A | B | C | D | |
---|---|---|---|---|
1 | Theme | ServiceSpace: xGPT | ChatGPT | Examples |
2 | Data | Reprioritize data proximity, weight by adding contextual embeddings. | You're clubbed in with the entire web, with an opaque logic. | "What is metta meditation?" Ask SharonBot vs. ChatGPT. |
3 | Personalization | Create personal identity, via a self-defined and editable "credo" | You're talking to a default, which you can temporarily edit with a prompt. | "What is a souffle?" |
4 | Context | Create an optional personalization layer; user can opt-out as well. | It's unclear what ChatGPT does. | |
5 | Sources | For each response, list sources wherever applicable. | No sources. Opaque. | Any post. |
6 | Aggregate Identities | Build meta-personalities, by combining multiple unique identities. | N/A | ServiceSpace + SharonBot + GGBot |
7 | Depersonalization | Create a universal-values (pro-social) layer on top of the contextual layer. (ie. general content + individual context + universal values) | Market logic biases towards indiviaul over collective, answers over questions, consumption over contribution, polarity over connection, sensationalism over contentment. | Manu's chat around dog, Ronak. |
8 | ||||
9 | Alignment | Eventually, we hope to fine-tune a data model that aims for alignment -- with a far deeper expression of what it means to be human. | OpenAI will dedicate 20% of its resources to "human alignment" problem; but alignment is based on very cheap definition of what it means to be human. | |
10 | Field of Motivations | Build a volunteer (intrinsically motivated) community that helps crowd-source collective wisdom around human alignment. | Data point -- OpenAI's first round of RLHF was done using very, very cheap labor. | See limitations of current RLHF approaches. |
11 | Pro-Social Platform | Design as a platform, so others can build compassion-centered apps on it. | They are doing this, but working with a different baseline of data; and in context of a market/consumption logic. | CompassionBot, for KarunaNews. |
12 | Integration | Add data-corpus + intelligence into large language models like ChatGPT/Bard. | N/A | Conversation with Bard folks. |
13 | LLM | Create a separate "wisdom" LLM. | N/A | Community of communities. |
14 | ||||
15 | ||||
16 | ||||
17 | ||||
18 | ||||
19 | ||||
20 | ||||
21 | ||||
22 | ||||
23 | ||||
24 | ||||
25 |