TalkTastic just launched a context-aware AI voice keyboard app. It works by integrating across all your macOS apps, from Slack to Messages to your browser, to not only transcribe your speech with accuracy but refine and rewrite what you say based on your screen’s context.
So imagine you’re going about your day-to-day. With Talktastic, you’d hit the little microphone button and start dictating your response back to an investor in an email. The tool will write your reply while fixing any vocal tics, adjust words for tone, and spell the names of your investors correctly. Then hop over to Slack and shoot your coworker a DM and TalkTastic adjusts for that audience and context.
Moshi is a new AI chatbot founded by French billionaire Xavier Niel. It’s designed for hyper-human-like conversations. It comes with voice mode by default and prioritizes ultra low latency.
ElevenLabs launched an API to compliment its Voice Isolator tool. It allows developers to easly implement the audio cleanup technology into any number of their own apps. The team even showcased how easy it is to build on top of it with Claude.
Runway Financial came out of beta last week. Backed by Garry Tan and others to the tune of $22.7 million, Runway Financial uses AI to help you automate your startups finances and important documents.
Sign AI is a virtual, realtime, AI-powered sign language interpreter that lives within video calls, conferences, and really anywhere it can be deployed. It’s a technological solution to the current shortage of interpreters.
Writebook is a platform for creating and publishing web-based books by 37signals — the team that also built Basecamp and Hey, co-founded by Jason Fried and David Heinemeier Hansson. Writebook is a downloadable software you can self-host, giving authors the power of ownership. Read more.
Suno is an AI-powered music app that secured $125 million in funding for its text-to-song method of generating entire tracks. It launched its new iOS app. You might have also seen its name around because Suno is being sued by music labels for copyright infringement. Read more.
next play is a new community network from Ben Lang, who’s known for helping lead Notion’s community growth from its early days. The next play community helps support people in finding their next role through introductions to startups, gatherings, and newsletter content.
Voice Isolator is another launch from AI company ElevenLabs that removes unwanted background noise and produces clear dialogue for audio clips and videos. You can see if it is as impressive as the demo is here.
Google launched Gemma 2. The lightweight models (9B and 27B) are being pitched to developers who want to incorporate AI into their apps or smart devices. Gemma 2 massively improves its performance on math-based questions, creative writing, cognitive reasoning, and more. Read more.
Anthropic launched its latest model, Claude 3.5 Sonnet. The launch follows a whopping $2.75 billion fundraise from Amazon earlier this year. Sonnet 3.5 matches and beats GPT-4o and Google’s Gemini across various tasks. Read more.
Intercom launched a new Copilot called FinAI. It can pull up answers to customer questions using your company’s knowledge base and previous history from your conversations with the customer, and references sources to its answers. Read more.
Figma’s Config conference came with several launches, including new AI features and an all-new product, Figma Slides. CEO Dylan Field says the goal isn’t to kill PowerPoint, but to offer a product for designers already using Figma — a pretty sizeable audience. Read more.
Notion launched Notion Sites, which adds more features for those who use Notion to externally publish their pages. Features include design customization (like favicons), custom domains, and SEO options. The launch steers clear of pivoting into a full-blown site builder. Read more.
Sherloq, a YC-backed startup, is a collaborative AI repository for SQL users. It gathers all your ad-hoc SQL queries into one place for easy retrieval. With the AI SQL repo plug-in, you can collaboratively manage, save, and share your SQL code, without leaving your IDE.
Remember when The Beatles got together for one last song? It got mixed reviews, but the fact that one of the biggest bands of all time embraced AI set a precedent. AI was going to change the music industry. From writing songs to generating entire tracks, many predicted an AI-powered musical revolution.
TwoShot is one of those AI apps looking to make its mark on the music scene. Launched last week, It is focused on making music accessible to all by letting you craft tracks with your voice, text prompt, or just by humming into a mic.
Say you’re working on a lo-fi track to stimulate productivity. You can instruct TwoShot to generate a “melody of flutes inspired by nature” and then pair it with a chill drum line by beatboxing into your mic. The AI will turn it into a full-fledged, professional-sounding track.
It also comes with a library of over 200,000 samples, ranging from rock to country music and everything in between, that you can grab for inspiration or even use the AI to remix into something new to create your next banger.
One of the most powerful features is TwoShot's plethora of different models. While building your track, you can swap your chosen AI model at any stage for a different one, including ones built by big names. Say you want your lyrics to have a female voice. You can load up the “Grimes” model with your prompt or existing sounds, and it will work it into a more Grimes-esque sound.
Alongside that, these models are “ethically trained,” according to the team, meaning they usually have attained artist permission to use their likeness or have worked with them to create the model itself, like in the case of the Grimes one above.
Dream Machine is a text-to-video model that competes with OpenAI’s Sora, outperforming Sora's output at 120 frames of video in around 120 seconds. Dream Machine is built by Luma AI, an a16z and Nvidia-backed startup. Read more.
ICYMI, Apple launched Apple Intelligence, iOS 18, a crazy new calculator app, and more. You can read our digestible breakdown of the best new AI features here.
MARS5 TTS from CAMB.AI is an open-source, text-to-speech model, even for “extremely tough prosodic scenarios.” That means the AI can replicate speech rhythm and intonation patterns when it matters (e.g. imagine sports commentators or a movie trailer voice-overs.)
MotherDuck is now Generally Available. The cloud SQL analytics platform is recognized for its fresh approach to data infrastructure, including blending local and cloud query execution. Read the story.
Writer launched an AI studio with a Python framework to help companies build AI apps. The startup emerged as a product to help companies use AI style guides and grew quickly, locking down enterprise customers who, as it turns out, need help building more AI stuff. Read more.
TeamCreate is a tool that creates various types of AI workers from a simple Slack integration. It was launched by the team at Sivo, a fast-growing, YC-backed debt-as-a-service startup.
Detail wants creators to share videos without editing software. Launched originally in 2020, Detail's latest update puts even more editing tools right into the recording experience, like text-based editing and remote recording. Detail is founded by Paul Veugen, whose past startups won Apple and Webby Awards. Read the story.
Descript, the popular audio & video editor, launched Underlord, an AI video assistant. If you’re a user, you know that Descript’s text-based editing was a game changer, but editing is still a time suck. Underlord can now redline changes for you, create clips, and more. Read the story.
Second is a codebase maintenance system from second-time YC founder Eric Rowell. Connect to your GitHub repo, select whatever maintenance module you need (e.g. Angular to React), review the AI agent's plan, modify as needed, and hit run. Read the Story.
Cello is a new all-in-one referral platform for B2B SaaS. Although a crowded space, the founders (a Forbes 30 under 30, serial founder + an ex-Twilio PM Director) say Cello is set apart through its seamless integration with your product, which motivated them to build it in the first place.
Elevate (formerly Bloom) offers US-based USD accounts for remote employees and freelancers, primarily in East and North Africa. Back in 2022, it became the first startup from Sudan to particpate in YC, and is now available in other emerging markets.
The ex-Head of Growth at Notion launched a tool for growth people: Roadway. It lets marketers self-serve data and pull insights with an AI assistant. Read the story.
Cartesia Sonic is an AI voice model – a state space model (SSM) that the founders invented while working as PhDs at the Stanford AI Lab. Sonic generates high-quality, lifelike speech fast (135ms model latency). Read the story.
Kino is a video app from the makers of Halide, the popular and Apple Design Award-winning photography app. Kino gives pros and amateurs more control, using the iPhone 15 Pro’s ability to shoot ProRes in Log format. Read the story.
🥇Oh, A Potato is a meal planner set apart by its AI-powered features. You can scan recipes from a book or save videos from Instagram — the app translates the instructions for you. The fun branding is prob thanks to the makers’ backgrounds in Product at Savage X Fenty, Adidas, Babbel, and Amazon.
Hex launched end-to-end no-code workflows. The data and analytics platform from ex-Palantir makers is already growing in popularity (we use it at PH) across people with varying technical backgrounds. This elevates its flexibility and collaborative potential.
Reforge launched a Chrome extension. It's an AI tool that integrates with tools like Notion and Google Docs to evaluate your work and give you advice based on experts like Andrew Chen or the ex-CPO of Tinder.
Stolkholm-based Mentimeter launched an AI builder to add polls and surveys to your interactive presentations.
Arc Search went old school with Call Arc. Picture this: You’re in your kitchen, pick up your phone, and ask Arc Search for an “authentic carbonara recipe.” It plays some hold music before providing a verbal answer and standing by for another question. Read more.
New Microsoft Windows laptops, called Copilot+, will have AI hardware and native AI software features enabling features like "Recall" for a searchable photographic memory on your PC, and the ability to run over 40 different AI models, including GPT-4o. Read more.
Maven is a new social media site that does away with followers. Backed by OpenAI CEO Sam Altman and Twitter co-founder Ev Williams, Maven is built around "open-endedness." Instead of following people, you follow interests like cooking or coding. Read more.
Canva redesigned its editing experience and launched Canva Enterprise at Canva Create. Enterprise features include enhanced security, single sign-on, dedicated support, folder customization, and a “suggestion mode” on Google Docs for reviewers to leave edits. Oh, and this happened. Read more.
Ivee, a B2B influencer marketing platform claimed Product of the Week. It lets companies find and evaluate LinkedIn, YouTube, Apple Podcast, and Substack influencers, and reach out for collaborations. Read more.
NocoBase, the top developer tool of the week, is built on the principle that 80% of requirements can be achieved with no-code tools. The 20% that require development can be implemented through the platform's plugins. NocoBase’s core and plug-ins are open source. The platform is also driven by data models, decoupling data and UI.
Voicenotes is a voice app built by the Buy Me a Coffee founder. There’s no login — open the app and hit record. Say you’re going shopping and can’t remember what deodorant you bought two weeks ago. Simply ask your AI to trawl your notes for you.
Read more | Voicenotes
🧑💻 Jijo Sunny, Aleesha John
AFFiNE, a Miro meets Notion tool, stands apart for its open-source and local-first approach. What’s under the hood of AFFiNE could potentially offer users the same exact features as Notion or Miro, but with customization and flexibility.
Read more | AFFiNE
🧑💻Jiachen He📍Singapore 💰Seed
Snaplet generates realistic seed data using AI to understand your database schema. Snaplet was founded by Peter Pistorius, who was also on the co-founding team of RedwoodJS with former GitHub founder Tom Preston-Werner.
Read more | Snaplet
🧑💻 Peter Pistorius📍Berlin 💰Pre-seed
Equals had its biggest release since its initial launch. There are three main parts to the update: a redesigned query experience, even more flexible spreadsheets, and more dashboard features.
Read more | Equals
🧑💻 Bobby Pinero, Ben McRedmond 📍SF 💰Series A 🏆GKA 2023 Best in Data
Plenty is a wealth management platform for couples. It's designed to help modern couples discuss, manage, and invest their money together. Plenty was started by a husband-wife duo in fintech (formerly Stripe and Even [acquired]).
Read more | Plenty
🧑💻 Emily Luk, Channing Allen 📍SF 💰Seed
OpenAI launched its latest model, GPT-4o. What really stood out was the voice handling. Users can ask the model a question and mid-answer, interrupt it to divert the conversation or correct it. OpenAI stressed how much more human-like this model is by showcasing its emotional reasoning capabilities. Read more | GPT-4o
Google launched a bunch of stuff including its answer to OpenAI's Sora. Veo outshines Sora by generating minute-long, 1080p videos from a single prompt. It can produce various visual styles, like landscape shots and time lapses, and allows for editing existing footage. Read more | Veo
There’s no shortage of AI coding copilots out there. I mean I only just covered GitHub’s latest coding copilot in this very newsletter, and Amazon just launched their own recently too. But where are the AI tools for non-coding related tasks like say, writing technical documents or engineering design?
Shin Kim, the founder of Eraser, an all-in-one diagrams and documentation designing tool, had the same question. So after helping build 1 million+ diagrams for customers, him and his team set to work on Eraser AI.
Eraser AI is a copilot that allows anyone to generate technical design assets by writing natural language prompts that outputs diagram code that you can save, edit, and share with your team.
Say you’re building a cloud infrastructure and you want to visually explain your thought process to your team. Instead of spending hours designing a diagram, you can instead tell Eraser what you need and it will get to work generating colorful, accurate, and icon-studded diagrams to your taste.
Or maybe you need a design doc to pitch to your manager. Drop in your specifications and Eraser will get to work generating outlines for planning documents like RFCs (request for comments) and full fledged design thinking documents.
Like a lot of AI copilots, Eraser shines when you know what you want and, in that spirit, you can make direct edits to any and all documents it generates before deciding on the final iteration.
If you want to make educated decisions on things like product direction, design variants, or copy, you need to back it up with data. Take this newsletter, for example. Every week, I dive into the data to see what resonated with you all and how I can make it even better.
For a long time, querying data has been left to people with experience in things like SQL, and if you’re a solo founder or a small team, this isn’t always possible. Since AI exploded onto the scene, though, that’s begun to change.
Originally launched back in 2023, Outerbase is a solution built to help founders and their teams easily navigate their data with AI. A big part of how it does this is with its AI chat assistant “EZQL” which lets you query your data in plain English.
Now, after participating in the 2023 YC winter batch, the team is back with Outerbase 2.0. So what’s new?
Mobile love: According to founder Brandon Strittmatter, Outerbase has been redesigned from the ground up with a specific focus on being mobile-friendly so you can have a data scientist in your pocket.
More databases: Outbase now supports over ten database connections, including MySQL, Postgres, and MongoDB, so there is no need for any pesky database migrations before using it.
New data studio: 2.0 comes with a new data studio with some new features, including an optimized GPT-4 client, an SQL co-pilot, and AI-powered data visualization tools.
Alongside that, the team also launched native Mac and Windows apps, an SQL sandbox so you can drop in a CSV to build your database, and more.
That’s how I would sum up Climatize, a new platform that’s like Republic or Wefunder but for solar projects. In other words, you can scroll the SEC-registered platform for solar projects that you want to invest in.
Climatize came about when co-founder Will Wiseman, who was working to develop solar panels, noticed that small to medium-sized solar projects struggled to raise funding, even though they qualified for grants. Now Climatize lets anyone with a US bank account scroll such projects and invest in them with as little as $10.
Examples on Climatize’s site include anything from a family-owned farm in South Carolina to a second-hand car dealership that wants to go solar and has already received part of what it needs in the form of a grant from the USDA.
The makers, who were recognized in Forbes 30 under 30, say that investors can gain up to a 10% return on their investment. And I love that — not because I’m a green money monster but because, as the makers also note, traditional activism has lacked tangible change over the last decade plus. Many of us who care about climate change watch our food consumption, shop sustainably, etc., and yet still Earth Days come and go and we feel limited by our impact.
Climatize takes advantage of the momentum that’s already growing in renewable energy (1 trillion dollars is projected to be spent on solar energy adoption) and aims to catalyze it. So, if you’d like to help more people go solar, check it out.
If you think of Clubhouse (the asynchronous voice-based chat room app that took off in the early days of the pandemic), and combine it with the serendipity of Twitter’s timeline, you’ll start to get the idea behind Airchat, the latest social app on everyone’s lips.
Originally hacked together last year, Airchat is built by AngelList founder Naval Ravikant and former Tinder executive Brian Norgard. They then rebuilt the app for both iOS and Android and recently launched it. To say it took social media by storm would be putting it lightly.
At the time Airchat was invite-only, but all the hype overwhelmed the app, and the team temporarily closed the doors in full. On Saturday, Naval posted on X that the app was open again, this time to all US and EU users with no invitation needed.
When you first open Airchat, you’ll notice that messages automatically start playing. From there, you can scroll a timeline filled with people sharing their thoughts, and you can quickly interact by replying in your voice, liking a note, and even reposting ones to your own profile.
It’s not really limited to just voice, either. You can swap between voice and video messages, and you can even add images to your post after the fact. Still, most people are zoned in on the voice chat functionality, given the tagline “a social walkie-talkie.”
One of the more interesting quirks is that voice notes automatically play at 2x the normal speed. I imagine this is to make it easier for users to consume and interact with more content, especially since the app lends itself to voice-led essays quite well. This was okay for me, but if it feels unnatural to you, you can reset the speed by holding down the play button.
Social is a hard market to break through. Over the last half-decade we have so many new takes on connecting people come and go with varying degrees of influence left over. Will Airchat be any different? It’s hard to tell. What do you think?
“Hey Spotify, make me a playlist inspired by Beyonce’s Cowboy Carter” - was the prompt I typed this morning while testing Spotify’s newest AI launch.
Since AI burst onto the tech scene, Spotify has been one of the larger companies that has fully embraced it. The music streaming service has already seen success with its popular AI DJ feature and Daylists, and now the team has announced AI playlists.
Like many AI integrations these days, it’s heavily based on prompts, but where it differs is in its complexity. In addition to some of the more basic requests like mine above, the playlist AI can also handle more creative requests like “beats to battle a zombie apocalypse” or “tunes to serenade my cat,” and Spotify insists that prompts can reference all sorts of things like animals, colors, activities but the best ones will include artists, genres, etc.
Spotify is also using its data on user tastes to customize the playlists it generates, and users themselves can use AI to edit and refine the generated playlist by instructing it to be “less poppy” or “more upbeat.”
How good is it? Moments after hitting enter, Spotify’s AI got to work crafting a pop-country-esque playlist for me that consisted of Beyonce’s latest album, complimented by the likes of Lil Nas X and even slightly more traditional sounds like Colby Acuff until I had a playlist fit for a morning of line-dancing.
Ultimately, it’s another piece of the music discovery puzzle that Spotify is hoping to complete, which is especially welcome considering most people stop discovering new music in their 30s.
In the meantime, I’ll be crafting a playlist to inspire me to finish my side projects.
Let’s be honest: AI-generated content is both incredible and sucks. It can clear brain fog in seconds but is also littering the internet.
In fact, earlier this month Google recognized just how bad your search experience is and announced some fine-tuning to their search algorithms to combat AI-generated content. Google will now downrank sites where AI is producing thousands of low-quality articles a day.
While I’m glad to see the changes, I will give it to Google that it's facing an uphill battle. Weeding out spam is like fighting a proliferating Agent Smith.
That’s why products like Arbor and Circle to Search catch my attention. They give you your own tools to save yourself from spam because even Neo needed some programs to learn Kung Fu.
Arbor’s mission is “to reindex the internet and save it from AI-generated SEO articles.” It aggregates and summarizes content across the web while duplicating and clustering content. For instance, it will give you a summary of a news event from all articles on the same topic and can highlight the differences between articles, too.
Circle to Search caught my eye because it took a familiar IRL action of circling content and powered it by AI. Though this tool is not as directly focused on spam content like Arbor is, it could definitely help you cut right to the chase — just circle an image of a steak to get results from recipes to Wikipedia articles on the topic, skipping Google entirely.
Try them out or stay stuck in a matrix of bots. The choice is yours.
Most of iOS users would agree — Siri sucks. YC President & CEO Garry Tan blames Apple’s closed system, recently tweeting that Siri's failure is due to "the sin of self preferencing and closed systems."
K, so how about an assistant that uses multiple LLMs?
Enter Martin, a new AI butler and iOS app. Martin not only comes with a delightfully posh British accent, he integrates with all the tools you already use, like your search engine, calendars, and emails.
The makers demoed a few promising features in Martin’s launch last week, including letting you cc Martin into your email thread to schedule meetings for you, instead of bouncing back and forth between tools. He’ll also notice events on your calendar and preemptively help you prep for them, like finding you a restaurant for your dinner meeting and sending you popular menu items ahead of time.
One huge benefit to Martin is that he’s easily reachable by text, email, and voice. Martin’s founders shouted out Deepgram for helping make Martin's voice capabilities fast and accurate. They also noted using AI models Claude-3 and GPT-4 Turbo.
And yes, YC is putting its money and support where its mouth is. Martin is YC-backed; a part of the Summer 2023 cohort.
As for Siri, “she” might not be along much longer — at least, not by name. Researchers from Apple are working on getting their assistant to work without needing a trigger phrase like “Siri.”
Don’t worry, old girl. We won’t forget you.
Remember Grok? Elon’s more sassy, Hitchhiker’s Guide-inspired alternative to OpenAI. Launched back in 2023 by xAI, Grok stood out from the AI crowd with features like access to real-time information via X (formerly Twitter) and a distinct lack of political correctness.
It was all quiet on the xAI front until March 11th, when Elon Musk announced the team will open source Grok. Fast forward to this week, and the company has fulfilled that promise by posting an open release of Grok-1 on GitHub, allowing researchers and developers more freedom to build on the platform and influence its direction.
According to a company blog post, the open release includes the “base model weights and network architecture” of the “314 billion parameter Mixture-of-Experts model, Grok-1.” It continues on to say that Grok-1 hasn’t undergone any fine-tuning training for any specific applications like conversational assistants.
Grok joins a growing number of AI models that have made the leap to open-source, including Meta’s LLaMa and Mistral. Google has even launched a suite of open-source sister models to Gemini.
Of course, makers are going to do what makers do best, and since the announcement, a number have taken the opportunity to dive into Grok and see what they can build, including Perplexity CEO Aravind Srinivas, who announced his company will fine-tune Grok for conversational search and make it available for their pro users.
All this is happening in the shadow of a battle going on between OpenAI and Elon Musk, who helped found OpenAI all those years ago alongside Sam Altman. The latter is suing the former over a supposed “betrayal” of its original not-for-profit mission statement. Of course, OpenAI has fired back.
AI and coding kind of go hand in hand. For the past few years, developers and non-developers alike have used tools like ChatGPT and dedicated code copilots to help them quickly build features, sites, and even whole apps.
For the most part, it’s taken the dynamic of developer and AI developer assistants. But what if the student becomes the teacher, or in this case, the AI assistant becomes the developer?
That’s the idea behind Devin.
Devin is a fully-fledged, AI-powered software engineer from Peter Thiel-backed AI startup Cognition. While there’s no shortage of coding assistants like GitHub Copilot, Devin takes it one step further, or maybe five, by being able to handle and ship entire projects, from setting up the structure, writing the code, fixing bugs, shipping updates, all on its own from start-to-finish.
How does it work? In the launch announcement, CEO Scott Wu explains how Devin can access some common developer tools, including its own code editor, shell, and even a browser within a sandboxed environment to plan and execute complex developer tasks.
Once you boot it up, you can type in a simple prompt using natural language, similar to other AI tools. From there, Devin gets to work by building a detailed, step-by-step plan to tackle the task. From there, it uses its own tools to start writing code, fixing bugs, commenting, and even testing and reporting on its progress.
According to the demo, It can do things as simple as building a static site to things as complex as fine-tuning a large language model.
Of course, it’s not sentient — yet, so if you see something that doesn’t look entirely right, you can step in and give the AI a quick nudge in the right direction.
Devin is currently only available to a select few users as the company staggers the rollout, but I'd wager a bet that it won't be long before a general release
Since tools like ChatGPT, Midjourney, and DALL-E burst onto the scene, creating high-quality content has become something anyone can do, regardless of their experience. It's a whole new world for those of us who love to create but might not have the skills or resources to do so traditionally.
Now, there’s a new tool on the block. Imagine being able to produce studio-quality audio for podcasts, audiobooks, ads, or any project you dream up. That's the idea behind Wondercraft.
Built by former Spotify and Plantir engineers, Wondercraft originally launched in beta back in May of 2023. Fast forward to January of 2024, and the team raised a $3m seed round led by Will Ventures and joined by voice-research company Eleven Labs.
The team describes Wondercraft as Canva for audio, and they want to do for audio what Canva did for design — make it accessible to everyone.
The idea is pretty simple: you type, and the platform, built on ElevenLabs AI, turns it into studio-quality audio. You can pick from different AI voices or even clone your own voice to make your projects more personal. They've also got templates to make things faster, like one for a daily news reader, which was used to make this Hacker News podcast.
Once you’re done, you can automatically translate or dub your audio to over 15 different languages and quickly generate a video to pair with your project just by typing in a few keywords.
So, if you’ve been mulling over the idea of starting a podcast but don’t want to spend money on some high-tech studio, it might be time to try…
