AI updates.

AI and product progress, in a nutshell

Direction: build mission-centered tools with AI as assistance and people accountable for evidence, public claims and release decisions. Our current AI use supports internal research and engineering; the public site does not offer autonomous mission matching, AI certification or a universal trust score.

Where development stands: Open Directory is available for source-aware public discovery. GameChangers is being developed in stages, starting with grid interconnection and transmission discovery in the Intermountain West. CuriousPi is separate from GameChangers and remains a local scripted prototype. Local-first experiments with open-weight models are a research direction; proposed Qwen and Gemma connections are not deployed, and no model-backed CuriousPi service is offered to users.

Next engineering gates, planned: define task-specific evaluations; review data boundaries, privacy and security; establish human approval and correction paths; then disclose a release stage before any limited user test. These are planned gates, not completed controls or a launch commitment.

Read our Responsible AI practice ↗ · Read the global mission roundup ↗

CuriousPi remains a local scripted prototype

Current status review: CuriousPi remains separate from GameChangers and Open Directory. It is a local scripted prototype; proposed Qwen and Gemma model connections are not deployed. No model-backed CuriousPi service or autonomous agent is offered to users, and CuriousPi is not connected to GameChangers data.

This update describes current product state. Any future model connection, user-facing capability, data flow or change in decision authority will require a separate review and public disclosure before release. This status statement is not a security certification or legal-compliance guarantee.

See current product status · Follow dated build updates

CuriousPi: proposed Qwen + Gemma advisory architecture

CuriousPi is an independent consulting-intelligence project, separate from GameChangers Marketplace OS, Open Directory and their data. This entry discloses an engineering design under evaluation, not an operating AI service or a GameChangers capability.

Proposed architecture — not deployed

1. Consulting workspace and intake: a light-blue and grey interface captures the business problem, objectives and voluntarily supplied context.
2. Private application gateway: planned authentication, access checks, request limits, model routing and human-approval gates before tools or data are used.
3. Local model inference: proposed Windows/NVIDIA development environment with Ollama; Qwen3.5 as the initial analyst candidate and Gemma 4 as an optional second-model reviewer for selected higher-stakes drafts. Model sizes and routing remain subject to tests.
4. Permission-scoped knowledge retrieval: planned retrieval from approved consulting references and authorized client materials, retaining source provenance and separation between clients.
5. Evidence and human review: distinguish cited facts from assumptions, surface missing evidence and uncertainty, and require a person to approve consequential advice or external actions.

Data, oversight and current limitations

Potential inputs include user-supplied business context and client documents shared with permission. Planned safeguards include data minimization, explicit authorization, client separation, secure storage, retention/deletion rules and no use of client documents for model fine-tuning by default. These are requirements for future implementation, not claims that production controls have passed security review. No CuriousPi client information is approved for transfer into GameChangers or Open Directory.

As of 7 October 2026, the workspace is a scripted local prototype; Qwen and Gemma are not yet connected, fine-tuned or serving production users. The two-model review pattern is unvalidated, and model agreement does not prove accuracy. AI cannot independently make binding legal, financial, hiring, compliance or other consequential decisions.

Change disclosure: this notice will need revision before a user-facing release or material change in model provider, hosting, data processing, retention, evaluation or decision authority. This is an architecture disclosure, not a claim of certification, commercial availability or endorsement by a model provider. Read our AI transparency principles.

Responsible AI governance package established

Assemble Teams formalized an internal Responsible AI governance package covering AI-assisted research and engineering, user-facing AI release states, human authority, evaluations, incident review, benchmark integrity and material public change disclosure.

AI cannot autonomously publish Benchmark ranks or certifications

Mission Benchmark research separates AI-assisted evidence work from publication authority. AI may help discover, organize and synthesize evidence, but listing, evidence qualification, comparability, future rank eligibility and any future verification or certification state remain separate human-governed gates.

AI-related capabilities use explicit release-state labels

Public AI capability claims are governed through explicit states so planned work is not presented as currently operating.

AI-generated research remains a lead until source-checked

AI may accelerate discovery and synthesis, but missing evidence is not converted into certainty. Material Benchmark records and public research claims require source review, and AI assistance does not itself mean verified, independently assessed, certified, endorsed or legally compliant.

Consequential decisions retain human authority

AI-only final decisions are not authorized for high-impact areas such as benchmark rank publication, certification, ambiguous user suspension, legal/compliance determinations, employment or selection decisions, financial awards or penalties, and high-impact mission participation unless a future narrow policy explicitly authorizes and validates such use.