Allison
@adjectiveallison.com
software engineering - systems/infrastructure - llms - information retrieval - long term memory - brain/mind - writing - talking - thinking (too much)
Oh you meant still on bsky not still on Christmas. Ummm I don't open this app as much anymore. Maintaining 1 feed is more than enough for me.
December 26, 2024 at 3:01 PM
Oh you meant still on bsky not still on Christmas. Ummm I don't open this app as much anymore. Maintaining 1 feed is more than enough for me.
I'm an American these timezones things are wild
December 26, 2024 at 3:00 PM
I'm an American these timezones things are wild
Reposted by Allison
December 5, 2024 at 10:35 PM
I heard trav is a WILD partier. BIG parties
November 30, 2024 at 4:18 PM
I heard trav is a WILD partier. BIG parties
The first paragraph of your links refer to personal data. Your posts are public. People aren't scraping your birthday or your email address.
November 28, 2024 at 6:50 PM
The first paragraph of your links refer to personal data. Your posts are public. People aren't scraping your birthday or your email address.
Okay yes, this is very cool and you're probably right.
November 28, 2024 at 3:44 PM
Okay yes, this is very cool and you're probably right.
I guess durability or density? No idea, I didn't know about Microsoft using quartz.
November 28, 2024 at 2:46 PM
I guess durability or density? No idea, I didn't know about Microsoft using quartz.
This was my favorite one! Still no likes on the other site. Thanks for the validation!
November 27, 2024 at 12:23 AM
This was my favorite one! Still no likes on the other site. Thanks for the validation!
I think other routes have a higher chance at success in generalization. I think reasoning is an important direction to explore for the time being. I think the amount of practice the labs get with synth data gen and large scale RL with these models is useful. I'm not Aidan tho.
November 26, 2024 at 9:58 PM
I think other routes have a higher chance at success in generalization. I think reasoning is an important direction to explore for the time being. I think the amount of practice the labs get with synth data gen and large scale RL with these models is useful. I'm not Aidan tho.
I'm speechless 😯
Are you fr or actually just summoning him? sonnet 3.5 is better for most of my use cases but I still go back to opus 3.0 from time to time now. Similar reasons for when I want to talk to llama 405B. I love the big models. This feels like slander.
Are you fr or actually just summoning him? sonnet 3.5 is better for most of my use cases but I still go back to opus 3.0 from time to time now. Similar reasons for when I want to talk to llama 405B. I love the big models. This feels like slander.
November 26, 2024 at 9:56 PM
I'm speechless 😯
Are you fr or actually just summoning him? sonnet 3.5 is better for most of my use cases but I still go back to opus 3.0 from time to time now. Similar reasons for when I want to talk to llama 405B. I love the big models. This feels like slander.
Are you fr or actually just summoning him? sonnet 3.5 is better for most of my use cases but I still go back to opus 3.0 from time to time now. Similar reasons for when I want to talk to llama 405B. I love the big models. This feels like slander.
Lots to learn from humans for sure!
November 26, 2024 at 9:44 PM
Lots to learn from humans for sure!
I believe Aidan is just hypothesizing that RL is the path that openAI took to create the o1 model. Maybe alternative routes to reasoning beyond RL wouldn't have the downside of being less likely to generalize? I'm not sure.
November 26, 2024 at 9:44 PM
I believe Aidan is just hypothesizing that RL is the path that openAI took to create the o1 model. Maybe alternative routes to reasoning beyond RL wouldn't have the downside of being less likely to generalize? I'm not sure.