{
  "title": "Putting AI to Work, Cloud1 and Vuo AI | Harri Puupponen, Pasi Jokinen | Neuvottelija 189",
  "titleOriginal": "Tekoäly käyttöön Cloud1 Vuo AI | Harri Puupponen Pasi Jokinen | Neuvottelija 189",
  "episodeNumber": "189",
  "guest": "Harri Puupponen, Pasi Jokinen",
  "publishedAt": "2023-04-27",
  "duration": "1:03:02",
  "durationIso": "PT1H3M2S",
  "youtube": "https://www.youtube.com/watch?v=8Y_MWfgxBFk",
  "fiCanonical": "https://www.neuvottelija.fi/fi/episodes/278-tekoaly-kayttoon-cloud1-vuo-ai-harri-puupponen-pasi-jokinen",
  "originalLanguage": "fi",
  "format": "full",
  "topics": [
    "ai_enterprise_tech",
    "saas_software",
    "negotiation"
  ],
  "description": "Harri Puupponen, founder of Cloud1, and Pasi Jokinen of Vuo AI discuss how Finnish companies and individuals put language models to work. Recorded in April 2023, weeks after GPT-4 shipped and AutoGPT broke out, and the interview questions were written by GPT-4 itself. Jokinen defines AutoGPT as an executive function, the prefrontal-lobe counterpart to GPT, and explains why its budget constraint is the interesting part. The most usable passage for a negotiator is perspective-shifting: ask the model to describe your case from the other party's point of view, because working memory cannot do the translation on the fly. One of the guests recounts telling a large Finnish bank whose lawyers had banned hallucinating applications that the bank employs 13,000 hallucinating instances, and that this one passes the California bar exam. A demonstration with the lion, sheep and cabbage river puzzle shows where GPT-4 fails systematically: change the rules and it still solves the standard version, admits the error, lists the constraints correctly and repeats the mistake. On M&A, Bengt Holmström's point that full disclosure creates informational equality is set against the Trojan-horse risk, and Jokinen proposes machine-run due diligence returning only pass flags before a fully transparent phase. The episode closes on the hardware ceiling: GPT-4 rate limits, Azure availability only in the United States, and NVIDIA's GPU forecasts falling short of demand.",
  "chapters": [
    {
      "t": "00:00",
      "title": "WhatsApp preparation for the episode, and ChatGPT-4.0's analysis of it",
      "seconds": 0
    },
    {
      "t": "00:26",
      "title": "ChatGPT 4.0 question 1 for Pasi Jokinen",
      "seconds": 26
    },
    {
      "t": "01:35",
      "title": "Pasi's Oulu-era friendship with the new MP Tere Sammallahti",
      "seconds": 95
    },
    {
      "t": "02:38",
      "title": "Voland's investment in Harri Puupponen's Cloud1. ChatGPT 4.0 question 1 for Harri",
      "seconds": 158
    },
    {
      "t": "04:14",
      "title": "Are Microsoft's cloud partners riding the AI boom thanks to the OpenAI investment?",
      "seconds": 254
    },
    {
      "t": "05:50",
      "title": "ChatGPT 4.0 question 2 for Pasi Jokinen",
      "seconds": 350
    },
    {
      "t": "09:04",
      "title": "ChatGPT 4.0 question 2 for Harri Puupponen",
      "seconds": 544
    },
    {
      "t": "11:57",
      "title": "Teaching AI the Neuvottelija material and the book Uusi neuvotteluvalta (WSOY)",
      "seconds": 717
    },
    {
      "t": "14:49",
      "title": "Closed AI, or networked cooperation?",
      "seconds": 889
    },
    {
      "t": "16:17",
      "title": "The trust account concept in AI cooperation",
      "seconds": 977
    },
    {
      "t": "18:03",
      "title": "The AI of AIs: AutoGPT",
      "seconds": 1083
    },
    {
      "t": "20:15",
      "title": "Did Microsoft win the AI game by accident with the OpenAI investment? Even Bing works",
      "seconds": 1215
    },
    {
      "t": "22:18",
      "title": "Will API capacity run out when everyone uses the ChatGPT lane?",
      "seconds": 1338
    },
    {
      "t": "24:43",
      "title": "Is it right to speak of AI hallucinating?",
      "seconds": 1483
    },
    {
      "t": "26:42",
      "title": "The limits of AI and the most-probable-word problem",
      "seconds": 1602
    },
    {
      "t": "29:42",
      "title": "Is it even worth building your own language models? The Neuvottelija-GPT case",
      "seconds": 1782
    },
    {
      "t": "34:27",
      "title": "From text to images and sound, expressions and gestures",
      "seconds": 2067
    },
    {
      "t": "39:15",
      "title": "Intelligent agents and trust networks",
      "seconds": 2355
    },
    {
      "t": "40:30",
      "title": "M&A negotiations — Bengt Holmström on complete openness through due diligence, and clean rooms run with AI",
      "seconds": 2430
    },
    {
      "t": "45:10",
      "title": "Trust, once again",
      "seconds": 2710
    },
    {
      "t": "47:30",
      "title": "Blackstone's global chief executive and AI playbooks",
      "seconds": 2850
    },
    {
      "t": "52:01",
      "title": "AI in Microsoft Office — the return of the paperclip",
      "seconds": 3121
    },
    {
      "t": "53:35",
      "title": "Power BI and data analytics in the grip of AI",
      "seconds": 3215
    },
    {
      "t": "57:12",
      "title": "AI as the perfect complement to lost attention",
      "seconds": 3432
    },
    {
      "t": "57:30",
      "title": "Where the hardware comes from — NVIDIA and edge computing",
      "seconds": 3450
    },
    {
      "t": "1:00:24",
      "title": "Cloud1's services",
      "seconds": 3624
    },
    {
      "t": "1:01:29",
      "title": "Vuo AI's services",
      "seconds": 3689
    },
    {
      "t": "1:02:42",
      "title": "Outro",
      "seconds": 3762
    }
  ],
  "sourceTranscriptType": "macwhisper_srt",
  "subtitleMethod": "none",
  "provenance": "Owner page assembled from YouTube metadata, the neuvottelija.fi episode record and a MacWhisper transcription of the Finnish audio. The chapter marks are the publisher's own list as released with the video, each verified to fall within ten seconds of a real transcript cue. No English caption track was produced, so no transcript is published here. The transcription contains one gap of roughly 27.7 seconds at 00:55:40, between the Tower of Hanoi request and the tamagotchi request in the WebGL demonstrations; the summary is written only from what the transcript contains and the gap has not been filled by inference. The MacWhisper transcript carries no speaker labels, so speakers are named only where the transcript gives direct evidence; two passages whose speaker could not be established are attributed to \"one of the guests\".",
  "slug": "ep189-tekoaly-kayttoon-cloud1-vuo-ai-puupponen-jokinen",
  "guestsCanonical": [
    "Harri Puupponen",
    "Pasi Jokinen"
  ],
  "guestPages": [
    "https://www.neuvottelija.com/podcast/guests/harri-puupponen/",
    "https://www.neuvottelija.com/podcast/guests/pasi-jokinen/"
  ],
  "page": "https://www.neuvottelija.com/podcast/episodes/ep189-tekoaly-kayttoon-cloud1-vuo-ai-puupponen-jokinen/",
  "markdown": "https://www.neuvottelija.com/podcast/episodes/ep189-tekoaly-kayttoon-cloud1-vuo-ai-puupponen-jokinen/index.md",
  "captions": null,
  "transcript": null,
  "videoId": "8Y_MWfgxBFk",
  "dbId": null,
  "qaCoveragePct": 0.0,
  "qaSemanticVerdict": "transcript_timecoded"
}