Over a two-week span, the team at Hunch built Coauthor which scaled #LinkedInRewind to over 300,000 users. @daviddbwilson explains the challenges they faced in scaling, cost, and output quality, live at @DataCouncilAI 2025.
๐ฅ Learn from David on April 24 at 2pm in Oakland, CA on how Hunch's LinkedIn year-in-review generator handled 1+ trillion tokens while reducing costs by 85%. Essential insights for anyone building scalable AI applications - from someone who's navigated the challenges of unexpected growth. He'll also hold office hours immediately after his talk when you can ask him anything.
https://t.co/6ez9a1ycT1
๐ฅBonus: Gemini 2.0 Flash now also available in Hunch - for improved speed, better multimodal generation, and a 1M token context window.
Already have access to Hunch? Just select o3-mini, DeepSeek-R1, or any model you prefer, in an AI Task block. Or use Auto AI to automatically select the best model for your task.
Whether you're coding, analyzing data, or solving complex problems, you now have more choices to prototype any AI functionality for your tasks. Quickly iterate across multiple models and test prompts in order to validate the right results.
๐What this means for you:
* Lower-cost, higher-confidence outputs
* Experiment with state-of-the-art reasoning capabilities across multiple models
* Speed up productivity on your most challenging tasks
๐ฃOpenAI o3-mini & DeepSeek-R1 now in Hunch, expanding our AI-first workspace with two powerful new reasoning models!
Reasoning models improve output accuracy and reduce hallucinations by spending extra time thinking. By separating tasks into steps, these models think through and confirm answers at various steps before producing the final output. This makes them particularly effective at STEM and coding tasks.๐ง
๐Try them at https://t.co/FVry9NVQQt
โ OpenAI o3-mini: Exceptional STEM reasoning in math, science, and coding tasks with significantly lower latency
โ DeepSeek-R1: An open-source model matching the performance of leading models at competitive pricing, hosted on Together AI & Fireworks AI
The recent class of purpose-built reasoning models, like OpenAI o1 and o1-mini, have proven to perform incredibly well for complex tasks. In fact, we've found them to be excellent at generating outputs requiring multiple steps, including ones that pull insights from data.
We're always thinking of ways to help you amplify your authentic voice. LinkedIn Rewind is a new tool we launched to let you share a compelling end-of-year LinkedIn summary of your best ideas and achievements from 2024. ๐
Join 25,000+ ppl and get yours: https://t.co/cEzFHXQJEl
It includes an eye-catching highlights card to increase your visibility and help you make more connections. Find more sales opportunities, collaborate with more partners, get more job leads, or simply gain more followers. ๐
What it does ๐ ๏ธ
- Creates a polished 2024 LinkedIn Rewind post (year-in-review)
- Generates a custom card image with your 2024 highlights
- Identifies your professional "superpower" from your achievements
- Wraps it all up in your authentic voice and style
Why you'll love it โค๏ธ
- Zero login required - just drop your LinkedIn URL and email
- Saves hours of reflection and writing time
- Makes you look awesome while everyone else is posting generic updates
- Early access to Coauthor's AI content platform
Perfect timing to stand out from all the year-end reflection posts flooding your feed! Skip digging through an entire year's worth of your posts and get a personalized recap that actually sounds like you wrote it. After all, it's repurposed from your actual LinkedIn content.
2024 LinkedIn Rewind is made by Hunch.
We're trending on Product Hunt!
Turn your LinkedIn activity into a year-in-review post + card to share your highlights from 2024.
>10,000 posts so far!
๐ Transform your Box content into new marketing and product assets. - We've launched an integration with @Box where Hunch users can now connect to their Box files.
๐๏ธ Repurpose any of your Box content into blog posts, announcements, PRDs, product briefs, and more, all in the same tone of voice, positioning, goals, and expertise. Build any workflow of AI models to accomplish far more sophisticated tasks than any single model can.
๐ Learn more:
https://t.co/5EQJ3o5XDg
โ๏ธ Connect Hunch to Box:
https://t.co/kW2GB6dAPc
๐ฃ With so many choices of AI models out there, learn from @daviddbwilson on how to navigate the best ones for your use case.
๐ฌ Watch - https://t.co/oBdaBADnN4
๐ Listen - https://t.co/Ujkk9FcNja
๐ Try - https://t.co/VdpEcR324d
๐ Note: Models included were based purely on usage behavior from the Hunch community. Those not included may also perform well for certain use cases.
๐ We see trends across thousands of AI tasks every day from our community. We're also constantly testing AI models in different use cases. With so many models to choose from, the classic tradeoffs between speed, quality and cost need to be considered.
๐ฏ The best approach is to try different models, even using multiple in parallel and synthesizing their outputs. Results also depend on iterative runs, prompting techniques, and how tasks are broken down into a workflow of subtasks.
๐ Unless you're an AI power user, you're probably using only one AI model like GPT or Claude. Maybe you're unsure about other models or how to get started with them. You could: 1) run thousands of your own prompts and compare the outputs, or 2) check out our little cheatsheet. ๐
๐ Something fun for your kids or loved ones this holiday - their own unique short story! Simply enter a phrase into this tool, made by @hunchtools, and run it.
๐ Use it - https://t.co/XypyWS61XT
๐ฌ Watch how - https://t.co/JExpdOTfoT
Just an example of a repeatable workflow you can build with @hunchtools
โถ๏ธ @daviddbwilson (@hunchtools Cofounder/CEO) is about to jump on a livestream with Isar Meitis to talk about how to choose the right AI model for any task. With so many choices out there he'll help you navigate the best ones for your use case.
๐ Join us now! (Starts Dec 5 at 9:00am Pacific time.)
https://t.co/wNSeRDUdVq