman models. Can we understand and control how strong models generalise from weak supervision?
- Interpretability: How can we understand the internal aspects of the model? And can we use this to, for example, build a lie detector with AI?
- Scalable supervision: How can we use AI systems to help humans evaluate the results of other AI systems in complex tasks?
- Many other research directions, including but not limited to: honesty, fidelity to the chain of thought, adversarial robustness, evaluations and testbeds, and much more.
Subsidies
https://googleads.g.doubleclick.net/pagead/ads?client=ca-pub-7238845118856992&output=html&h=186&slotname=3100905470&adk=2596809427&adf=3674528092&pi=t.ma~as.3100905470&w=742&fwrn=4&lmt=1706832729&rafmt=11&format=742×186&url=https%3A%2F%2Fopportunitydesk.org%2F2023%2F12%2F27%2Fopenai-superalignment-fast-grants-2024%2F&wgl=1&uach=WyJXaW5kb3dzIiwiMTAuMC4wIiwieDg2IiwiIiwiMTA2LjAuNDk5OC42MSIsbnVsbCwwLG51bGwsIjY0IixbWyJOb3RfQSBCcmFuZCIsIjguMC4wLjAiXSxbIkNocm9taXVtIiwiMTIwLjAuNjA5OS4yMTciXSxbIk9wZXJhIEdYIiwiMTA2LjAuNDk5OC42MSJdXSwwXQ..&dt=1706869811994&bpp=1&bdt=747&idt=333&shv=r20240131&mjsv=m202401300101&ptt=9&saldr=aa&abxe=1&cookie=ID%3D982bccc1650bbfbd%3AT%3D1702979975%3ART%3D1706869577%3AS%3DALNI_MYUpDHIHjCvdAd9qpjwkIAoHR_Dmw&gpic=UID%3D00000ce6a34beb5d%3AT%3D1702979975%3ART%3D1706869577%3AS%3DALNI_Mbd8_6PMZZnaNxdD61mpTbhPuKlmg&eo_id_str=ID%3D41a60e3297e4b646%3AT%3D1706710309%3ART%3D1706869577%3AS%3DAA-AfjbY6WJEKbhZwY_O3PaRUgmV&prev_fmts=0x0%2C728x90%2C742x280&nras=1&correlator=5190292105312&frm=20&pv=1&ga_vid=669982640.1702979975&ga_sid=1706869812&ga_hid=1251832202&ga_fc=1&rplot=4&u_tz=-60&u_his=1&u_h=768&u_w=1366&u_ah=728&u_aw=1366&u_cd=24&u_sd=0.9&dmc=8&adx=169&ady=1737&biw=1455&bih=696&scr_x=0&scr_y=0&eid=44759876%2C44759927%2C44808398%2C31080779%2C31080818%2C95322183%2C95323006&oid=2&pvsid=3972707384887977&tmod=95914282&uas=0&nvt=1&ref=https%3A%2F%2Fwww.google.com%2F&fc=1920&brdim=0%2C0%2C0%2C0%2C1366%2C0%2C1366%2C728%2C1472%2C696&vis=1&rsz=%7C%7Cebr%7C&abl=CS&pfx=0&fu=1152&bc=31&bz=0.93&psd=W251bGwsbnVsbCxudWxsLDNd&ifi=4&uci=a!4&btvi=2&fsb=1&dtd=338
- They are offering grants of between US$ 100,000 and US$ 2 million to academic laboratories, non-profit organisations and individual researchers.
- For postgraduate students, they are sponsoring an OpenAI Superalignment Fellowship of US$ 150,000 for one year: US$ 75,000 in scholarship and US$ 75,000 in computer and research funding.
Eligibility
- Open to academic laboratories, non-profit organisations and individual researchers.
- No previous experience working on alignment is necessary; they are actively looking to support researchers who are enthusiastic about working on alignment for the first time.
Application
The deadline for applications is 18 February 2024.
