
Couch Detective is a text-based deduction game hosted by AI. Ask questions to uncover bizarre yet logical truths. You can also feed the AI your ideas to generate new puzzles and share them with friends.
Modeled estimate, not a reported figure. How we estimate →
Media

About
Ask, deduce, unravel—the truth is just one question away.
Have you heard of "lateral thinking puzzles"? Also known as "yes/no riddles"—
First, you read a vague and bizarre scenario. Then, players can ask any number of questions, and the host only answers "yes," "no," or "irrelevant," until the players piece together the full story. Couch Detective is just like that, except this time, the host is AI, and you're sitting on the couch, ready to unravel the mystery.
Read a super weird prompt (sometimes even a bit funny)
Ask any yes/no questions, try to deduce
AI will answer "yes," "no," or "irrelevant" based on your questions, and you piece together the story step by step
By the way, we have a creation mode! In creation mode, you just need to input a few keywords, a scenario, or even an ending, and AI will automatically generate a complete lateral thinking puzzle for you—the prompt, the answer, and the logic all set.
You can play it yourself or share it with friends to challenge them. Instantly become a puzzle designer!
Genres
Stats
All-time low is the lowest price we've recorded for this game since we started tracking it.
Languages: English, French, German, Spanish - Spain, Russian, Japanese, Simplified Chinese, Traditional Chinese, Portuguese - Brazil, Korean
Reviews
Couch Detective is a collection of situation puzzles. The player types in questions, and an AI host answers yes/no/irrelevant. CD is not the first game to do it, but still, this idea is possibly one of the most interesting uses of AI in games we've seen thus far. To complete a puzzle, the player needs to switch to answer mode and provide a short summary of the solution. The AI can give more elaborate answers than yes/no if there is a possible ambiguity, ask for a clarification, or explain why an answer provided by the player is incomplete. In fact, currently asking an open-ended question or providing an incomplete answer can be used as a way to obtain hints. Lastly, each puzzle has predefined built-in hints. One issue is that all those ways to obtain feedback and hints are not well balanced yet, sometimes revealing too much information or spoiling the puzzle outright. There's a log of previous questions and answers, and there's a separate section for collected information. The main game currently contains more than 100 puzzles. In addition, there are user-made puzzles and guess-the-item puzzles (I believe guess-the-item puzzles are generated automatically). However, the puzzles after Classic Soup I-IV are not up to par. Instead of being built around one interesting catch or twist, they require the player to reconstruct some contrived details, which isn't rewarding at all. There's also the usual issue of the game being online only, hanging indefinitely if it doesn't have Internet access, and not bothering to inform the player about what user data is being collected for analysis.
While this is a cool idea, the game is buggy. It is supposed to answer no/yes/irrelevant truthfully. There was one puzzle where is kept saying "no" to my questions but when I checked the notes where it keeps all the relevant answers, suddenly all the answers said "yes". But when I solved the puzzle, the solution showed that the notes were wrong. Example: Me: "was he murdered?" AI bot: "no" Hint notes: "he was murdered" Solution: "he was not murdered" Its just not consistent and makes it hard to solve the puzzles.
이 롯같은 새끼 자기가 파파야가 아니라고 해놓고 정답은 파파야를 넣어놔? 진짜 무친놈 나만 겪을수 없다
바다거북 스프를 좋아하지만, 같이 할 사람 찾기가 정말 어렵다. 이런 게임이 만들어졌다는 건 참 좋다. 근데 힌트 부분이 번역이 안 됩니다.
진짜 지X하지마
AI會亂回,一直回有時候有時候 抹布有時候是一款廚具嗎?
스트리머가 바다거북 수프 컨텐츠를 하는건 재미있는데 내가 혼자 이 게임을 하는건 재미가 없었네. 그걸 몰랐네.
친구가 없어도 바다거북스프게임을 할 수 있어서 좋아요 ㅠㅠ
[h1][b]探偵モノ、推理モノ、というより水平思考クイズなので 文脈を追いながら推理するのではなく 閃きと知識が問われるゲーム[/b][/h1] ストーリー ★★☆☆☆(サンプルとして配置されているものを対象) 構成 ★★☆☆☆ UI ★☆☆☆☆ BGM ★★☆☆☆ 推理性 ★★☆☆☆ 自由度 ★★★☆☆ ■気になる点 まず、はっきりとした原因はわからないが、 文字入力後にEnterキーで決定(プロンプト送信)で 20回に1回くらいの割合でフリーズする。 (私は初回のチュートリアルのウミガメのスープの問答でいきなりフリーズした) つぎに、プロモーションにあるような、 安楽椅子探偵きどりの動物がグラフィカルに口を開いたり、 音声でのやり取りはないし、 クイズそのものはほぼ文章のみでの推察となり、 映像(イメージ画像があるものはあるにはあるが)は殆どヒントとは呼べないものである。 さらにいうと、正解したあとの達成感はあまりない (ポイント形式とかランキングとか、ビハインドスコアが実はあるのか知らないが) そしてクイズに正解したあとの真相が表示されるウィンドウを 初見ではどうやったら閉じれるのかというガイダンスやボタンプロンプトがない。 (Escキーを押せば抜けられる) つづいて、日本語表記では、や。の句読点が、 いわゆる繁体字デフォなので真ん中に配置されて非常に読みにくい。 最後に中身。 5行くらいの文脈をよみとって真相を推理し、解答するゲームとしては こちらが質問したり、サンプルとして用意されているヒントを開いても 「そんなの正解にすんなりたどり着けるわけないだろ」が多すぎて向いていない。 ■良い点 生成AIを仕込んでいるゲームはまだまだ数少ないので足がかりとしては期待大。 あと中身。 前述のとおり文脈読み取り系にはまだまだAIの作りこみとしても向いているとはいえないが (たとえばリンゴとか果物をあてるなどの)単純なモノをあてるクイズとしては 論理性が高いのでとっつきやすい。 (が、それはそれで、だいぶ昔のゲームですでに実現されているので わざわざお金を払ったAIを噛ましてモノあてクイズ問答ごっこをするアドバンテージはない気がする) 最大の良い点は、オンラインで プレイヤーがそのクイズをゲーム内プラットフォームを通じて投稿し、 他人に評価してもらえる点だろう。 というわけで、バグは改修されることを期待しつつだが 中身は、クイズではなく文脈を読み取った推理を楽しめるなら高評価を出せた。 そうではない時点で、個人的には及第点とはいいがたい。 現代に大きな革新をもたらしている生成AIの可能性は、 まだまだこんなものではない。もっともっと拡張できるはず。
讓AI當海龜湯的題主還滿聰明的,不過使用作答的話AI會超級容易把答案說出來。有些題目感覺因果太過薄弱,有點可惜。遊戲很便宜,作為小品還是不錯。
Related
Related
Trends
Concurrent players on Steam, sampled daily since tracking started.
Related
| The Scroll of Taiwu : Beyond The Dome | $62.3M | 69.7K | 66% | $29.99 |
| Supermarket Simulator | $47.7M | 83.3K | 92% | $19.99 |
| Wallpaper Engine | $42.3M | 1M | 98% | $4.99 |
| Beat Saber | $38.1M | 76.2K | 96% | $29.99 |
| Universe Sandbox | $31.8M | 31.8K | 96% | $29.99 |
| Dyson Sphere Program | $30.9M | 91.2K | 97% | $15.99 |
Related
| FIND KITTENS 3: Steampunk | $1.7K | 141 | 92% | $0.73 |
| Space Hotel | $1.8K | 58 | 38% | $1.99 |
| BuildMoreCubes | $1.7K | 143 | 72% | $4.99 |
| DOSMan: Space Aliens in Space! | $1.7K | 82 | 84% | $1.04 |
| Spaceguy | $1.7K | 125 | 51% | $0.69 |
| Run Dude | $1.7K | 52 | 85% | $1.99 |