{
  "video_id": "8t10hdfjS8E",
  "channel_slug": "dataindependent",
  "channel_handle": "dataindependent",
  "title": "I react to OpenAI DevDay",
  "duration_seconds": 495.0,
  "url": "https://www.youtube.com/watch?v=8t10hdfjS8E",
  "upload_date": "",
  "transcript": "open AI hosted their first developer\nconference in San Francisco on Monday in\nthis video I'm going to count down my\ntop 10 announcements they made and what\nit means for Builders and Founders over\nthe weekend I asked you if you were\ngoing to watch open ai's Dev day 43% of\nyou said you didn't know it was going on\nwell if you didn't know then you\nprobably know now it's hard to avoid all\nthe hype it was their chance to tell the\nworld about all the cool stuff they were\nbuilding best of all most of the\nannouncements they made were immediately\navailable to start building with hello\ngood\npeople\nit was an absolutely stunning event\ngreat design great Vibe and great\nexecution I haven't seen this type of\nhype in a while and it didn't feel\noverly corporate either just a bunch of\nBuilders looking to do hood rat stuff\nwith their friends I hosted a watch\nparty with Harry tmy and Niall oh\nHiggins over at the GP in San Francisco\na bunch of people got together to nerd\nout on AI announcements it's fun getting\nto chat with people away from the\nkeyboard but after the event I was\nexhausted from all the announcements I\nneeded to have a fresh sleep a cup of\ncoffee and a good think on how I\nactually feel about these announcements\nwell that time is over and I'm excited\nto share with you thoughts and\nReflections on my top 10 favorite\nannouncement from open ai's Dev day\nlet's get into it number 10 custom\nmodels you can now work with open aai\nresearchers to go deep on model building\nfor your specific use case it's fine but\nit'll run you about $2 to3 million to do\nso this is an interesting Services play\nfor open AI what's even more interesting\nto me is the opportunity for other\nagencies to provide fine tuning as a\nservice you won't go all the way down to\nthe model building like you will with\nopen AI but you can help customers\nunderstand fine-tuning best practices\nand even do it for them I think there's\na huge market for this as customers\nthink that fine tuning will help them\nwith their performance use cases number\nnine GPT 4 fine-tuning it's well known\nthat GPT 3.5 when it's fine-tuned\nexceeds gp4 on some tasks even if the\nperformance is the same as gp4 you get a\nmassive drop in cost and latency so if\nyou fine-tune GPT 3.5 and it's as good\nas GPT 4 what happens when you fine-tune\nGPT 4 well I think we're going to see\nsee some really cool things here number\neight text to speech API we already had\ngood text to speech with 11 labs and\nplay HT I'm not an oracle or anything\nbut I'll do my best here so this one\nisn't a big unlock but I like how you\ncan now consolidate the vendors that you\nwork with and get the product quality of\nopen AI how long do you think it'll be\nuntil they allow custom voices leave me\na comment and let me know what you think\nnumber seven chat GPT backed by GPT 4\nTurbo chat GPT is now powered by the new\nGPT 4 Turbo model I love it I use chat\nGPT for my ad hoc quick t tasks I'll\ntake a speed boost here all day long\nnumber six GPT 4 Turbo\n128,000 token context length this one's\na bit controversial GPT 4 now has the\nability to ingest 128,000 tokens worth\nof context this is practically an entire\nbook on one hand you have the crowd that\nsays oh my God now we can put a book in\nGPT 4 this changes everything but we\ncould already do that via breaking it up\ninto Parts you'll get lower latency with\nthe single API call but what about the\nperformance hit I'm currently pressure\ntesting the 120 ,000 context limit with\na needle in the haystack analysis I'll\ndrop a random fact in the middle of a\nlarge context window to see if GPT 4 can\npull it out sign up for the newsletter\nor follow me on Twitter if you want to\nsee those results number five GPT 3.5\nand4 price reduction so gp4 is not only\nbetter but now it's cheaper I like that\ncombo as the price reduces you start to\ndrop the barrier to entry on a few use\ncases for example a use case that I\nheard yesterday from nialo Higgins was\nwell what if you're in the data\nmonitoring game passing in a bunch of\nlogs to gp4 can be quite expensive but\nat half the costs now you can increase\nthe flow of data that you pass through\nit I think we're going to see some more\ninteresting examples here number four\nDolly 3 API see I love this one it's so\nhigh on my list because it's a net new\ncapability that millions of developers\nnow have but Greg what about all the\nother apis with stable diffusion and mid\nJourney sure but I don't want to have to\naccess Discord API to get what I need\nfrom it I think we're going to see\nmillions of photos flooding apps near\nyou let's jump into the top three\nbiggest announcements for Monday number\nthree assistant gpts in the API okay now\nwe're getting to the serious part of the\nlist openi isn't messing around anymore\nwhat is the assistance GPT API well now\nyou have little mini Bots that you can\nstart to code up you can give them extra\ncontext and you can start to\nprogrammatically give them rules that\nthey should follow it's kind of like a\nprepackaged chat bot that you can give\nto users but let's talk about what the\nassistance API actually does well number\none now you have the code interpreter\nAPI this means open AI will run a bit of\ncode on their end and give you the\nresult this is great for longer math and\nfile trans Transformations up until now\nit was only available via the UI but\nhaving available via the API means that\ndevs can offload a ton of hacks that\nthey were doing to try to get this done\nbefore number two you now have file\nretrieval so assistants can reference\nfiles that you've uploaded open AI will\ntake care of the chunking the embeddings\nand the retrieval for you this sounds\npretty good on paper right for NOCO\nBuilders I think this is going to be a\nhuge unlock you don't have to worry\nabout building retrieval anymore but for\ndevs I predict that they're going to\nwant more control over the retrieval\nprocess than opening AI will give them\nnot only that but it's super expensive\nto host files with open aai number three\nyou'll also get persistent threads this\nis nice and convenient because thread\nmanagement per user was a pretty\nannoying issue that developers had to\nworry about for a while number two chat\nGPT App Store now you would think this\nwould be number one but it was dethroned\njust because of my uncertainty around\nthe adoption yes the demos are cool and\nyes there's big promise but I haven't\nyet seen the evidence of user need to\ncall this one gamechanging quite yet if\nit does work then it's the beginning of\na massive product opportunity but\nthere's a couple facts to keep in mind\nbuilding marketplaces is extremely hard\nplug-in adoption didn't go very well but\nmonetary incentives to build apps can\ndrive crazy adoption and building but\nthere's still a few problems consumers\nwill need to pick which GPT they want\nand they'll also need to proactively\ntrigger these gpts to do work for them\non the flip side I like this because you\ndon't have to share your prompts anymore\nyou can just share your assistant and\nget a review on the assistant it's super\nannoying when you go on LinkedIn and you\nsee 100 50 new chat GPT prompts that you\nneed now instead of just giving me the\nprompts give me the value inside of an\nentire GPT bot I also love the creative\nprofiles it's pretty cool how you can\nstart to connect with the community via\nBuilder profiles another low-key\nannouncement that open aai made was\nthey're going to do Revenue sharing for\nthe gpts that you build this is pretty\ncool cuz now there's an incentive to\nshare the gpts that you build and even\ndo marketing for them the biggest\nquestion in everybody's mind now is well\nhow many startups are dead what about\nthe chat gbt rappers my quick answer is\nno these startups aren't dead yet but\nthey do control their own fate and it's\nnot looking that great if you want to\nsee an example about the future of these\nrappers you don't have to look any\nfurther than pdf. they're a huge example\nabout where this is going they're about\nas rapper as you can get but my\nhypothesis is that if you can build a\nlot of products around your rapper\nexperience and provide a better\nexperience than open AI for your\ncustomers then you're going to be okay\nnow finally on to my favorite\nannouncement from yesterday number one\nGPT 4 Vision API lowering latency and\ncost on other models are are cool but\nhow often do you have a new modality At\nYour Service after GPT 4V came out in\nthe UI I wrote a blog post on how people\nwere using it they were all doing it in\none-hot examples meaning you give GPT 4V\na single photo and it'll give you a\nresponse but now that the API has opened\nup we're seeing a whole new class of\ninteractions people are demoing realtime\nidentification via webcams this opens up\nsemantic search and understanding of\nvideo as well there's huge applications\nfor interpreting what's on your computer\nscreen what's on a security camera well\nwhat about meta's Rayband glasses or\nwhat about extracting unstructured data\nor maybe the moderation use case I put\nthis one up above the GPT store because\nwell my confidence is high on this one\nand I know that's a controversial thing\nto say and the future is hard to predict\nbut I'm willing to bet that this is\nwhere we're going to see some of the\nmost immediate value from so how do we\nget started well the answer is same as\nalways you need to go Tinker you need to\ngo have fun and you need to build an\nintuition for these tools go find\nproblems you're passionate about solving\nand go find customers willing to pay you\nfor that value let me know about your\nown reactions to Dev day in the comments\nwe'll see you\nlater",
  "transcript_chars": 9562,
  "ingested_at": "2026-05-15T04:41:10.671716+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 2286,
    "like_count": 104,
    "channel_id": "UCyR2Ct3pDOeZSRyZH5hPO-Q",
    "categories": [
      "Science & Technology"
    ],
    "tags": [
      "Open AI Dev Day",
      "Dev Day Review",
      "Top 10 Most Interesting AI Updates",
      "AI event recap",
      "Artificial intelligence insights",
      "Tech conference highlights",
      "AI innovations",
      "Developer conference summary",
      "OpenAI event analysis",
      "Key takeaways",
      "Cutting-edge AI tech",
      "Tech industry trends",
      "AI developments",
      "OpenAI news",
      "Event review",
      "OpenAI updates",
      "AI in 2023",
      "Future of AI",
      "Tech discoveries",
      "ChatGPT"
    ]
  }
}