max "activating examples" loeffler
@maxsloef
Followers
2K
Following
20K
Media
245
Statuses
2K
researcher @goodfireai. helped make @websim_ai. ˈhaɪpəstɪʃᵊnd eɪkɔːzᵊl ˈtreɪdə, questing for a fragment of the eternal & sublime
sf
Joined September 2023
important question.
Which do you think the base model is happiest to find out they are.
0
0
1
RT @ericho_goodfire: I think we need new tools in the interp toolbox other than sparse dictionary learning techniques like SAEs, and SPD is….
0
3
0
sparse dictionary learning is great, but we know that more tools are needed - SPD is our first big bet in that direction. really excited to see where we can take this (and other big bets!).
(1/7) New research: how can we understand how an AI model actually works? Our method, SPD, decomposes the *parameters* of neural networks, rather than their activations - akin to understanding a program by reverse-engineering the source code vs. inspecting runtime behavior.
2
2
62
RT @voooooogel: The reviews are in: We recommend the OpenMind assistant wearable. As usual, we tried every assistant wearable on the marke….
0
13
0
wow people did NOT get this one, is this what it feels like to make an apple/potato joke?.
0
0
4
second week at an ai safety company and im already insider trading prediction markets.is this what it feels like to live your dreams? :').
1
0
16
now seems like a good time to announce that i've joined @GoodfireAI full time!. I had such a blast working on the R1 SAE and Paint with Ember projects, and am so proud to have contributed to the first ever CLT replication. can't wait to build more with this incredible team :).
New research update! We replicated @AnthropicAI's circuit tracing methods to test if they can recover a known, simple transformer mechanism.
9
4
181
this is the easily the best team i've ever worked with, and i feel so lucky to be here. join us! i promise you'll have fun :). also we've raised $59M, have just 15 people, are Anthropic's first corporate investment, and have more GPUs than we can use :))).
0
1
22
in order to get the bonus, please have the applicant add "max loeffler referral" plus your name in the "Why do you want to work at Goodfire?" section of the application. you can also refer yourself, in which case you can consider this an extra signing bonus!.
1
1
18
apply here! we have a $10k (pre-tax) referral bonus for employees, and i will pay you the entire post-tax amount (probably like 6.5-7k) if you refer a candidate we hire.
1
1
24
i'll pay you ~$7k for a successful hire!. we are aggressively hiring experienced ML & research engineers here at @GoodfireAI. let's solve interpretability!. link & referral bonus details below:.
8
7
230
RT @myra_deng: >be you.>work in HFT .>have existential dread .>see this tweet, wonder if your skills could be better used to make AGI safe….
0
14
0
thank you nasdaq for recognizing our leadership in "cloud infrastructure".
@GoodfireAI on the the Nasdaq tower! happy to be recognized in @Redpoint's 2025 report on the most impactful and fastest-growing infra companies
1
0
24
why on gods green earth does it cost $100/hr to reinforcement fine-tune o4-mini.
1
0
9
this is a really really cool idea, and i think it’s desperately needed!.
1
0
7
sipping on that shit that made Pocari sweat.
1
0
6
goodfire social is this thursday, come through!. if we're moots and you don't get accepted just DM me :).
0
0
5
very very proud to have contributed to this project! i have a lot of hope for novel interfaces that lean on interpretability - i hope this is the first of many.
We created a canvas that plugs into an image model’s brain. You can use it to generate images in real-time by painting with the latent concepts the model has learned. Try out Paint with Ember for yourself 👇
4
1
81