softmaxresearch Profile Banner
Softmax Profile
Softmax

@softmaxresearch

Followers
1K
Following
75
Media
10
Statuses
46

Softmax's mission is to scale organic alignment. We approach this problem with multi-agent reinforcement learning population-based simulations.

San Francisco, CA
Joined February 2025
Don't wanna be here? Send us removal request.
@softmaxresearch
Softmax
6 months
Is a monthly cadence right for this? So far, the experiment seems successful. But we are at the very dawn of organizational metadesign. Maybe it should be 4 days and Cooldown Fridays. Or maybe there should be two cooling months per year. We run Softmax as a living experiment.
1
0
23
@softmaxresearch
Softmax
6 months
During Annealing Week, we aren’t trying to make progress against our goals. Instead, we care about simplifying things. Removing steps. Killing processes. Deleting code. Replacing two features with one. Cutting meetings. Pruning the list of channels. Reducing company complexity.
2
0
30
@softmaxresearch
Softmax
6 months
This pattern is everywhere for learning systems. Curiously it’s usually 4:1 ratio of heating to cooling. The business cycle (before MMT) was 4:1 years of boom:bust. Bulk:cut in training is often recommended around 4:1 month cycles.
1
0
19
@softmaxresearch
Softmax
6 months
It’s Annealing Week at Softmax! Humans are awake for 16 hours learning, cooling for 4 hours in light sleep, and in deep sleep for 4. An organic mental annealing cycle, heating to cooling. At Softmax, we do the same. It’s four weeks sprinting towards goals, one week consolidating.
1
2
82
@softmaxresearch
Softmax
6 months
Our little Cogs grow up so fast. Cogbert has never seen this exact production chain before, but with only a couple missteps he begins to execute it correctly. Our in-context learner takes its first baby steps!
10
7
138
@softmaxresearch
Softmax
6 months
If you’ve written interactive prompts that help guide the user through making a plan or giving feedback or documenting their thought process, what have you learned doing it? What are the very best active process prompts you’ve made or used, and what made them great?
5
1
42
@softmaxresearch
Softmax
6 months
We are building organic alignment at Softmax. Not just with reinforcement learning, but within our company we try to use these same principles for our work. We are implementing this as an organizational operations system (OrgOS), a prompt library covering our internal processes.
6
2
93
@softmaxresearch
Softmax
7 months
InstaDeep, Africa’s foremost AI frontier lab, is doing some of the most compelling MARL work in the world. Sable is a genuine breakthrough. Worth checking out.
@instadeepai
InstaDeep
7 months
We're proud to be presenting our latest research at ICML, 2025: “Sable: a Performant, Efficient and Scalable Sequence Model for MARL” 🧵
0
2
8
@softmaxresearch
Softmax
7 months
Fechner’s Elements of Psychophysics is the latest addition to the Softmax library
0
0
6
@softmaxresearch
Softmax
7 months
Coming soon: BE NOT AFRAID
4
1
51
@softmaxresearch
Softmax
8 months
Our CEO, Emmett Shear, appeared on BuzzRobot and shared a bit more about our vision of the future
@sopharicks
Sophia
8 months
The problem that @eshear is working on deeply resonates with me: How to align AI and humans together so both see each other as part of their tribe. This doesn't mean aligning AI to human preferences, which is what AI labs seek to do today, by imposing a system of control on AI.
0
1
11
@chris_percy
Chris Percy
9 months
Wonderful to be invited to the @softmaxresearch research community day yesterday - my lightning talk and unconference session were about artificial minds and the difficulties in getting 'complex' consciousness out of a stepwise algorithm...
1
2
11
@softmaxresearch
Softmax
9 months
OH at the office: “What’s GitHub? Oh, it’s like Facebook for nerds”
0
0
3
@softmaxresearch
Softmax
10 months
Our CEO, Emmett Shear, gave a talk on alignment protocols: the engineered ways that parts communicate in order to align their trajectories. https://t.co/7tXKqAATde
2
0
33
@softmaxresearch
Softmax
10 months
tired: inductive bias wired: forgetting bias inspired: coherence bias
0
3
25
@softmaxresearch
Softmax
10 months
Wishing all learning agents a happy Easter
1
2
12
@softmaxresearch
Softmax
10 months
The Softmax Arcana of Choice and Perception Element: Air Numerology: 6 Astrological Parallel: Gemini (Duality, contextual truth) Associated Deck Archetype: The Lovers (many thanks to ChatGPT, our Tarot guide)
1
1
17
@softmaxresearch
Softmax
10 months
I found "Illuminating search spaces by mapping elites" (MAP-elites) phenomenally inspiring. Instead of trying to find a globally optimal solution, it turns out better to find every locally optimal solution! The power of maintaining diversity. https://t.co/04TGcPs4Qb
0
0
9
@softmaxresearch
Softmax
10 months
|>frontier-friday:activate Friday night: the optimal time to share our favorite papers from the frontiers of learning, intelligence, and alignment in all their manifold forms.
3
1
14