[
    {
        "role": "user",
        "content": "I%20was%20listening%20to%20this%20week%20in%20Inteligent%20Machines%20Podcast%2C%20I%20guess%20it%20was%20last%20weeks%20episode%20and%20Leo%20talked%20about%20a%20different%20type%20of%20a%20AI%20that%20just%20filters%20or%20gives%20a%20yes%20or%20no%2C%20what%20is%20that%20caleld%20I%20as%20thinking%20it%20was%20something%20like%20a%20Discion%20AI%20or%20something%3F"
    },
    {
        "role": "assistant",
        "content": "What you heard Leo talking about is almost certainly \u201cdecision models\u201d (also called \u201cdecision AI\u201d) \u2014 a newer type of AI that doesn\u2019t generate long text but instead answers a well\u2010defined question (like yes/no) with probabilities or labels. ([systemonemodels.org](https://systemonemodels.org/guides/what-is-a-decision-model/?utm_source=openai))\n\nHere are the more precise terms and how they differ:\n\n---\n\n## Key Terms and What They Mean\n\n**Binary classifier**  \n- A classic term in machine learning. It\u2019s a model that takes input data and decides between two classes (yes/no, positive/negative, true/false). ([en.wikipedia.org](https://en.wikipedia.org/wiki/Binary_classification?utm_source=openai))  \n- It typically outputs a probability or score, which you then threshold to pick one of the two answers.\n\n**Decision model / Decision AI**  \n- A newer name for models built specifically for structured decision outputs. Instead of generating text (like a typical LLM), they output typed, calibrated decisions (yes/no, a choice from multiple categories, or a score) with confidence values. ([systemonemodels.org](https://systemonemodels.org/guides/what-is-a-decision-model/?utm_source=openai))  \n- These models are often cheaper, faster, and more reliable for tasks that need a clear decision rather than prose. ([roai.us](https://roai.us/research/the-decision-layer?utm_source=openai))  \n\n**Yes/No (or Boolean) decision**  \n- This is the simplest type of decision model output: just \u201cyes\u201d or \u201cno.\u201d  \n- In some frameworks this is called a \u201cpredicate,\u201d \u201cyes_no\u201d type, or \u201cnoul.\u201d ([docs.classifyre.com](https://docs.classifyre.com/detectors/custom-detectors/decision/?utm_source=openai))  \n\n---\n\n## Recent Developments\n\n- **TypeSafe AI** released \u201cJev\u201d, a model they describe as a \u201cSystem One\u201d model: it takes in input content and a typed question (e.g. \u201cIs this text threatening?\u201d) and outputs a decision (yes/no) with a probability. ([jev.tibor.io](https://jev.tibor.io/what-is-jev?utm_source=openai))  \n- OpenAI has a \u201cDecisions API\u201d that similarly lets you ask questions and get structured answers instead of text generation. ([docs.empiriolabs.ai](https://docs.empiriolabs.ai/decisions?utm_source=openai))  \n\n---\n\n## What Leo Likely Meant\n\nWhen Leo said something about \u201ca different type of AI that just filters or gives a yes or no,\u201d he was probably referring to these decision models \u2013 especially ones like Jev or OpenAI\u2019s Decisions API. They are designed not to have the model compose long answers, but to decide with yes/no or pick from options, often to be used in automated pipelines. \n\n---\n\nIf you want, I can pull up the exact episode or link where Leo talked about this to confirm what term he used. Would you like me to do that?"
    },
    {
        "role": "user",
        "content": "Can%20you%20give%20me%20some%20examples%20online%20where%20I%20could%20experiment%20or%20see%20how%20they%20process%20info%3F"
    },
    {
        "role": "assistant",
        "content": "Here are some live tools and demos where you can try out or see how decision / \u201cyes-no / choice / score\u201d AI models process information. Most involve TypeSafe\u2019s Jev model, which is one of the leading \u201cdecision-only\u201d AI models today:\n\n---\n\n### \ud83c\udfaf Live Tools & Demos\n\n1. **Jev Playground** (TypeSafe)\n   \u2022 A hands-on interface to test \u201cchoice\u201d, \u201cscore\u201d, or \u201cnoul\u201d (yes/no) decision types on example tasks. ([jev.works](https://jev.works/?utm_source=openai))  \n   \u2022 You can tweak state and questions, see the outputs, and inspect how the decision is shaped. ([jevtypesafe.org](https://jevtypesafe.org/jev-tutorial/?utm_source=openai))  \n\n2. **Jev Use Cases & Production Notes Gallery**  \n   \u2022 Collections of real uses \u2014 Jev playing \u201cSnake\u201d or \u201c2048\u201d live (every move is a decision), moderating content, routing support tickets, etc. ([jevtypesafe.org](https://jevtypesafe.org/use-cases/?utm_source=openai))  \n   \u2022 Watch how these are used in workflows to gate actions or classify things. ([jevtypesafe.org](https://jevtypesafe.org/use-cases/?utm_source=openai))  \n\n3. **Jev \u201cDemos: Typed Decisions vs LLMs\u201d**  \n   \u2022 A demo comparing outputs when using generative LLMs vs using Jev. You can see all shapes of output (choice, score, yes/no), seeing how quickly and deterministically Jev responds. ([jevdemos.vercel.app](https://jevdemos.vercel.app/?utm_source=openai))  \n\n4. **\u201cFlappy Bird\u201d Demo with Jev**  \n   \u2022 A game where your only decisions are whether the bird should \u201cflap\u201d or \u201cwait.\u201d Jev makes that choice. ([github.com](https://github.com/hosseintoussi/jev-flappy-bird?utm_source=openai))  \n\n5. **Smart Home Assistant Demo / other interactive tasks**  \n   \u2022 Tasks where you give Jev a request (e.g. adjusting lights, turning on devices) and it decides what action or label to use, often with fallback to an LLM when needed. ([ningg.top](https://ningg.top/jev-docs-CN/demos/?utm_source=openai))  \n\n---\n\n### \u2699\ufe0f Other Tools & Examples\n\n- The **\u201cAwesome Jev Use Cases\u201d** GitHub list: dozens of community-built demos/apps, with code, metrics, and explainer repos. ([github.com](https://github.com/walidboulanouar/awesome-jev-use-cases?utm_source=openai))  \n- Reddit and side-projects where people are using Jev to score ideas, build agent routers, etc. ([reddit.com](https://www.reddit.com/r/SideProject/comments/1wiw6tk/i_built_a_side_project_to_test_typesafes_jev/?utm_source=openai))  \n\n---\n\nIf you want, I can send you some of these with direct links so you can try them out quickly. Do you want live links or ones you can use without needing an API key?"
    }
]