Bellairs Invitational Workshop on Contemporary, Foreseeable and Catastrophic Risks of Large Language Models

29 March 2024 -- 5 April 2024

Program Reading Participants Venue Travel

Artificial intelligence in the form of Large Language Models (LLMs) has advanced more rapidly than most researchers expected. As a result, estimates of how long it will take to develop human-level AI have shortened, accelerating speculation about the long-term dangers of AI. Far future predictions are valuable, but it is at least as important to engage with the concrete risks AI poses now. Especially since our capacity to solve long-term dangers will be determined by immediate risks: effective alignment algorithms must at least work for the easy cases we encounter at present. And, to make good collective decisions for the future, we must ensure that short-term proliferation of AI technologies does not irrevocably harm our economies and societies.

The objective of this workshop is to make progress on foreseeable and quantifiable risks of AI on the five year time scale. We will convene experts from diverse domains such as natural language processing, machine learning, safety, privacy, and law within a singular forum to understand the concrete, demonstrable risks of LLMs. Together, will examine strategies for mitigating these risks and pinpoint areas that need further exploration and development. Foremost, we seek to ground the discourse around LLM risks in a pragmatic and actionable framework.

Scientific Program

The institute will be open from March 29 to April 5.

The scientific program will take place from March 31 to April 4.

Each day of the workshop will consist of:

Day Time Topic Leads

Overview, Social biases in LLMs, Bias mitigation, Jailbreaking

Siva Reddy

7:30pm−9pm Alignment Maja Trębacz
9:30am−noon Uncertainty Communication Sylvie Delacroix, Neil Lawrence

7:30pm−9pm Explainability, Interpretability, Trust Ana Marasović
9:30am−noon Watermarking, Responsible Deployment Boaz Barak, Rich Zemel
7:30pm−9pm Privacy, Security Nicolas Papernot
9:30am−noon Normativity, Human Evolution and LLMs Gillian Hadfield
7:30pm−9pm Data Governance Leandro Von Werra, Harm de Vries
9:30am−noon Policy, Miscellaneous Jess Montgomery
7:30pm−9pm Conclusion, Outcomes Siva Reddy

Background Reading

Participants are encouraged to consult the following references in advance of the workshop:



The workshop will be held at the Bellairs Research Institute of McGill University, Holetown, St. James, Barbados.

For accommodation pricing, see the official page.


The Most Important House Rules

Kitchen and Food

Showers and Sand

Locked Doors and Valuables


Bellairs Survival Hints

Food and Snacks

Beach, Sun, Snorkeling, and SCUBA diving



Flying in

Please see the Barbados Official Travel Protocols for the rules that are currently in place on the island.

As of January 10, 2023, that site said "Effective midnight, Thursday September 22, 2022, Barbados will discontinue all COVID-19 related travel protocols. Therefore, there will be no testing requirements for entering Barbados whether you are vaccinated or unvaccinated."

Details for travel from the airport will be provided by email.

Map of Bellairs

For questions please contact