[[🚘def(tesla)]] | - | [iai_g](https://github.com/Data4DM/BayesSD/discussions/174) BayesWorkflow-HMC | [iai_l](https://github.com/Data4DM/BayesSD/discussions/174) Probabilistic Programming-SMC | [o4s_l](https://github.com/Data4DM/BayesSD/discussions/159)ops for startup<br>[o4e_g](https://github.com/Data4DM/BayesSD/discussions/161) ops for entrepreneur + ecosystem | [[flagship_pioneering]] | | ---------------------------------- | ------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | agenet level | statistical modeler with dgp (including prior knowledge and likelihood), algorithm, observed data | system with hardware and software component | individual+ai<br> | small exploration team to a fully independent venture with external leadership.<br><br> | | phase0 | | | | exploration - Aligns with the setup phase where the "agent level" is established. In Flagship's model, this is where they brainstorm about opportunities within emerging areas of science and identify a "venture hypothesis." This maps to the initial setup where statistical models, systems, or individual+AI foundations are established. | | phase1 (quality measure) | learn step size $\epsilon$ | structure learning $P_\mathcal{L}(s,\theta)$ | 🌳nail (product-market fit)<br>before graduate with energy and time but no money | ProtoCompany<br>conduct proof-of-concept experiments with ~$1 million and 1 year timeline. Both focus on validation of core hypotheses and establishing fundamental viability.<br><br>structure learning" - determining if the foundational structure of the idea has merit.<br> | | phase2 (quality measure) | learn curvature covariance matrix M | exact inference $P_\mathcal{L}(y\|s,\theta)$ | ⛰️scale (growth given product-market fit)<br>adults with energy and money but no time | NewCo<br>they build business strategy, product plans, and assemble a larger team (20-30 people). Both focus on building upon validated foundations and preparing for significant scaling. In probabilistic terms, this is like "exact inference" - building precision on top of the validated structure. | | phase3 (quality measure) | | efficient inference | 🌊sail<br>elderly with time and money but no energy | Venture<br><br>recruit external CEOs, operate as a fully spun-out entity, and attract significant external capital. Both represent mature stages where the venture can operate more independently. In probabilistic terms, this is like "efficient inference" - optimizing processes that are now well-established. | | bottleneck | supply (assuming target dbn) | demand (reversible dgp-inf.alg) | demand, supply | killer experiments | | monitor(sense, eval) local optimal | n_eff, energy bfmi | | CTO vs CMO perceives higher demand vs supply uncertainty (market vs product) | using numbers instead of names for ProtoCompanies to make discontinuation easier. | | prevent(align, act) local optimal | parameterization depending on data amount (non-centered for large data) | | acculturate (frequent synthesize) | | | diagram | ![[Pasted image 20231222072332.png \|400]] | ![[Pasted image 20231222070323.png \|500]] | ![[Pasted image 20240122063934.png]]<br>![[Pasted image 20231222071150.png \|300]]![[Pasted image 20231226155653.png \|300]] | ![[Pasted image 20250413052819.png]] | | agnet center (from ai2human) | ![[Pasted image 20231222063048.png\|50]] | | ![[Pasted image 20231222063424.png\|50]] | - Structure learning = Exploration phase (determining if the structure has merit)<br>- Exact inference = ProtoCompany/NewCo phases (validating and refining)<br>- Efficient inference = Venture phase (optimizing a proven model) |