PhD student at QUT in computational Bayesian statistics
Joined October 2022
- Tweets13
- Following900
- Followers72
- Likes433
Ryan Kelly retweeted
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Ryan Kelly retweeted
The past month has revealed a clear difference between three kinds of people who previously dismissed catastrophic risks associated with AI:
1. People who recognize that AI development and its associated risks are moving faster than they expected, acknowledge that their previous views may have reflected biases and heuristics such as anchoring, normalcy bias, and social conformity, and think proactively about how to correct for those tendencies moving forward.
2. People who recognize that AI development and its associated risks are moving faster than they expected, but frame these developments as largely unforeseeable given the evidence previously available. As a result, they remain one step behind the curve: acknowledging risks once they happen while continuing to dismiss risks that remain in the future.
3. People who fail to recognize that AI development and its associated risks are moving faster than they expected at all, and thus remain stuck in 2022 talking points: AI systems are nothing more than stochastic parrots, and the only harms worth discussing are algorithmic bias, copyright infringement, economic disruption, and so on.
We all want to think that we would fall into the first category, but in practice this kind of epistemic virtue is rare. Sincere gratitude to the people who make these updates and publicly acknowledge them. We need many more people to set this kind of example for the societal conversation about AI risks to go well.
Ryan Kelly retweeted
Overall, I’m very surprised at how little media coverage there’s been around the OpenAI / Hugging Face attack. It’s clearly one of the most important things to happen this year.
Ryan Kelly retweeted
New post: going into our investigation of the HF attack (before Black Hat), I was very wrong about what basically happened. This incident was far more serious than I expected, and far more serious than previous documented misalignment incidents. planned-obsolescence.org/p/t…
Ryan Kelly retweeted
Janet Egan has spent the last month back home in Australia talking to policymakers about AI.
I asked her: Any differences between how Aus and US policymakers are thinking about AI?
Janet: "The first thing that is so distinct between Washington, D.C. and Australia is how seriously people take AI."
"... There are a lot of very senior people in positions in [the Australian] government who still aren’t grappling with the realities of AI progress."
The geostrategic case for compute in Australia:
Janet Egan (@janet_e_egan) has been doing a bunch of quiet work trying to nudge Australian policymakers to think more seriously about AI.
She works at the Center for a New American Security in Washington, but was back home in Australia over the past month. Was lucky to catch her on the day before she returned to the US.
We pressure-test various strategic arguments for a large-scale compute industry and talk about the steps we would need to take to enable it.
Watch below, or wherever you get your podcasts!
Timestamps:
(0:00:00) – Introduction
(0:00:33) – Lessons from the AdBlue crisis
(0:09:02) – How seriously do Australian policymakers take AI?
(0:12:48) – How much does frontier access matter?
(0:28:55) – Compute-for-access
(0:43:08) – What does sufficient leverage look like?
(0:53:00) – What does AI sovereignty actually mean?
(1:03:15) – Taiwan, undersea cables and data centres as military targets
(1:10:30) – Could compute buy Australia influence over AI safety?
(1:15:19) – Verifying international AI agreements
(1:24:00) – Copyright: the red line?
(1:37:49) – The maximally ambitious path
Ryan Kelly retweeted
The Bayesian Workflow book is out in print now, just in time to distract you from *gestures all around*. Some links below to code and other resources:
Ryan Kelly retweeted
- i'm angry about this because i personally and for others want access to fable, and simultaneously believe anthropic's safeguards were sufficient and the US government badly misunderstood the information they were presented
- but in abstract this is in fact exactly what I want. it's heartening to see the USG treat artificial intelligence with the seriousness and immediacy it deserves. this kind of swift action is what might have a chance of saving us from unaligned RSI.
- but i also very much don't trust *this* government to handle this well, to take sane unilateral action, to chart any kind of correct path.
- and this escalates the global race enormously. this is as strong a signal as you can get to, not just China but the EU and even our closest allies, that the US will not be sharing this advantage. that if they want sovereignty they're going to have to fight for it
- obviously, that was always the case, and it was always going to happen eventually. but i don't think now was the time to send that signal. it would have been better to delay as long as possible.
very mixed feelings today
Ryan Kelly retweeted
"A Comprehensive Guide to Simulation-based Inference in Computational Biology" (by Xiaoyu Wang, Ryan P. Kelly, Adrianne L. Jenner, David J. Warne, Christopher Drovandi): arxiv.org/abs/2409.19675
Ryan Kelly retweeted
tweeting this so I can also find it again.
Pen and Paper exercises in ML *WITH SOLUTIONS*
By Michael Gutmann
arxiv.org/abs/2206.13446
Excited to see our paper accepted in TMLR. This work, a collaborative effort with David Nott, @dtfraz, @DavidJWarne, and @chris_drovandi, proposes robust sequential neural likelihood, a simulation-based inference method that is robust to model misspecification.
Misspecification-robust Sequential Neural Likelihood for Simulation-based Inference
Ryan P. Kelly, David J Nott, David Tyler Frazier, David J Warne, Christopher Drovandi.
Action editor: George Papamakarios.
openreview.net/forum?id=tbOY…
#models #likelihoods #like
Ryan Kelly retweeted
New pre-print by PhD student @RyanKel26511337 on how to make a recent neural approach to simulation-based inference robust to misspecification. @dtfraz @DavidJWarne @QUTDataScience
arxiv.org/abs/2301.13368
