Dario’s Picks

I. Silent ChatGPT upgrade: @ mentions for GPTs

Without any public announcements (so far), OpenAI has shipped a new feature to a select few ChatGPT plus users that allows tagging GPTs directly in the chat to activate them.

Why it matters: I haven’t got the feature myself yet but, by the look of it, this upgrade makes the experience of using GPTs a lot smoother. We’re so used to @-mentions from other apps, that I think it will be a much more intuitive way to work, rather than having to go to the GPT store each time you want to activate a GPT.

Reportedly this also make the combination of multiple GPTs in the same chat possible (similar to how ChatGPT plugins used to work), which could unlock new ways of working. Looking forward to testing this!

II. API upgrades from OpenAI and new embedding models

This one is especially relevant to the ones building products on top of OpenAI’s API.

Here’s the TLDR:

• GPT-4 Turbo’s update is set to end to end “lazy” responses from ChatGPT (see example below).

• GPT-3.5 Turbo gets 50% cheaper on input and 25% cheaper on output.

• New embedding models: text-embedding-3-small and text-embedding-3-large.

• The free text moderation model got an upgrade that makes it better at identifying harmful content.

• A new dashboard giving improved visibility into API usage, and more control over API access.

Why it matters: If you’ve increasingly experienced responses like the above screenshot, you know that this update might mean a improvement in efficiency (let’s see how good it actually is first).

It’s also a solid overall update to OpenAI’s tools for builders. This is much needed for OpenAI, who is being relentlessly challenged by alternatives, including open source, high-performing alternatives like Mistral.

III. Elevent Labs now has unicorn status

Eleven Lab is a new AI unicorn, following an $80m Series B round. The company is a pioneer in AI voice technology and lifelike text-to-speech.

They also just revealed a lineup of new products: Dubbing Studio, Voice Library marketplace, and a preview of a Mobile Reader App. Additionally, they are launching new models with improved speed and language coverage.

Why it matters: AI voice tech is changing how we digest media and holds a big market opportunity for creators and investors. Voice cloning in particular is opening up totally new ways of using and scaling voice content (both amazing and concerning ones).

Companies like Eleven Labs and Speechify are scaling their solutions to the mainstream with apps that make it possible to have text on nearly any platform (websites, your e-mail, etc.) read back to you in a convincingly lifelike way. Their APIs are also increasingly popular for adding voice content to range of applications. The possibility of greatly reduced production costs of voice content is also alluring to most publishers and creators.

 In Focus 

How AI can help anyone code (great article, and my own story)

I recently read an article from Ben Tossell on how non-technical people are using AI to code (partially paywalled). If you’ve found yourself wanting to create something in the digital sphere, but don’t know how to code, I warmly recommend giving it a read.

I’ll add to the topic with a glimpse on how I’ve personally used AI to help me build projects with code:

Before ChatGPT, I was a business analyst with a lot of ideas and entrepreneurial drive, but few tangible ways to build something by myself to share with the world. I had a bit of experience building a basic site in Webflow (no-code website builder), but I had ambitions of creating something of more complexity, and I kept running into roadblocks that required custom code to solve. No-code tools can be great place to start, but they all have their constraints on the out-of-the-box functionality and customisation possible, which might leave a lot of the potential of your project on the table.

I started out asking ChatGPT how to build a small feature using Javascript (which I had zero knowledge of at the time), and after a few chats and failed tries it worked. I was hooked. This got me started building whatplugin.ai. I’ve continuously iterated on what started as a small project, having AI help me with features in Javascript, improved aesthetics with advanced CSS, made scripts and formulas for data transformation, deployed cloud functions, set up integrations and more. Along the way, I’ve actually learnt quite a bit of code as well. I now consider it more as a way of interactive learning, rather than AI simply generating what I want. Over time, especially as you try to debug the stuff that doesn’t work, you develop an understanding for the underlying principles, syntax and workflows needed to make it work.

It’s not surprising that Sam Altman cites coding as the single most effective use case of ChatGPT so far. We’re certainly not at the point yet where AI can just make a pitch-perfect app from a single prompt (no matter what the hype-masters of the interwebs try to make you believe). Yet, with some dedication, AI can empower non-technical individuals to bring their ideas to life.

GPTs  

💬 = total chats, ↑ = chats last 7 days

Editor’s Choice

Weekly picks from me to you

Crafts personalized Spotify playlists based on user preferences.

❞ A new favourite for me. Tell it the vibe and it delivers the goods. It also connects directly to your Spotify account so it adds the playlist automatically.

Generates similar images using DALL·E 3 and provides detailed prompts based on any uploaded image.

❞ The uploaded photo came straight from my camera roll. Big ups to DALL-E for recognising the surroundings from Machu Picchu. And that llama is so dang cool.

Research in minutes, database of 282 million articles, while providing citation-backed answers to questions.

❞ This one is a great alternative to Consensus. Does very similar things, and is quite fast. (And yeah, it’s best to wait with the ice bathing until a few hours after your strength training if you wanna keep those gainz)

Trending GPTs

The GPTs with the most chats in the last 7 days

1. Consensus ↑400K    Academic Research

2. Grimoire ↑200K Coding

3. Canva  ↑300K Image Generation

4. AI PDF  ↑100K   PDF Conversations

5. AskYourPDF Research Assistant  ↑100K Academic Research

Up and coming

Lesser known (not featured in the GPT store) but rapidly growing GPTs

1. World Class Software Engineer  ↑300    AI-Assisted Development

2. Short Video Viral Caption  ↑300    Entertainment

3. Coupon Finder  ↑500    Online Shopping & Deals

4. Digital Models  ↑500    Image Generation

5. PokedexGPT V3  ↑300    Gaming

Bytes 

• Google Bard (with Gemini Pro) just surpassed GPT-4 on Hugging Face’s Chatbot Arena Leaderboard. GPT-4 Turbo is still ahead, though.

• Hugging Face and Google partner for open AI collaboration.

• Google unveils Lumiere, a video diffusion model. And users prefer it to Runway, Pika labs and Stable Video Diffusion, according to the company. It also has features beyond text-to-video generation, including image-to-video generation, making videos in a specific style, animating only a portion of a video, masking out just an area of the video, and more.

• Three new experimental AI features available directly in Chrome: smartly organize your tab, personalise your browser with AI generated themes, help me write (AI powered writer).

• Samsung’s new Galaxy S24 line arrives with Google Gemini-powered AI features. Circle to Search is the most notable AI feature (circle a piece of text with a finger and trigger a Google search on the word).

• Google is using a special stock compensation to avoid OpenAI stealing its top AI researchers (paywalled).

• Microsoft’s new AI research team is on a mission to create smaller, cheaper models (paywalled). A newly established team is focusing specifically on Small Language Models (SLMs). With appropriate fine-tuning and narrow applications, these smaller models can achieve great results at the fraction of the compute required from an LLM.

• While other Big Tech companies outsource manufacturing, OpenAI CEO Sam Altman is aiming to raise billions to create a network of semiconductor factories (AI chip factories). It’s a forward-looking move, as AI adoption is likely to accelerate, but the most important hardware is likely to be remain scarce.

• OpenInterpreter - a new tool that lets you control your computer using natural language. This is pretty cool – it runs LLMs to understand your needs and executes task on your computer (e.g. “converts the PDFs in this folder to word docs”). It’s currently available as a Github repo (40k stars), but also is inviting users to early version of a desktop app.