The Community

Build AI that belongs to us.

We’re just getting started. Become a founding member to follow, test and help shape suum-exse as it moves from outside APIs, to our own hardware, to a model of our own. Chat and share ideas in our own space, on our own site, as the community grows.

See the roadmap: our own models, 1M to 1T

  • Build log Public
    Notes from us as we build: Level I to Level III, our own models and everything in between. Anyone can read it; only we post.
  • Lab
    Early builds and experiments. Try them, break them, tell us.
  • General
    Say hello. Anything suum-exse.
  • The build
    Models, hardware and the road from outside APIs to a model of our own.
  • Games
    ARC-2055 and whatever comes next: runs, tips and ideas.
  • Ideas and feedback
    Vote on what we build next and suggest ideas. We answer the top ones. Bugs in a game go in its bug form.
  • AMA
    Questions for our next AMA session. Vote up the ones you want answered first.

Membership · starts today

Community

$15 / month

or $150 a year — two months free

  • Member chat: every channel
  • Shared tools and resources, and early builds as they become available
  • A vote on what we build next, and a place to ask us questions
  • Occasional members’ newsletters, by email

Membership is for community and shared resources. We may occasionally share member-only instructional videos, separate from our public social posts. Members can find our Mark I and Mark II mini-courses in The Build. Videos have no fixed release schedule.

Already a member? Sign in. 18+. Renews until you cancel. Community terms. Compare with a reservation.

Public

Build log

Notes from us as we build: Level I to Level III, our own models and everything in between. Anyone can read it; only we post.

Mark II · The 10M training run

We trained Mark II from scratch in 54 minutes. On the same 200 reserved stories, prediction loss fell from 1.626 bits per byte for Mark I to 0.502 for Mark II. On 50 fresh openings, Mark II reached its end signal 44 times within the length limit, compared with 32 for Mark I. Some continuations still repeated phrases or lost track of the plot.

Training details
Parameters
10,026,112
Started
11 Nov 2026
Finished
12 Nov 2026 · 54 minutes
Hardware
NVIDIA GeForce GTX 1070 · 8 GB VRAM
Trained on · text
TinyStories · 437,380 complete cleaned stories · 389,555,204 text characters
Trained on · tokens
99,999,744 training predictions
Vocabulary
4,096 byte-level BPE tokens, including 3 special markers
Weights
Open (on request)
What it’s for
Continue short stories and compare Mark I and Mark II on the same opening.
To earn the next step
Beat Mark I on shared tests for text prediction, plot continuity, repetition, and story endings.

MK1 · The 1M training run

We trained our first model from scratch. After 2,400 optimizer steps, validation loss fell from 4.637 to 1.125. On 200 reserved test stories, loss was 1.128 and perplexity was 3.09. MK1 can continue a story, though it still misspells words and sometimes loses the plot.

Training details
Parameters
1,005,536
Started
22 Oct 2026 · project kickoff
Finished
10 Nov 2026 · training run
Hardware
Intel Core i5-6500 CPU @ 3.20 GHz
Trained on · text
TinyStories · 3,686 cleaned stories · 3,122,441 characters
Trained on · tokens
3,129,813 corpus tokens · 3,686,400 training predictions including repeats
Vocabulary
99 tokens · 97 characters and 2 story markers
Weights
Open (on request)
What it’s for
Prove the pipeline end to end, from preparing stories to training, testing, and generating text.
To earn the next step
Train and test cleanly, verify browser generation, and share what we learned in the Build log.