Juan Bello’s Post

OpenAI is ‘going to build a search product’... Makes complete sense, I already use OpenAI 4o for 80% of my search. If I need an answer to something, a fact, a translation, grammar help etc. I use CGPT, if I just need a list of websites for options then I still use search, but its gettign less and less. If I am looking for an answer to something CGPT gets the answer quicker and more efficiently 100% of the time... Yes I'm paying $30 a month, but I get a whole lot more than search. I suspect price will be a barrier for most, me I'm already converted... btw My favourite GPT is currently 'Diagrams <Show Me>', check it out... https://lnkd.in/gNPVi8Zv

We tested OpenAI's GPT 3.5 & 4, Meta's Llama 3, Google's Gemini 1.5 Pro, Cohere's Command R+ and 130+ other LLMs on writing code. Deep dive blog post soon 🏇 In the meantime: a screenshot of the results (more points is better)

In this post, I explain how to add speech-to-text capabilities to the Jarvis AI assistant using OpenAI Whisper API. With this feature in place, you can know speak back and forth with Jarvis. A truly magical experience.

OpenAI’s GPT Store: A Promising Start but with Some Key Flaws! 🚒 🚨 I've explored the new GPT Store by OpenAI and here's the deal... -Promising Start 👍: A nice and simple way to find new GPTs and provide creators with additional exposure. -But Needs Polishing 👎: Needs user reviews/ratings, better analytics for creators, more GPT categories, search results pages, etc., to truly excel. It's a step forward but with room to grow. Read more about it at the link below! #gptstore #gpts #chatgpt https://lnkd.in/dCXXiZA7

We just submitted our first Custom GPT to the OpenAI store, after I've had this on my list for a couple of weeks. 'EUDR Expert' will help affected companies navigate the complexities of the EU deforestation regulation. I am curious to see if and how the user numbers develop and -if they develop well- explore how this tool can potentially double for lead gen. It would be great if OpenAI provided an easy method to embed them directly on your website, though. But maybe that's coming up? The link: https://lnkd.in/eMepcZsD

Did OpenAI publish an unfinished product??🤯 I'm sure most of you guys followed the OpenAI DevDay and the announcement of the GPT Builder Beta. I wanted to see if it really is up to what OpenAI promises and well.. it definitely deserves the name "Beta". You can read about my experience and why I still think that OpenAI will pull through and deliver a great experience in the future in my most recent blog article.

Google Gemini Flash is turning out to be possibly the most interesting recent LLM announcement. It’s one of those small models that doesn’t compete with GPT-4, so it’s not had so much press attention. I don’t think that’s fair and here’s why... Gemini Flash is cheap (cheaper than GPT-3.5-turbo) and has a crazy-long input size, so you can get it to read entire books. That alone would make it interesting, but I noticed that Google doesn’t charge for fine-tuning and there’s no premium to use your fine-tuned model. Not charging for fine-tuning is wild - OpenAI charge a significant premium, between 4x and 6x depending on your input/output mix. The bottom line is that your own fine-tuned version of Gemini Flash is cheaper than the base off-the-shelf GPT-3.5-turbo and *massively* cheaper than a fine-tuned version. I could never see the point of fine-tuning GPT-3.5-turbo, as the big cost penalty means there’s very few uses where it makes financial sense. With Gemini Flash, Google just got my attention - fine-tuning suddenly got interesting!

With the rapid growth of artificial intelligence technology, converting spoken language into text has become an incredibly useful skill. OpenAI’s Whisper API is a powerful tool for doing just this—it can accurately turn your spoken words into written text. https://lnkd.in/gUTQB3Ah

Will this be a threat to OPENAI..? Claude 2.1 is a direct competitor It’s offering unique features distinguishing from it’s rivals. It’s focus is on handling complex multi-step instructions. (It’s clever and can handle hard tasks). Claude has up-to-date data Although compared to GPT, it only supports English and a few other languages. Meanwhile GPT offers output in 90 different languages.

I've been aware since ive been writing code against public LLMS such as OpenAI's GPT API that something strange is happening. You can get it write a game of Snake, but not a text editor. A game of breakout, but not a simple spreadsheet. Why ? I guess they don't want to to be disruptive ? And it changes over time. It wasn't much of annoyance until today : but I was asked to write a small training app : and ive already done one before : get some base data in a PDF on the subject matter, encode it into a vector database, use a vector router with a LMM to ask questions about the subjects matter, and rate the users answers. LLM prompt is of the form : "Can you give me 5 questions about PANDAS. Make them Multiple Choice. Pause for each question to let me answer. Tell me if I get it right. At the end tell me how many i have got right in total. " No big deal. So I pulled up the app again today, and there is no need for privacy or confidentiality, so using GPT4 API should be be fine. It was last week. But today, the API just refuses : Chat GPT cannot generate a quiz, you should use specific software for this. Is it really safe to write production software against an API which can change its behaviour over time to just refuse to answer for no reason ? TBF : it didn't take long to jailbreak the prompt into working, but how long will that last before the behaviour changes again ? Obviously re-wrote to use Llama2, but annoying to have to use expensive local inference to do a task that a public GPT3.5 turbo API call have answered almost for free. Developers beware of the last minute refusal ! :)