allie lawsen
@lexlawsen.bsky.social
AI grantmaking at Coefficient Giving
Previously 80,000 Hours
lawsen.substack.com
Previously 80,000 Hours
lawsen.substack.com
Anthropic before deploying a new model
May 9, 2025 at 8:43 AM
Anthropic before deploying a new model
xAI before releasing a new model
May 9, 2025 at 8:22 AM
xAI before releasing a new model
Google before deploying a new model
May 9, 2025 at 8:11 AM
Google before deploying a new model
OpenAI before deploying a new model
May 9, 2025 at 7:44 AM
OpenAI before deploying a new model
Interesting...
Proxy-generator is here in case you're still using the other place: substack-proxy.glitch.me
Proxy-generator is here in case you're still using the other place: substack-proxy.glitch.me
April 24, 2025 at 3:42 PM
Interesting...
Proxy-generator is here in case you're still using the other place: substack-proxy.glitch.me
Proxy-generator is here in case you're still using the other place: substack-proxy.glitch.me
If only this was the AI that had been asked for tariff setting advice.
April 9, 2025 at 10:32 PM
If only this was the AI that had been asked for tariff setting advice.
What does it take for people to actually, reliably use a whistleblowing function? 1/8
March 24, 2025 at 1:05 PM
What does it take for people to actually, reliably use a whistleblowing function? 1/8
Just published a new post on the Claude Projects I'm currently maintaining. I've had a lot of interest in how I'm using AI, so wrote this breakdown.
March 6, 2025 at 11:45 AM
Just published a new post on the Claude Projects I'm currently maintaining. I've had a lot of interest in how I'm using AI, so wrote this breakdown.
3. It's a direct test of whether an AI can rapidly develop and execute long-horizon plans in novel environments. This capability seems core to loss of control concerns, especially *active* loss of control as defined here.
February 26, 2025 at 1:41 PM
3. It's a direct test of whether an AI can rapidly develop and execute long-horizon plans in novel environments. This capability seems core to loss of control concerns, especially *active* loss of control as defined here.
1. I'm unironically excited about Anthropic's Pokémon eval. It's a step towards the kind of thing I was hoping we'd get in response to this section of our recent RFP.
February 26, 2025 at 1:41 PM
1. I'm unironically excited about Anthropic's Pokémon eval. It's a step towards the kind of thing I was hoping we'd get in response to this section of our recent RFP.
1. I wrote this a while ago, but it's come up in a few conversations since so it seemed worth a short thread.
February 26, 2025 at 11:42 AM
1. I wrote this a while ago, but it's come up in a few conversations since so it seemed worth a short thread.
1/8 Really happy to share Open Phil’s new Request For Proposals to improve AI capability evaluations. It's been a big project for me and my colleague Catherine. Getting evaluations right is crucial — and we're ready to fund serious work to make that happen. 🧵
February 6, 2025 at 7:22 PM
1/8 Really happy to share Open Phil’s new Request For Proposals to improve AI capability evaluations. It's been a big project for me and my colleague Catherine. Getting evaluations right is crucial — and we're ready to fund serious work to make that happen. 🧵
December 14, 2024 at 9:25 AM
OK, who the hell is Jacob and what does his Pie shop have to do with Pot Limit Omaha?
December 10, 2024 at 10:42 PM
OK, who the hell is Jacob and what does his Pie shop have to do with Pot Limit Omaha?
Teething problems with my first test...
December 5, 2024 at 8:28 PM
Teething problems with my first test...
December 4, 2024 at 12:21 PM
Now she's watching football
November 26, 2024 at 9:04 PM
Now she's watching football
Playing with logo designs for the AI Governance and Policy team.
What I like about this one is how original it is. Completely unlike any existing logo.
What I like about this one is how original it is. Completely unlike any existing logo.
November 24, 2024 at 12:11 PM
Playing with logo designs for the AI Governance and Policy team.
What I like about this one is how original it is. Completely unlike any existing logo.
What I like about this one is how original it is. Completely unlike any existing logo.
Red sky at night
Shepherd's delight
Blue sky is dawning
Still no good morning :(
Shepherd's delight
Blue sky is dawning
Still no good morning :(
November 20, 2024 at 11:31 PM
Red sky at night
Shepherd's delight
Blue sky is dawning
Still no good morning :(
Shepherd's delight
Blue sky is dawning
Still no good morning :(
I'll balance it out by posting a picture which is going to be horribly cropped in preview and then get low engagement as a result.
November 19, 2024 at 11:37 PM
I'll balance it out by posting a picture which is going to be horribly cropped in preview and then get low engagement as a result.
