docs: arch: Inventory #1207
docs: arch: Inventory #1207
Conversation
Signed-off-by: John Andersen <[email protected]>
| knowing resources and processes. For example, how do you make coffee? You know a | ||
| coffee machine gives you brewed coffee. (Need machine, grinder, store sells | ||
| coffee, other store sells machine, delivery, etc. or coffee shop). | ||
|
|
There was a problem hiding this comment.
Trying to come up with a standard format for:
- Pieces of information, assets, resources, entities, processes (operations, operation implementations).
- Connections between those pieces
- Considerations which need to be taken into account when
a. The piece is used in a flow
b. The piece is used in relation to other specific pieces
c. Flucuating circumstances applied at flow execution
There was a problem hiding this comment.
Phased Approach to Universal Blueprint Creation and Execution
- Phase 1: Inventory of all operations, definitions, types of connections, connection information between operations and definitions, considerations for use of each operation, definition, connection
- Phase 2: Creation of all possible dataflows which take given inputs and produce desired outputs
- Phase 3: Prune set of possible dataflows
- Based on static considerations, considerations specific to when two specific operations are connected, considerations given during phase 3
- Phase 4: Execution
- OperationImplementation selection based on considerations at instantiation time
There was a problem hiding this comment.
What do we do in our heads when we're learning? We're trying to get information into a form in our heads that when presented with a problem allows us to connect the dots between the things we know so as to use the most relevant information to tackle the problem at hand in the most effective way possible.
information as it currently exists and is expressed does not capture connections to other pieces of information in a meaningful way so as to enable the effective use of relevant information when solving for a task at had.
Using note taking as an example. How would we most effectively take notes to organize the information we're consuming so as to optimize the recollection process later. Optimize for skimming. You see a chemistry problem, it involves certain units, elements or compounds in different states. What equations relate to those units, filter by what relate to those states, or get you from one state to another.
If we understand how concepts are related we can tailor the teaching of those concepts, the forming of those conceptual links, to the learning style of the individual needed to learn that link.
There was a problem hiding this comment.
POC to prod model
- Prototype
- Refactor into operations
There was a problem hiding this comment.
There was a problem hiding this comment.
2022-10-08 @pdxjohnny Engineering Logs
- Upstream
Alice is you. What do you have access too?
- webrtc media stream of desktop
- extension in browser
- search
- vetting of information (gatekeeper/prioritizer)
- codegen synthesis
- offline RL
- copy on write dataflow / system contexts for strategic plan evaluation for RL training on those predicted outputs
- start with max_ctxs=1
You ask codegen in generic terms for the prompt then you use open architecture plus codegen trained on open architecture to build deployments: system contexts, sometimes with overlays applied.\
We don't need codegen, to progress on this thought, it's just the
Everything is an operation. See thread, what are all the parameter sets its been called with before. We add feedback by enabling dynamic dataflow.auto_flow / by_origin called on opimpn run of gather inputs and operations.
This would be sweet in something as fast as rust. Could allow for rethinking with everything as operations and dataflow as class off the bat
- https://medium.com/@hugojm/from-text-to-a-knowledge-graph-hands-on-dd68e9d42939
- https://gist.github.com/pdxjohnny/1cd906b3667d8e9c956dd624f295aa2f
- https://fathy.fr/carbonyl
- This renders chrome to a terminal, we'll want to play with it eventually
- https://github.com/fathyb/carbonyl
- https://github.com/browserless/chrome#playwright
- TODO
- OS DecentrAlice: Fedora and Wolfi on different partitions. Boot to fedora, sshd via systemd-nspawn into wofli partition.
- Updates Similar
There was a problem hiding this comment.
- https://csarven.ca/dokieli-rww#architecture-and-technologies
- https://github.com/linkeddata/dokieli
-
There is no root, authority, or centralisation here. Control yourself!
-
There was a problem hiding this comment.
- https://github.com/cayleygraph/cayley
- https://github.com/inventaire/inventaire
- https://codeberg.org/forgejo/discussions/issues/16 - code.forgejo.org/actions as a catalog of Free Software actions
There was a problem hiding this comment.
Hmmmm, if we use thc.org/segfault and throw compute on devcloud under it that could be cool too
There was a problem hiding this comment.
think associative thinking is similar to:
#1207 (comment)
Thank you Steve!: https://deepai.org/machine-learning-glossary-and-terms/association-learning
|
Didn't mean to close this |
|
AI/LLM Agent Tool Catalog: https://github.com/scitt-community/scitt-api-emulator/blob/f52fb3c4316c3edc3abe456d1f8f8a891c59e65a/scitt_emulator/policy_engine.py#L8-L20 graph TD
subgraph Tool_Catalog[Tool Catalog]
subgraph Third_Party[3rd Party Catalog - Open Source / External OpenAPI Endpoints]
run_kubernetes_get_pod[kubectl get pod]
run_kubernetes_delete_deployment[kubectl delete deploy $deployment_name]
end
subgraph Second_Party[2nd Party Catalog - Org Local OpenAPI Endpoints]
query_org_database[Query Org Database]
end
query_org_database --> tool_catalog_list_tools
run_kubernetes_get_pod --> tool_catalog_list_tools
run_kubernetes_delete_deployment --> tool_catalog_list_tools
tool_catalog_list_tools[#47;tools#47;list]
end
subgraph llm_provider_Endpoint[LLM Endpoint - https://api.openai.com/v1/]
llm_provider_completions_endpoint[#47;chat#47;completions]
llm_provider_completions_endpoint --> query_org_database
llm_provider_completions_endpoint --> run_kubernetes_get_pod
llm_provider_completions_endpoint --> run_kubernetes_delete_deployment
end
subgraph Transparency_Service[Transparency Service]
Transparency_Service_Statement_Submission_Endpoint[POST #47;entries]
Transparency_Service_Policy_Engine[Decide admicability per Registration Policy]
Transparency_Service_Receipt_Endpoint[GET #47;receipts#47;urn...qnGmr1o]
Transparency_Service_Statement_Submission_Endpoint --> Transparency_Service_Policy_Engine
Transparency_Service_Policy_Engine --> Transparency_Service_Receipt_Endpoint
end
subgraph LLM_Proxy[LLM Proxy]
llm_proxy_completions_endpoint[#47;chat#47;completions]
intercept_tool_definitions[Intercept tool definitions to LLM]
add_tool_definitions[Add tools from tool catalog to tool definitions]
make_modified_request_to_llm_provider[Make modified request to llm_provider]
validate_llm_reponse_tool_calls[Validate LLM reponse tool calls]
llm_proxy_completions_endpoint --> intercept_tool_definitions
intercept_tool_definitions --> add_tool_definitions
tool_catalog_list_tools --> add_tool_definitions
add_tool_definitions --> make_modified_request_to_llm_provider
make_modified_request_to_llm_provider --> llm_provider_completions_endpoint
llm_provider_completions_endpoint --> validate_llm_reponse_tool_calls
validate_llm_reponse_tool_calls --> Transparency_Service_Statement_Submission_Endpoint
Transparency_Service_Receipt_Endpoint --> validate_llm_reponse_tool_calls
validate_llm_reponse_tool_calls --> llm_proxy_completions_endpoint
end
subgraph AI_Agent[AI Agent]
langchain_agent[langchain.ChatOpenAI] --> llm_proxy_completions_endpoint
end
llm_proxy_completions_endpoint -->|Return proxied response| langchain_agent
|
The knowledge graph